CPUID

In the x86 architecture, the CPUID instruction (identified by a CPUID opcode) is a processor supplementary instruction (its name derived from CPU Identification) allowing software to discover details of the processor. It was introduced by Intel in 1993 with the launch of the Pentium and SL-enhanced 486 processors.

A program can use the CPUID to determine processor type and whether features such as MMX/SSE are implemented.

History

Prior to the general availability of the CPUID instruction, programmers would write esoteric machine code which exploited minor differences in CPU behavior in order to determine the processor make and model. With the introduction of the 80386 processor, EDX on reset indicated the revision but this was only readable after reset and there was no standard way for applications to read the value.

Outside the x86 family, developers are mostly still required to use esoteric processes (involving instruction timing or CPU fault triggers) to determine the variations in CPU design that are present.

For example, in the Motorola 680x0 family — that never had a CPUID instruction of any kind — certain specific instructions required elevated privileges. These could be used to tell various CPU family members apart. In the Motorola 68010 the instruction MOVE from SR became privileged. This notable instruction (and state machine) change allowed the 68010 to meet the Popek and Goldberg virtualization requirements. Because the 68000 offered an unprivileged MOVE from SR the 2 different CPUs could be told apart by a CPU error condition being triggered.

While the CPUID instruction is specific to the x86 architecture, other architectures (like ARM) often provide on-chip registers which can be read in prescribed ways to obtain the same sorts of information provided by the x86 CPUID instruction.

Calling CPUID

The CPUID opcode is 0F A2.

In assembly language, the CPUID instruction takes no parameters as CPUID implicitly uses the EAX register to determine the main category of information returned. In Intel's more recent terminology, this is called the CPUID leaf. CPUID should be called with EAX = 0 first, as this will store in the EAX register the highest EAX calling parameter (leaf) that the CPU implements.

To obtain extended function information CPUID should be called with the most significant bit of EAX set. To determine the highest extended function calling parameter, call CPUID with EAX = 80000000h.

CPUID leaves greater than 3 but less than 80000000 are accessible only when the model-specific registers have IA32_MISC_ENABLE.BOOT_NT4 [bit 22] = 0 (which is so by default). As the name suggests, Windows NT 4.0 until SP6 did not boot properly unless this bit was set, but later versions of Windows do not need it, so basic leaves greater than 4 can be assumed visible on current Windows systems. As of April 2024, basic valid leaves go up to 23h, but the information returned by some leaves are not disclosed in the publicly available documentation, i.e. they are "reserved".

Some of the more recently added leaves also have sub-leaves, which are selected via the ECX register before calling CPUID.

EAX=0: Highest Function Parameter and Manufacturer ID

This returns the CPU's manufacturer ID string – a twelve-character ASCII string stored in EBX, EDX, ECX (in that order). The highest basic calling parameter (the largest value that EAX can be set to before calling CPUID) is returned in EAX.

Here is a list of processors and the highest function implemented.

Highest Function Parameter

Processors	Basic	Extended
Earlier Intel 486	CPUID Not Implemented
Later Intel 486 and Pentium	0x01	Not Implemented
Pentium Pro, Pentium II and Celeron	0x02	Not Implemented
Pentium III	0x03	Not Implemented
Pentium 4	0x02	0x8000 0004
Xeon	0x02	0x8000 0004
Pentium M	0x02	0x8000 0004
Pentium 4 with Hyper-Threading	0x05	0x8000 0008
Pentium D (8xx)	0x05	0x8000 0008
Pentium D (9xx)	0x06	0x8000 0008
Core Duo	0x0A	0x8000 0008
Core 2 Duo	0x0A	0x8000 0008
Xeon 3000, 5100, 5200, 5300, 5400 (5000 series)	0x0A	0x8000 0008
Core 2 Duo 8000 series	0x0D	0x8000 0008
Xeon 5200, 5400 series	0x0A	0x8000 0008
Atom	0x0A	0x8000 0008
Nehalem-based processors	0x0B	0x8000 0008
Ivy Bridge-based processors	0x0D	0x8000 0008
Skylake-based processors (proc base & max freq; Bus ref. freq)	0x16	0x8000 0008
System-On-Chip Vendor Attribute Enumeration Main Leaf	0x17	0x8000 0008
Meteor Lake-based processors	0x23	0x8000 0008

The following are known processor manufacturer ID strings:

• "AuthenticAMD" – AMD
• "CentaurHauls" – IDT WinChip/Centaur (Including some VIA and Zhaoxin CPUs)
• "CyrixInstead" – Cyrix/early STMicroelectronics and IBM
• "GenuineIntel" – Intel
• "GenuineIotel" – Intel (rare)
• "TransmetaCPU" – Transmeta
• "GenuineTMx86" – Transmeta
• "Geode by NSC" – National Semiconductor
• "NexGenDriven" – NexGen
• "RiseRiseRise" – Rise
• "SiS SiS SiS " – SiS
• "UMC UMC UMC " – UMC
• "Vortex86 SoC" – DM&P Vortex86
• "  Shanghai  " – Zhaoxin
• "HygonGenuine" – Hygon
• "Genuine  RDC" – RDC Semiconductor Co. Ltd.
• "E2K MACHINE" – MCST Elbrus
• "VIA VIA VIA " – VIA
• "AMD ISBETTER" – early engineering samples of AMD K5 processor

The following are ID strings used by open source soft CPU cores:

• "GenuineAO486" – ao486 CPU (old)
• "MiSTer AO486" – ao486 CPU (new)
• "GenuineIntel" – v586 core (this is identical to the Intel ID string)

The following are known ID strings from virtual machines:

• "MicrosoftXTA" – Microsoft x86-to-ARM
• "GenuineIntel" – Apple Rosetta 2
• "VirtualApple" – Newer versions of Apple Rosetta 2

For instance, on a GenuineIntel processor values returned in EBX is 0x756e6547, EDX is 0x49656e69 and ECX is 0x6c65746e. The following example code displays the vendor ID string as well as the highest calling parameter that the CPU implements.

	.intel_syntax noprefix
	.text
.m0: .string "CPUID: %x\n"
.m1: .string "Largest basic function number implemented: %i\n"
.m2: .string "Vendor ID: %s\n"

    .globl main

main:
	push    r12
	mov	    eax, 1
	sub	    rsp, 16
    cpuid
    lea	    rdi, .m0[rip]
	mov	    esi, eax
	call	printf
	mov     eax, 0
    cpuid
	lea	    rdi, .m1[rip]
	mov	    esi, eax
	mov	    r12d, edx
	mov	    ebp, ecx
	call    printf
	mov     3[rsp], ebx
	lea	    rsi, 3[rsp]
    lea	    rdi, .m2[rip]
    mov     7[rsp], r12d
    mov     11[rsp], ebp
	call	printf
	add	    rsp, 16
	pop	    r12
	ret

    .section .note.GNU-stack,"",@progbits

On some processors, it is possible to modify the Manufacturer ID string reported by CPUID.(EAX=0) by writing a new ID string to particular MSRs (Model-specific registers) using the WRMSR instruction. This has been used on non-Intel processors to enable features and optimizations that have been disabled in software for CPUs that don't return the GenuineIntel ID string. Processors that are known to possess such MSRs include:

Processors with Manufacturer ID MSRs

Processor	MSRs
IDT WinChip	108h-109h
VIA C3, C7	1108h-1109h
VIA Nano	1206h-1207h
Transmeta Crusoe, Efficeon	80860001h-80860003h
AMD Geode GX, LX	3000h-3001h
DM&P Vortex86EX2	52444300h-52444301h

EAX=1: Processor Info and Feature Bits

This returns the CPU's stepping, model, and family information in register EAX (also called the signature of a CPU), feature flags in registers EDX and ECX, and additional feature info in register EBX.

CPUID EAX=1: Processor Version Information in EAX

EAX
31	30	29	28	27	26	25	24	23	22	21	20	19	18	17	16	15	14	13	12	11	10	9	8	7	6	5	4	3	2	1	0
Reserved				Extended Family ID								Extended Model ID				Reserved		Processor Type		Family ID				Model				Stepping ID

• Stepping ID is a product revision number assigned due to fixed errata or other changes.
• The actual processor model is derived from the Model, Extended Model ID and Family ID fields. If the Family ID field is either 6 or 15, the model is equal to the sum of the Extended Model ID field shifted left by 4 bits and the Model field. Otherwise, the model is equal to the value of the Model field.
• The actual processor family is derived from the Family ID and Extended Family ID fields. If the Family ID field is equal to 15, the family is equal to the sum of the Extended Family ID and the Family ID fields. Otherwise, the family is equal to the value of the Family ID field.
• The meaning of the Processor Type field is given in the table below.

Processor Type

Type	Encoding in Binary
Original equipment manufacturer (OEM) Processor	00
Intel Overdrive Processor	01
Dual processor (applicable to Intel P5 Pentium processors only)	10
Reserved value	11

As of October 2023, the following x86 processor family IDs are known:

CPUID EAX=1: Processor Family IDs

Family ID + Extended Family ID	Intel	AMD	Other
0h	—	—	—
1h	—	—	—
2h	—	—	—
3h		—	—
4h	486	486, 5x86, Élan SC4xx/5xx	Cyrix 5x86, Cyrix MediaGX, UMC Green CPU, MCST Elbrus (most models), MiSTer ao486
5h	Pentium, Pentium MMX, Quark X1000	K5, K6	Cyrix 6x86, Cyrix MediaGXm, Geode (except NX), NexGen Nx586, IDT WinChip, IDT WinChip 2, IDT WinChip 3, Transmeta Crusoe, Rise mP6, SiS 550, DM&P Vortex86 (early), RDC IAD 100, MCST Elbrus-8C2
6h	Pentium Pro, Pentium II, Pentium III, Pentium M, Intel Core (all variants), Intel Atom (all variants), Xeon (except NetBurst variants), Xeon Phi (except KNC)	K7: Athlon, Athlon XP	Cyrix 6x86MX/MII, VIA C3, VIA C7, VIA Nano, DM&P Vortex86 (DX3,EX2), Zhaoxin ZX-A/B/C/C+, (Centaur CNS), MCST Elbrus-12C/16C/2C3
7h	Itanium (in IA-32 mode)	—	Zhaoxin KaiXian, Zhaoxin KaisHeng
8h		—	—
9h	—	—	—
0Ah	—	—	—
0Bh	Xeon Phi (Knights Corner)	—	—
0Ch	—	—	—
0Dh	—	—	—
0Eh	—	—	—
0Fh	NetBurst (Pentium 4)	K8/Hammer (Athlon 64)	Transmeta Efficeon
10h	—	K10: Phenom	—
11h	Itanium 2 (in IA-32 mode)	Turion X2	—
12h	—	Llano	—
13h	—	—	—
14h	—	Bobcat	—
15h	—	Bulldozer, Piledriver, Steamroller, Excavator	—
16h	—	Jaguar, Puma	—
17h	—	Zen 1, Zen 2	—
18h	—	Hygon Dhyana
19h	—	Zen 3, Zen 4	—
1Ah	—	(Zen 5)	—

CPUID EAX=1: Additional Information in EBX

Bits	EBX	Valid
7:0	Brand Index
15:8	CLFLUSH line size (Value * 8 = cache line size in bytes)	if CLFLUSH feature flag is set. CPUID.01.EDX.CLFSH [bit 19]= 1
23:16	Maximum number of addressable IDs for logical processors in this physical package; The nearest power-of-2 integer that is not smaller than this value is the number of unique initial APIC IDs reserved for addressing different logical processors in a physical package. Former use: Number of logical processors per physical processor; two for the Pentium 4 processor with Hyper-Threading Technology.	if Hyper-threading feature flag is set. CPUID.01.EDX.HTT [bit 28]= 1
31:24	Local APIC ID: The initial APIC-ID is used to identify the executing logical processor.	Pentium 4 and subsequent processors.

The processor info and feature flags are manufacturer specific but usually, the Intel values are used by other manufacturers for the sake of compatibility.

CPUID EAX=1: Feature Information in EDX and ECX

EDX			ECX
Bit	Short	Feature	Short	Feature	Bit
0	fpu	Onboard x87 FPU	sse3	SSE3 (Prescott New Instructions - PNI)	0
1	vme	Virtual 8086 mode extensions (such as VIF, VIP, PVI)	pclmulqdq	PCLMULQDQ (carry-less multiply) instruction	1
2	de	Debugging extensions (CR4 bit 3)	dtes64	64-bit debug store (edx bit 21)	2
3	pse	Page Size Extension (4 MB pages)	monitor	MONITOR and MWAIT instructions (PNI)	3
4	tsc	Time Stamp Counter and RDTSC instruction	ds-cpl	CPL qualified debug store	4
5	msr	Model-specific registers and RDMSR/WRMSR instructions	vmx	Virtual Machine eXtensions	5
6	pae	Physical Address Extension	smx	Safer Mode Extensions (LaGrande) (GETSEC instruction)	6
7	mce	Machine Check Exception	est	Enhanced SpeedStep	7
8	cx8	CMPXCHG8B (compare-and-swap) instruction	tm2	Thermal Monitor 2	8
9	apic	Onboard Advanced Programmable Interrupt Controller	ssse3	Supplemental SSE3 instructions	9
10	(mtrr)	(reserved)	cnxt-id	L1 Context ID	10
11	sep	SYSENTER and SYSEXIT fast system call instructions	sdbg	Silicon Debug interface	11
12	mtrr	Memory Type Range Registers	fma	Fused multiply-add (FMA3)	12
13	pge	Page Global Enable bit in CR4	cx16	CMPXCHG16B instruction	13
14	mca	Machine check architecture	xtpr	Can disable sending task priority messages	14
15	cmov	Conditional move: CMOV, FCMOV and FCOMI instructions	pdcm	Perfmon & debug capability	15
16	pat	Page Attribute Table	(reserved)		16
17	pse-36	36-bit page size extension	pcid	Process context identifiers (CR4 bit 17)	17
18	psn	Processor Serial Number supported and enabled	dca	Direct cache access for DMA writes	18
19	clfsh	CLFLUSH cache line flush instruction (SSE2)	sse4.1	SSE4.1 instructions	19
20	(nx)	No-execute (NX) bit (Itanium only)	sse4.2	SSE4.2 instructions	20
21	ds	Debug store: save trace of executed jumps	x2apic	x2APIC (enhanced APIC)	21
22	acpi	Onboard thermal control MSRs for ACPI	movbe	MOVBE instruction (big-endian)	22
23	mmx	MMX instructions (64-bit SIMD)	popcnt	POPCNT instruction	23
24	fxsr	FXSAVE, FXRSTOR instructions, CR4 bit 9	tsc-deadline	APIC implements one-shot operation using a TSC deadline value	24
25	sse	Streaming SIMD Extensions (SSE) instructions (aka "Katmai New Instructions"; 128-bit SIMD)	aes-ni	AES instruction set	25
26	sse2	SSE2 instructions	xsave	Extensible processor state save/restore: XSAVE, XRSTOR, XSETBV, XGETBV instructions	26
27	ss	CPU cache implements self-snoop	osxsave	XSAVE enabled by OS	27
28	htt	Max APIC IDs reserved field is Valid	avx	Advanced Vector Extensions (256-bit SIMD)	28
29	tm	Thermal monitor automatically limits temperature	f16c	Floating-point conversion instructions to/from FP16 format	29
30	ia64	IA64 processor emulating x86	rdrnd	RDRAND (on-chip random number generator) feature	30
31	pbe	Pending Break Enable (PBE# pin) wakeup capability	hypervisor	Hypervisor present (always zero on physical CPUs)	31

Reserved fields should be masked before using them for processor identification purposes.

EAX=2: Cache and TLB Descriptor information

This returns a list of descriptors indicating cache and TLB capabilities in EAX, EBX, ECX and EDX registers.

On processors that support this leaf, calling CPUID with EAX=2 will cause the bottom byte of EAX to be set to 01h and the remaining 15 bytes of EAX/EBX/ECX/EDX to be filled with 15 descriptors, one byte each. These descriptors provide information about the processor's caches, TLBs and prefetch. This is typically one cache or TLB per descriptor, but some descriptor-values provide other information as well - in particular, 00h is used for an empty descriptor, FFh indicates that the leaf does not contain valid cache information and that leaf 4h should be used instead, and FEh indicates that the leaf does not contain valid TLB information and that leaf 18h should be used instead. The descriptors may appear in any order.

For each of the four registers (EAX,EBX,ECX,EDX), if bit 31 is set, then the register should not be considered to contain valid descriptors (e.g. on Itanium in IA-32 mode, CPUID(EAX=2) returns 80000000h in EDX - this should be interpreted to mean that EDX contains no valid information, not that it contains a 512K L2 cache.)

The table below provides, for known descriptor values, a condensed description of the cache or TLB indicated by that descriptor value (or other information, where that applies). The suffixes used in the table are:

• K,M,G : binary kilobyte, megabyte, gigabyte (capacity for caches, page-size for TLBs)
• E : entries (for TLBs; e.g. 64E = 64 entries)
• p : page-size (e.g. 4Kp for TLBs where each entry describes one 4 KB page, 4K/2Mp for TLBs where each entry can describe either one 4 KB page or one 2 MB hugepage)
• L : cache-line size (e.g. 32L = 32-byte cache line size)
• S : cache sector size (e.g. 2S means that the cache uses sectors of 2 cache-lines each)
• A : associativity (e.g. 6A = 6-way set-associative, FA = fully-associative)

Legend for cache/TLB descriptor byte encodings

Level-1 instruction or data cache

Level-2 cache

Level-3 cache

Instruction or data TLB

Level-2 shared TLB

Other information

(reserved)

CPUID EAX=2: Cache/TLB descriptor byte encodings

	x0	x1	x2	x3	x4	x5	x6	x7		x8	x9	xA	xB	xC	xD	xE	xF
0x	null descriptor	ITLB: 32E, 4Kp, 4A	ITLB: 2E, 4Mp, FA	DTLB: 64E, 4Kp, 4A	DTLB: 8E, 4Mp, 4A	DTLB: 32E, 4Mp, 4A	L1I: 8K, 4A, 32L		0x	L1I: 16K, 4A, 32L	L1I: 32K, 4A, 64L	L1D: 8K, 2A, 32L	ITLB: 4E, 4Mp, FA	L1D: 16K, 4A, 32L	L1D: 16K, 4A, 64L	L1D: 24K, 6A, 64L		0x
1x	(L1D: 16K, 4A, 32L)					(L1I: 16K, 4A, 32L)			1x			(L2C: 96K, 6A, 64L)			L2C: 128K, 2A, 64L			1x
2x		L2C: 256K, 8A, 64L	L3C: 512K, 4A, 64L, 2S	L3C: 1M, 8A, 64L, 2S	L2C: 1M, 16A, 64L	L3C: 2M, 8A, 64L, 2S	(128-byte prefetch)	(128-byte prefetch)	2x	(128-byte prefetch)	L3C: 4M, 8A, 64L, 2S			L1D: 32K, 8A, 64L				2x
3x	L1I: 32K, 8A, 64L								3x		L2C: 128K, 4A, 64L, 2S	L2C: 192K, 6A, 64L, 2S	L2C: 128K, 2A, 64L, 2S	L2C: 256K, 4A, 64L, 2S	L2C: 384K, 6A, 64L, 2S	L2C: 512K, 4A, 64L, 2S		3x
4x	no L3 cache present	L2C: 128K, 4A, 32L	L2C: 256K, 4A, 32L	L2C: 512K, 4A, 32L	L2C: 1M, 4A, 32L	L2C: 2M, 4A, 32L	L3C: 4M, 4A, 64L	L3C: 8M, 8A, 64L	4x	L2C: 3M, 12A, 64L	L2C/L3C: 4M, 16A, 64L	L3C: 6M, 12A, 64L	L3C: 8M, 16A, 64L	L3C: 12M, 12A, 64L	L3C: 16M, 16A, 64L	L2C: 6M, 24A, 64L	ITLB: 32E, 4Kp	4x
5x	ITLB: 64E,FA, 4K/2M/4Mp	ITLB: 128E,FA, 4K/2M/4Mp	ITLB: 256E,FA, 4K/2M/4Mp			ITLB: 7E, 2M/4Mp, FA	DTLB: 16E, 4Mp, 4A	DTLB: 16E, 4Kp, 4A	5x		DTLB: 16E, 4Kp, FA	DTLB: 32E, 2M/4Mp, 4A	DTLB: 64E 4K/4Mp, FA	DTLB: 128E, 4K/4Mp, FA	DTLB: 256E, 4K/4Mp, FA			5x
6x	L1D: 16K, 8A, 64L	ITLB: 48E, 4Kp, FA		Two DTLBs: 32E, 2M/4Mp, 4A + 4E, 1Gp, FA	DTLB: 512E, 4Kp, 4A		L1D: 8K, 4A, 64L	L1D: 16K, 4A, 64L	6x	L1D: 32K, 4A, 64L		DTLB: 64E, 4Kp, 8A	DTLB: 256E, 4Kp, 8A	DTLB: 128E, 2M/4Mp, 8A	DTLB: 16E, 1Gp, FA			6x
7x	Trace cache, 12K-μop, 8A	Trace cache, 16K-μop, 8A	Trace cache, 32K-μop, 8A	Trace cache, 64K-μop, 8A			ITLB: 8E, 2M/4Mp, FA	(L1I: 16K, 4A, 64L)	7x	L2C: 1M, 4A, 64L	L2C: 128K, 8A, 64L, 2S	L2C: 256K, 8A, 64L, 2S	L2C: 512K, 8A, 64L, 2S	L2C: 1M, 8A, 64L, 2S	L2C: 2M, 8A, 64L	(L2C: 256K, 8A, 128L)	L2C: 512K, 2A, 64L	7x
8x	L2C: 512K, 8A, 64L	(L2C: 128K, 8A, 32L)	L2C: 256K, 8A, 32L	L2C: 512K, 8A, 32L	L2C: 1M, 8A, 32L	L2C: 2M, 8A, 32L	L2C: 512K, 4A, 64L	L2C: 1M, 8A, 64L	8x	(L3C: 2M, 4A, 64L)	(L3C: 4M, 4A, 64L)	(L3C: 8M, 4A, 64L)			(L3C: 3M, 12A, 128L)			8x
9x	(ITLB: 64E,FA, 4K-256Mp)						(DTLB: 32E,FA, 4K-256Mp)		9x				(DTLB: 96E,FA, 4K-256Mp)					9x
Ax	DTLB: 32E, 4Kp, FA								Ax									Ax
Bx	ITLB: 128E, 4Kp, 4A	ITLB: 8E, 2M/4Mp, 4A	ITLB: 64E, 4Kp, 4A	DTLB: 128E, 4Kp, 4A	DTLB: 256E, 4Kp, 4A	ITLB: 64E, 4Kp, 8A	ITLB: 128E, 4Kp, 8A		Bx			DTLB: 64E, 4Kp, 4A						Bx
Cx	DTLB: 8E, 4K/4Mp, 4A	L2TLB: 1024E, 4K/2Mp, 8A	DTLB: 16E, 2M/4Mp, 4A	Two L2 STLBs: 1536E, 4K/2Mp, 6A + 16E, 1Gp, 4A	DTLB: 32E, 2M/4Mp, 4A				Cx			L2TLB: 512E, 4Kp, 4A						Cx
Dx	L3C: 512K, 4A, 64L	L3C: 1M, 4A, 64L	L3C: 2M, 4A, 64L				L3C: 1M, 8A, 64L	L3C: 2M, 8A, 64L	Dx	L3C: 4M, 8A, 64L				L3C: 1.5M, 12A, 64L	L3C: 3M, 12A, 64L	L3C: 6M, 12A, 64L		Dx
Ex			L3C: 2M, 16A, 64L	L3C: 4M, 16A, 64L	L3C: 8M, 16A, 64L				Ex			L3C: 12M, 24A, 64L	L3C: 18M, 24A, 64L	L3C: 24M, 24A, 64L				Ex
Fx	64-byte prefetch	128-byte prefetch							Fx							Leaf 2 has no TLB info, use leaf 18h	Leaf 2 has no cache info, use leaf 4	Fx
	x0	x1	x2	x3	x4	x5	x6	x7		x8	x9	xA	xB	xC	xD	xE	xF

EAX=3: Processor Serial Number

See also: Pentium III § Controversy about privacy issues

This returns the processor's serial number. The processor serial number was introduced on Intel Pentium III, but due to privacy concerns, this feature is no longer implemented on later models (the PSN feature bit is always cleared). Transmeta's Efficeon and Crusoe processors also provide this feature. AMD CPUs however, do not implement this feature in any CPU models.

For Intel Pentium III CPUs, the serial number is returned in the EDX:ECX registers. For Transmeta Efficeon CPUs, it is returned in the EBX:EAX registers. And for Transmeta Crusoe CPUs, it is returned in the EBX register only.

Note that the processor serial number feature must be enabled in the BIOS setting in order to function.

EAX=4 and EAX=8000001Dh: Cache hierarchy and topology

These two leaves are used to provide information about the cache hierarchy levels available to the processor core on which the CPUID instruction is run. Leaf 4 is used on Intel processors and leaf 8000001Dh is used on AMD processors - they both return data in EAX, EBX, ECX and EDX, using the same data format except that leaf 4 returns a few additional fields that are considered "reserved" for leaf 8000001Dh. They both provide CPU cache information in a series of sub-leaves selected by ECX - to get information about all the cache levels, it is necessary to invoke CPUID repeatedly, with EAX=4 or 8000001Dh and ECX set to increasing values starting from 0 (0,1,2,...) until a sub-leaf not describing any caches (EAX[4:0]=0) is found. The sub-leaves that do return cache information may appear in any order, but all of them will appear before the first sub-leaf not describing any caches.

In the below table, fields that are defined for leaf 4 but not for leaf 8000001Dh are highlighted with yellow cell coloring and a (#4) item.

CPUID EAX=4 and 8000001Dh: Cache property information in EAX, EBX and EDX

Bit	EAX		EBX		EDX	Bit
0	Cache Type: 0: (No more caches) 1: Data Cache 2: Instruction Cache 3: Unified Cache 4-31: (reserved)		System coherency line size in bytes, minus 1		WBINVD cache invalidation execution scope. A value of 0 indicates that the INVD/WBINVD instructions will invalidate all lower-levels caches of this cache, including caches that belong to sibling processors sharing this cache. A value of 1 indicates that lower-level caches of sibling processors that are sharing this cache are not guaranteed to be all cleared.	0
1					Cache inclusiveness. If 1, then cache is inclusive of lower-level caches.	1
2					Complex cache indexing. If 1, then cache uses a complex function for cache indexing, else the cache is direct-mapped. (#4)	2
3					(reserved)	3
4					(reserved)	4
7:5	Cache Level (starting from 1)				(reserved)	7:5
8	Self initializing cache level (1=doesn't need software initialization after reset)				(reserved)	8
9	Fully Associative Cache				(reserved)	9
10	(WBINVD cache invalidation execution scope) (#4)				(reserved)	10
11	(Cache Inclusiveness) (#4)				(reserved)	11
13:12	(reserved)		Physical line partitions (number of cache lines that share a cache address tag), minus 1		(reserved)	13:12
21:14	Maximum number of addressable IDs for logical processors sharing this cache, minus 1				(reserved)	21:14
25:22			Ways of cache associativity, minus 1		(reserved)	25:22
31:26	Maximum number of addressable IDs for processor cores in physical package, minus 1 (#4)				(reserved)	31:26

For any caches that are valid and not fully-associative, the value returned in ECX is the number of sets in the cache minus 1. (For fully-associative caches, ECX should be treated as if it return the value 0.) For any given cache described by a sub-leaf of CPUID leaf 4 or 8000001Dh, the total cache size in bytes can be computed as:
CacheSize = (EBX[11:0]+1) * (EBX[21:12]+1) * (EBX[31:22]+1) * (ECX+1)

EAX=4 and EAX=Bh: Intel thread/core and cache topology

These two leaves are used for processor topology (thread, core, package) and cache hierarchy enumeration in Intel multi-core (and hyperthreaded) processors. As of 2013 AMD does not use these leaves but has alternate ways of doing the core enumeration.

Unlike most other CPUID leaves, leaf Bh will return different values in EDX depending on which logical processor the CPUID instruction runs; the value returned in EDX is actually the x2APIC id of the logical processor. The x2APIC id space is not continuously mapped to logical processors, however; there can be gaps in the mapping, meaning that some intermediate x2APIC ids don't necessarily correspond to any logical processor. Additional information for mapping the x2APIC ids to cores is provided in the other registers. Although the leaf Bh has sub-leaves (selected by ECX as described further below), the value returned in EDX is only affected by the logical processor on which the instruction is running but not by the subleaf.

The processor(s) topology exposed by leaf Bh is a hierarchical one, but with the strange caveat that the order of (logical) levels in this hierarchy doesn't necessarily correspond to the order in the physical hierarchy (SMT/core/package). However, every logical level can be queried as an ECX subleaf (of the Bh leaf) for its correspondence to a "level type", which can be either SMT, core, or "invalid". The level id space starts at 0 and is continuous, meaning that if a level id is invalid, all higher level ids will also be invalid. The level type is returned in bits 15:08 of ECX, while the number of logical processors at the level queried is returned in EBX. Finally, the connection between these levels and x2APIC ids is returned in EAX[4:0] as the number of bits that the x2APIC id must be shifted in order to obtain a unique id at the next level.

As an example, a dual-core Westmere processor capable of hyperthreading (thus having two cores and four threads in total) could have x2APIC ids 0, 1, 4 and 5 for its four logical processors. Leaf Bh (=EAX), subleaf 0 (=ECX) of CPUID could for instance return 100h in ECX, meaning that level 0 describes the SMT (hyperthreading) layer, and return 2 in EBX because there are two logical processors (SMT units) per physical core. The value returned in EAX for this 0-subleaf should be 1 in this case, because shifting the aforementioned x2APIC ids to the right by one bit gives a unique core number (at the next level of the level id hierarchy) and erases the SMT id bit inside each core. A simpler way to interpret this information is that the last bit (bit number 0) of the x2APIC id identifies the SMT/hyperthreading unit inside each core in our example. Advancing to subleaf 1 (by making another call to CPUID with EAX=Bh and ECX=1) could for instance return 201h in ECX, meaning that this is a core-type level, and 4 in EBX because there are 4 logical processors in the package; EAX returned could be any value greater than 3, because it so happens that bit number 2 is used to identify the core in the x2APIC id. Note that bit number 1 of the x2APIC id is not used in this example. However, EAX returned at this level could well be 4 (and it happens to be so on a Clarkdale Core i3 5x0) because that also gives a unique id at the package level (=0 obviously) when shifting the x2APIC id by 4 bits. Finally, you may wonder what the EAX=4 leaf can tell us that we didn't find out already. In EAX[31:26] it returns the APIC mask bits reserved for a package; that would be 111b in our example because bits 0 to 2 are used for identifying logical processors inside this package, but bit 1 is also reserved although not used as part of the logical processor identification scheme. In other words, APIC ids 0 to 7 are reserved for the package, even though half of these values don't map to a logical processor.

The cache hierarchy of the processor is explored by looking at the sub-leaves of leaf 4. The APIC ids are also used in this hierarchy to convey information about how the different levels of cache are shared by the SMT units and cores. To continue our example, the L2 cache, which is shared by SMT units of the same core but not between physical cores on the Westmere is indicated by EAX[26:14] being set to 1, while the information that the L3 cache is shared by the whole package is indicated by setting those bits to (at least) 111b. The cache details, including cache type, size, and associativity are communicated via the other registers on leaf 4.

Beware that older versions of the Intel app note 485 contain some misleading information, particularly with respect to identifying and counting cores in a multi-core processor; errors from misinterpreting this information have even been incorporated in the Microsoft sample code for using CPUID, even for the 2013 edition of Visual Studio, and also in the sandpile.org page for CPUID, but the Intel code sample for identifying processor topology has the correct interpretation, and the current Intel Software Developer's Manual has a more clear language. The (open source) cross-platform production code from Wildfire Games also implements the correct interpretation of the Intel documentation.

Topology detection examples involving older (pre-2010) Intel processors that lack x2APIC (thus don't implement the EAX=Bh leaf) are given in a 2010 Intel presentation. Beware that using that older detection method on 2010 and newer Intel processors may overestimate the number of cores and logical processors because the old detection method assumes there are no gaps in the APIC id space, and this assumption is violated by some newer processors (starting with the Core i3 5x0 series), but these newer processors also come with an x2APIC, so their topology can be correctly determined using the EAX=Bh leaf method.

EAX=5: MONITOR/MWAIT Features

This returns feature information related to the MONITOR and MWAIT instructions in the EAX, EBX, ECX and EDX registers.

CPUID EAX=5: MONITOR/MWAIT feature information in EAX, EBX, EDX

Bit	EAX		EBX		EDX	Bit
3:0	Smallest monitor-line size in bytes		Largest monitor-line size in bytes		Number of C0 sub-states supported for MWAIT	3:0
7:4					Number of C1 sub-states supported for MWAIT	7:4
11:8					Number of C2 sub-states supported for MWAIT	11:8
15:12					Number of C3 sub-states supported for MWAIT	15:12
19:16	(reserved)		(reserved)		Number of C4 sub-states supported for MWAIT	19:16
23:20					Number of C5 sub-states supported for MWAIT	23:20
27:24					Number of C6 sub-states supported for MWAIT	27:24
31:28					Number of C7 sub-states supported for MWAIT	31:28

CPUID EAX=5: MONITOR/MWAIT extension enumeration in ECX

Bit	ECX
	Short	Feature
0	EMX	Enumeration of MONITOR/MWAIT extensions in ECX and EDX supported
1	IBE	Supports treating interrupts as break-events for MWAIT even when interrupts are disabled
2	(reserved)
3	Monitorless_­MWAIT	Allow MWAIT to be used for power management without setting up memory monitoring with MONITOR
31:4	(reserved)

EAX=6: Thermal and power management

This returns feature bits in the EAX register and additional information in the EBX, ECX and EDX registers.

CPUID EAX=6: Thermal/power management feature bits in EAX

Bit	EAX
	Short	Feature
0	DTS	Digital Thermal Sensor capability
1		Intel Turbo Boost Technology capability
2	ARAT	Always Running APIC Timer capability
3	(reserved)
4	PLN	Power Limit Notification capability
5	ECMD	Extended Clock Modulation Duty capability
6	PTM	Package Thermal Management capability
7	HWP	Hardware-controlled Performance States. MSRs added: IA32_PM_ENABLE(770h) IA32_HWP_CAPABILITIES(771h) IA32_HWP_REQUEST(774h) IA32_HWP_STATUS(777h
8	HWP_Notification	HWP notification of dynamic guaranteed performance change - IA32_HWP_INTERRUPT(773h) MSR
9	HWP_Activity_­Window	HWP Activity Window control - bits 41:32 of IA32_HWP_REQUEST MSR
10	HWP_Energy_­Performance_­Preference	HWP Energy/performance preference control - bits 31:24 of IA32_HWP_REQUEST MSR
11	HWP_Package_­Level_Request	HWP Package-level control - IA32_HWP_REQUEST_PKG(772h) MSR
12	(reserved)
13	HDC	Hardware Duty Cycling supported. MSRs added: IA32_PKG_HDC_CTL (DB0h) IA32_PM_CTL1 (DB1h) IA32_THREAD_STALL (DB2h)
14		Intel Turbo Boost Max Technology 3.0 available
15		Interrupts upon changes to IA32_HWP_CAPABILITIES.Highest_Performance (bits 7:0) supported
16		HWP PECI override supported - bits 63:60 of IA32_HWP_PECI_REQUEST_INFO(775h) MSR
17		Flexible HWP - bits 63:59 of IA32_HWP_REQUEST MSR
18		Fast access mode for IA32_HWP_REQUEST MSR supported
19	HW_FEEDBACK	Hardware Feedback Interface. Added MSRs: IA32_HW_FEEDBACK_PTR(17D0h) IA32_HW_FEEDBACK_CONFIG(17D1h) (bit 0 enables HFI, bit 1 enables Intel Thread Director)
20		IA32_HWP_REQUEST of idle logical processor ignored when only one of two logical processors that share a physical processor is active.
21	(reserved)
22		IA32_HWP_CTL(776h) MSR supported
23		Intel Thread Director supported. Added MSRs: IA32_THREAD_FEEDBACK_CHAR(17D2h) IA32_HW_FEEDBACK_THREAD_CONFIG(17D4h)
24		IA32_THERM_INTERRUPT MSR bit 25 supported
31:25	(reserved)

CPUID EAX=6: Thermal/power management feature fields in EBX, ECX and EDX

Bit	EBX		ECX		EDX	Bit
0	Number of Interrupt Thresholds in Digital Thermal Sensor		Effective frequency interface supported - IA32_MPERF(0E7h) and IA32_APERF(0E8h) MSRs		Hardware Feedback reporting: Performance Capability Reporting supported	0
1			(ACNT2 Capability)		Hardware Feedback reporting: Efficiency Capability Reporting supported	1
2			(reserved)		(reserved)	2
3			Performance-Energy Bias capability - IA32_ENERGY_PERF_BIAS(1B0h) MSR			3
7:4	(reserved)		(reserved)			7:4
11:8			Number of Intel Thread Director classes supported by hardware		Size of Hardware Feedback interface structure (in units of 4 KB) minus 1	11:8
15:12					(reserved)	15:12
31:16			(reserved)		Index of this logical processor's row in hardware feedback interface structure	31:16

EAX=7, ECX=0: Extended Features

This returns extended feature flags in EBX, ECX, and EDX. Returns the maximum ECX value for EAX=7 in EAX.

CPUID EAX=7,ECX=0: Extended feature bits in EBX, ECX and EDX

Bit	EBX			ECX			EDX		Bit
	Short	Feature		Short	Feature		Short	Feature
0	fsgsbase	Access to base of %fs and %gs		prefetchwt1	PREFETCHWT1 instruction		(sgx-tem)	?	0
1		IA32_TSC_ADJUST MSR		avx512-vbmi	AVX-512 Vector Bit Manipulation Instructions		sgx-keys	Attestation Services for Intel SGX	1
2	sgx	Software Guard Extensions		umip	User-mode Instruction Prevention		avx512-4vnniw	AVX-512 4-register Neural Network Instructions	2
3	bmi1	Bit Manipulation Instruction Set 1		pku	Memory Protection Keys for User-mode pages		avx512-4fmaps	AVX-512 4-register Multiply Accumulation Single precision	3
4	hle	TSX Hardware Lock Elision		ospke	PKU enabled by OS		fsrm	Fast Short REP MOVSB	4
5	avx2	Advanced Vector Extensions 2		waitpkg	Timed pause and user-level monitor/wait instructions (TPAUSE, UMONITOR, UMWAIT)		uintr	User Inter-processor Interrupts	5
6	fdp-excptn-only	x87 FPU data pointer register updated on exceptions only		avx512-vbmi2	AVX-512 Vector Bit Manipulation Instructions 2		(reserved)		6
7	smep	Supervisor Mode Execution Prevention		cet_ss/shstk	Control flow enforcement (CET): shadow stack (SHSTK alternative name)		(reserved)		7
8	bmi2	Bit Manipulation Instruction Set 2		gfni	Galois Field instructions		avx512-vp2intersect	AVX-512 vector intersection instructions on 32/64-bit integers	8
9	erms	Enhanced REP MOVSB/STOSB		vaes	Vector AES instruction set (VEX-256/EVEX)		srbds-ctrl	Special Register Buffer Data Sampling Mitigations	9
10	invpcid	INVPCID instruction		vpclmulqdq	CLMUL instruction set (VEX-256/EVEX)		md-clear	VERW instruction clears CPU buffers	10
11	rtm	TSX Restricted Transactional Memory		avx512-vnni	AVX-512 Vector Neural Network Instructions		rtm-always-abort	All TSX transactions are aborted	11
12	rdt-m/pqm	Intel Resource Director (RDT) Monitoring or AMD Platform QOS Monitoring		avx512-bitalg	AVX-512 BITALG instructions		(reserved)		12
13		x87 FPU CS and DS deprecated		tme_en	Total Memory Encryption MSRs available		rtm-force-abort	TSX_FORCE_ABORT (MSR 0x10f) is available	13
14	mpx	Intel MPX (Memory Protection Extensions)		avx512-vpopcntdq	AVX-512 Vector Population Count Double and Quad-word		serialize	SERIALIZE instruction	14
15	rdt-a/pqe	Intel Resource Director (RDT) Allocation or AMD Platform QOS Enforcement		(fzm)	?		hybrid	Mixture of CPU types in processor topology (e.g. Alder Lake)	15
16	avx512-f	AVX-512 Foundation		la57	5-level paging (57 address bits)		tsxldtrk	TSX load address tracking suspend/resume instructions (TSUSLDTRK and TRESLDTRK)	16
17	avx512-dq	AVX-512 Doubleword and Quadword Instructions		mawau	The value of userspace MPX Address-Width Adjust used by the BNDLDX and BNDSTX Intel MPX instructions in 64-bit mode		(reserved)		17
18	rdseed	RDSEED instruction					pconfig	Platform configuration (Memory Encryption Technologies Instructions)	18
19	adx	Intel ADX (Multi-Precision Add-Carry Instruction Extensions)					lbr	Architectural Last Branch Records	19
20	smap	Supervisor Mode Access Prevention					cet-ibt	Control flow enforcement (CET): indirect branch tracking	20
21	avx512-ifma	AVX-512 Integer Fused Multiply-Add Instructions					(reserved)		21
22	(pcommit)	(PCOMMIT instruction, deprecated)		rdpid	RDPID (Read Processor ID) instruction and IA32_TSC_AUX MSR		amx-bf16	AMX tile computation on bfloat16 numbers	22
23	clflushopt	CLFLUSHOPT instruction		kl	AES Key Locker		avx512-fp16	AVX-512 half-precision floating-point arithmetic instructions	23
24	clwb	CLWB (Cache line writeback) instruction		bus-lock-detect	Bus lock debug exceptions		amx-tile	AMX tile load/store instructions	24
25	pt	Intel Processor Trace		cldemote	CLDEMOTE (Cache line demote) instruction		amx-int8	AMX tile computation on 8-bit integers	25
26	avx512-pf	AVX-512 Prefetch Instructions		(mprr)	?		ibrs / spec_ctrl	Speculation Control, part of Indirect Branch Control (IBC): Indirect Branch Restricted Speculation (IBRS) and Indirect Branch Prediction Barrier (IBPB)	26
27	avx512-er	AVX-512 Exponential and Reciprocal Instructions		movdiri	MOVDIRI instruction		stibp	Single Thread Indirect Branch Predictor, part of IBC	27
28	avx512-cd	AVX-512 Conflict Detection Instructions		movdir64b	MOVDIR64B (64-byte direct store) instruction		L1D_FLUSH	IA32_FLUSH_CMD MSR	28
29	sha	SHA-1 and SHA-256 extensions		enqcmd	Enqueue Stores and EMQCMD/EMQCMDS instructions			IA32_ARCH_CAPABILITIES MSR (lists speculative side channel mitigations)	29
30	avx512-bw	AVX-512 Byte and Word Instructions		sgx-lc	SGX Launch Configuration			IA32_CORE_CAPABILITIES MSR (lists model-specific core capabilities)	30
31	avx512-vl	AVX-512 Vector Length Extensions		pks	Protection keys for supervisor-mode pages		ssbd	Speculative Store Bypass Disable, as mitigation for Speculative Store Bypass (IA32_SPEC_CTRL)	31

EAX=7, ECX=1: Extended Features

This returns extended feature flags in all four registers.

CPUID EAX=7,ECX=1: Extended feature bits in EAX, EBX, ECX, and EDX

Bit	EAX			EBX			ECX			EDX		Bit
	Short	Feature		Short	Feature		Short	Feature		Short	Feature
0	sha512	SHA-512 extensions			Intel PPIN (Protected Processor Inventory Number): IA32_PPIN_CTL (04Eh) and IA32_PPIN (04Fh) MSRs.		(reserved)			(reserved)		0
1	sm3	SM3 hash extensions		pbndkb	Total Storage Encryption: PBNDKB instruction and TSE_CAPABILITY (9F1h) MSR.		(reserved)			(reserved)		1
2	sm4	SM4 cipher extensions		(reserved)			legacy_reduced_isa	X86S		(reserved)		2
3	rao-int	Remote Atomic Operations on integers: AADD, AAND, AOR, AXOR instructions		(reserved)			(reserved)			(reserved)		3
4	avx-vnni	AVX Vector Neural Network Instructions (VNNI) (VEX encoded)		(reserved)			sipi64	64-bit SIPI		avx-vnni-int8	AVX VNNI INT8 instructions	4
5	avx512-bf16	AVX-512 instructions for bfloat16 numbers		(reserved)			(reserved)			avx-ne-convert	AVX no-exception FP conversion instructions (bfloat16↔FP32 and FP16→FP32)	5
6	lass	Linear Address Space Separation (CR4 bit 27)		(reserved)			(reserved)			(reserved)		6
7	cmpccxadd	CMPccXADD instructions		(reserved)			(reserved)			(reserved)		7
8	archperfmonext	Architectural Performance Monitoring Extended Leaf (EAX=23h)		(reserved)			(reserved)			amx-complex	AMX support for "complex" tiles (TCMMIMFP16PS and TCMMRLFP16PS)	8
9	(dedup)	?		(reserved)			(reserved)			(reserved)		9
10	fzrm	Fast zero-length REP MOVSB		(reserved)			(reserved)			avx-vnni-int16	AVX VNNI INT16 instructions	10
11	fsrs	Fast short REP STOSB		(reserved)			(reserved)			(reserved)		11
12	rsrcs	Fast short REP CMPSB and REP SCASB		(reserved)			(reserved)			(reserved)		12
13	(reserved)			(reserved)			(reserved)			utmr	User-timer events: IA32_UINTR_TIMER (1B00h) MSR	13
14	(reserved)			(reserved)			(reserved)			prefetchi	Instruction-cache prefetch instructions (PREFETCHIT0 and PREFETCHIT1)	14
15	(reserved)			(reserved)			(reserved)			user_msr	User-mode MSR access instructions (URDMSR and UWRMSR)	15
16	(reserved)			(reserved)			(reserved)			(reserved)		16
17	fred	Flexible Return and Event Delivery		(reserved)			(reserved)			uiret-uif-from-rflags	If 1, the UIRET (User Interrupt Return) instruction will set UIF (User Interrupt Flag) to the value of bit 1 of the RFLAGS image popped off the stack.	17
18	lkgs	LKGS Instruction		(reserved)			(reserved)			cet-sss	If 1, then Control-Flow Enforcement (CET) Supervisor Shadow Stacks (SSS) are guaranteed not to become prematurely busy as long as shadow stack switching does not cause page faults on the stack being switched to.	18
19	wrmsrns	WRMSRNS instruction (non-serializing write to MSRs)		(reserved)			(reserved)			avx10	AVX10 Converged Vector ISA (see also leaf 24h)	19
20	nmi_src	NMI source reporting		(reserved)			(reserved)			(reserved)		20
21	amx-fp16	AMX instructions for FP16 numbers		(reserved)			(reserved)			APX_F	Advanced Performance Extensions, Foundation (adds REX2 and extended EVEX prefix encodings to support 32 GPRs, as well as some new instructions)	21
22	hreset	HRESET instruction, IA32_HRESET_ENABLE (17DAh) MSR, and Processor History Reset Leaf (EAX=20h)		(reserved)			(reserved)			(reserved)		22
23	avx-ifma	AVX IFMA instructions		(reserved)			(reserved)			mwait	MWAIT instruction	23
24	(reserved)			(reserved)			(reserved)			(reserved)		24
25	(reserved)			(reserved)			(reserved)			(reserved)		25
26	lam	Linear Address Masking		(reserved)			(reserved)			(reserved)		26
27	msrlist	RDMSRLIST and WRMSRLIST instructions, and the IA32_BARRIER (02Fh) MSR		(reserved)			(reserved)			(reserved)		27
28	(reserved)			(reserved)			(reserved)			(reserved)		28
29	(reserved)			(reserved)			(reserved)			(reserved)		29
30	invd_disable_­post_bios_done	If 1, supports INVD instruction execution prevention after BIOS Done.		(reserved)			(reserved)			(reserved)		30
31	(reserved)			(reserved)			(reserved)			(reserved)		31

EAX=7, ECX=2: Extended Features

This returns extended feature flags in EDX.

EAX, EBX and ECX are reserved.

CPUID EAX=7,ECX=2: Extended feature bits in EDX

Bit	EDX
	Short	Feature
0	psfd	Fast Store Forwarding Predictor disable supported. (SPEC_CTRL (MSR 48h) bit 7)
1	ipred_ctrl	IPRED_DIS controls supported. (SPEC_CTRL bits 3 and 4) IPRED_DIS prevents instructions at an indirect branch target from speculatively executing until the branch target address is resolved.
2	rrsba_ctrl	RRSBA behavior disable supported. (SPEC_CTRL bits 5 and 6)
3	ddpd_u	Data Dependent Prefetcher disable supported. (SPEC_CTRL bit 8)
4	bhi_ctrl	BHI_DIS_S behavior enable supported. (SPEC_CTRL bit 10) BHI_DIS_S prevents predicted targets of indirect branches executed in ring0/1/2 from being selected based on branch history from branches executed in ring 3.
5	mcdt_no	If set, the processor does not exhibit MXCSR configuration dependent timing.
6		UC-lock disable feature supported.
31:7	(reserved)

EAX=0Dh: XSAVE features and state-components

This leaf is used to enumerate XSAVE features and state-components.

The XSAVE instruction set extension is designed to save/restore CPU extended state (typically for the purpose of context switching) in a manner that can be extended to cover new instruction set extensions without the OS context-switching code needing to understand the specifics of the new extensions. This is done by defining a series of state-components, each with a size and offset within a given save area, and each corresponding to a subset of the state needed for one CPU extension or another. The EAX=0Dh CPUID leaf is used to provide information about which state-components the CPU supports and what their sizes/offsets are, so that the OS can reserve the proper amount of space and set the associated enable-bits.

The state-components can be subdivided into two groups: user-state (state-items that are visible to the application, e.g. AVX-512 vector registers), and supervisor-state (state items that affect the application but are not directly user-visible, e.g. user-mode interrupt configuration). The user-state items are enabled by setting their associated bits in the XCR0 control register, while the supervisor-state items are enabled by setting their associated bits in the IA32_XSS (0DA0h) MSR - the indicated state items then become the state-components that can be saved and restored with the XSAVE/XRSTOR family of instructions.

The XSAVE mechanism can handle up to 63 state-components in this manner. State-components 0 and 1 (x87 and SSE, respectively) have fixed offsets and sizes - for state-components 2 to 62, their sizes, offsets and a few additional flags can be queried by executing CPUID with EAX=0Dh and ECX set to the index of the state-component. This will return the following items in EAX, EBX and ECX (with EDX being reserved):

CPUID EAX=0Dh, ECX≥2: XSAVE state-component information

Bit	EAX	EBX	ECX	Bit
0	Size in bytes of state-component	Offset of state-component from the start of the XSAVE/XRSTOR save area (This offset is 0 for supervisor state-components, since these can only be saved with the XSAVES/XRSTORS instruction, which use compacting.)	User/supervisor state-component: 0=user-state (enabled through XCR0) 1=supervisor-state (enabled through IA32_XSS)	0
1			64-byte alignment enable when state save compaction is used. If this bit is set for a state-component, then, when storing state with compaction, padding will be inserted between the preceding state-component and this state-component as needed to provide 64-byte alignment. If this bit is not set, the state-component will be stored directly after the preceding one.	1
31:2			(reserved)	31:2

Attempting to query an unsupported state-component in this manner results in EAX,EBX,ECX and EDX all being set to 0.

Sub-leaves 0 and 1 of CPUID leaf 0Dh are used to provide feature information:

CPUID EAX=0Dh,ECX=0: XSAVE features

EBX	ECX	EDX:EAX
Maximum size (in bytes) of XSAVE save area for the set of state-components currently set in XCR0.	Maximum size (in bytes) of XSAVE save area if all state-components supported by XCR0 on this CPU were enabled at the same time.	64-bit bitmap of state-components supported by XCR0 on this CPU.

CPUID EAX=0Dh,ECX=1: XSAVE extended features

EAX	EBX	EDX:ECX
XSAVE feature flags (see below table)	Size (in bytes) of XSAVE area containing all the state-components currently set in XCR0 and IA32_XSS combined.	64-bit bitmap of state-components supported by IA32_XSS on this CPU.

EAX=0Dh,ECX=1: XSAVE feature flags in EAX

Bit	EAX
	Short	Feature
0	xsaveopt	XSAVEOPT instruction: save state-components that have been modified since last XRSTOR
1	xsavec	XSAVEC instruction: save/restore state with compaction
2	xgetbv_ecx1	XGETBV with ECX=1 support
3	xss	XSAVES and XRSTORS instructions and IA32_XSS MSR: save/restore state with compaction, including supervisor state.
4	xfd	XFD (Extended Feature Disable) supported
31:5	(reserved)

As of July 2023, the XSAVE state-components that have been architecturally defined are:

XSAVE State-components

Index	Description	Enabled with
0	x87 state	XCR0
1	SSE state: XMM0-XMM15 and MXCSR	XCR0
2	AVX state: top halves of YMM0 to YMM15
3	MPX state: BND0-BND3 bounds registers
4	MPX state: BNDCFGU and BNDSTATUS registers
5	AVX-512 state: opmask registers k0-k7
6	AVX-512 "ZMM_Hi256" state: top halves of ZMM0 to ZMM15
7	AVX-512 "Hi16_ZMM" state: ZMM16-ZMM31
8	Processor Trace state	IA32_XSS
9	PKRU (User Protection Keys) register	XCR0
10	PASID (Process Address Space ID) state	IA32_XSS
11	CET_U state (Control-flow Enforcement Technology: user-mode functionality MSRs)
12	CET_S state (CET: shadow stack pointers for rings 0,1,2)
13	HDC (Hardware Duty Cycling) state
14	UINTR (User-Mode Interrupts) state
15	LBR (Last Branch Record) state
16	HWP (Hardware P-state control) state
17	AMX tile configuration state: TILECFG	XCR0
18	AMX tile data registers: tmm0-tmm7
19	APX extended general-purpose registers: r16-r31
20 to 61	(reserved)
62	Lightweight Profiling (LWP) (AMD only)	XCR0
63	(reserved)

EAX=12h: SGX capabilities

This leaf provides information about the supported capabilities of the Intel Software Guard Extensions (SGX) feature. The leaf provides multiple sub-leaves, selected with ECX.

Sub-leaf 0 provides information about supported SGX leaf functions in EAX and maximum supported SGX enclave sizes in EDX; ECX is reserved. EBX provides a bitmap of bits that can be set in the MISCSELECT field in the SECS (SGX Enclave Control Structure) - this field is used to control information written to the MISC region of the SSA (SGX Save State Area) when an AEX (SGX Asynchronous Enclave Exit) occurs.

CPUID EAX=12h,ECX=0: SGX leaf functions, MISCSELECT and maximum-sizes

Bit	EAX			EBX			EDX		Bit
	Short	Feature		Short	Feature		Short	Feature
0	sgx1	SGX1 leaf functions		EXINFO	MISCSELECT: report information about page fault and general protection exception that occurred inside enclave		MaxEnclave­Size_Not64	Log2 of maximum enclave size supported in non-64-bit mode	0
1	sgx2	SGX2 leaf functions		CPINFO	MISCSELECT: report information about control protection exception that occurred inside enclave				1
2	(reserved)			(reserved)					2
3	(reserved)			(reserved)					3
4	(reserved)			(reserved)					4
5	oss	ENCLV leaves: EINCVIRTCHILD, EDECVIRTCHILD, and ESETCONTEXT		(reserved)					5
6		ENCLS leaves: ETRACKC, ERDINFO, ELDBC, ELDUC		(reserved)					6
7		ENCLU leaf: EVERIFYREPORT2		(reserved)					7
8	(reserved)			(reserved)			MaxEnclave­Size_64	Log2 of maximum enclave size supported in 64-bit mode	8
9	(reserved)			(reserved)					9
10		ENCLS leaf: EUPDATESVN		(reserved)					10
11		ENCLU leaf: EDECSSA		(reserved)					11
12	(reserved)			(reserved)					12
13	(reserved)			(reserved)					13
14	(reserved)			(reserved)					14
15	(reserved)			(reserved)					15
31:16	(reserved)			(reserved)			(reserved)		31:16

Sub-leaf 1 provides a bitmap of which bits can be set in the 128-bit ATTRIBUTES field of SECS in EDX:ECX:EBX:EAX (this applies to the SECS copy used as input to the ENCLS[ECREATE] leaf function). The top 64 bits (given in EDX:ECX) are a bitmap of which bits can be set in the XFRM (X-feature request mask) - this mask is a bitmask of which CPU state-components (see leaf 0Dh) will be saved to the SSA in case of an AEX; this has the same layout as the XCR0 control register. The other bits are given in EAX and EBX, as follows:

CPUID EAX=12h,ECX=1: SGX settable bits in SECS.ATTRIBUTES

Bit	EAX			EBX		Bit
	Short	Feature		Short	Feature
0	(INIT)	(must be 0)		(reserved)		0
1	DEBUG	Permit debugger to read and write enclave data using EDBGRD and EDBGWR				1
2	MODE64BIT	64-bit-mode enclave				2
3	(reserved)					3
4	PROVISIONKEY	Provisioning key available from EGETKEY				4
5	EINITTOKEN_KEY	EINIT token key available from EGETKEY				5
6	CET	CET (Control-Flow Enforcement Technology) attributes enable				6
7	KSS	Key Separation and Sharing				7
8	(reserved)					8
9	(reserved)					9
10	AEXNOTIFY	Threads inside enclave may receive AEX notifications				10
31:11	(reserved)					31:11

Sub-leaves 2 and up are used to provide information about which physical memory regions are available for use as EPC (Enclave Page Cache) sections under SGX.

CPUID EAX=12h,ECX≥2: SGX Enclave Page Cache section information

Bits	EAX		EBX		ECX		EDX	Bits
3:0	Sub-leaf type: 0000: Invalid 0001: EPC section other: reserved		Bits 51:32 of physical base address of EPC section		EPC Section properties: 0000: Invalid 0001: Has confidentiality, integrity and replay protection 0010: Has confidentiality protection only 0011: Has confidentiality and integrity protection other: reserved		Bits 51:32 of size of EPC section	3:0
11:4	(reserved)				(reserved)			11:4
19:12	Bits 31:12 of physical base address of EPC section				Bits 31:12 of size of EPC section			19:12
31:20			(reserved)				(reserved)	31:20

EAX=14h, ECX=0: Processor Trace

This sub-leaf provides feature information for Intel Processor Trace (also known as Real Time Instruction Trace).

The value returned in EAX is the index of the highest sub-leaf supported for CPUID with EAX=14h. EBX and ECX provide feature flags, EDX is reserved.

CPUID EAX=14h,ECX=0: Processor Trace feature bits in EBX and ECX

Bit	EBX			ECX		Bit
	Short	Feature		Short	Feature
0		CR3 filtering supported		topaout	ToPA (Table of Physical Addresses) output mechanism for trace packets supported	0
1		Configurable PSB (Packet Stream Boundary) packet rate and Cycle-Accurate Mode (CYC packets) supported		mentry	ToPA tables can contain hold multiple output entries	1
2		IP filtering, TraceStop filtering and preservation of PT MSRs across warm reset supported		snglrngout	Single-Range Output scheme supported	2
3		MTC (Mini Time Counter) timing packets supported, and suppression of COFI (Change of Flow Instructions) packets supported.			Output to Trace Transport subsystem supported	3
4	ptwrite	PTWRITE instruction supported		(reserved)		4
5		Power Event Trace supported		(reserved)		5
6		Preservation of PSB and PMI (performance monitoring interrupt) supported		(reserved)		6
7		Event Trace packet generation supported		(reserved)		7
8		TNT (Branch Taken-Not-Taken) packet generation disable supported.		(reserved)		8
9		PTTT (Processor Trace Trigger Tracing) supported		(reserved)		9
30:10	(reserved)			(reserved)		30:10
31	(reserved)				IP (Instruction Pointer) format for trace packets that contain IP payloads: 0=RIP (effective-address IP) 1=LIP (linear-address IP, with CS base address added)	31

EAX=15h and EAX=16h: CPU, TSC, Bus and Core Crystal Clock frequencies

These two leaves provide information about various frequencies in the CPU in EAX, EBX and ECX (EDX is reserved in both leaves).

CPUID EAX=15h: TSC and Core Crystal frequency information

EAX	EBX	ECX
Ratio of TSC frequency to Core Crystal Clock frequency, denominator	Ratio of TSC frequency to Core Crystal Clock frequency, numerator	Core Crystal Clock frequency, in units of Hz

CPUID EAX=16h: Processor and Bus specification frequencies

Bits	EAX	EBX	ECX	Bits
15:0	Processor Base Frequency (in MHz)	Processor Maximum Frequency (in MHz)	Bus/Reference frequency (in MHz)	15:0
31:16	(reserved)	(reserved)	(reserved)	31:16

If the returned values in EBX and ECX of leaf 15h are both nonzero, then the TSC (Time Stamp Counter) frequency in Hz is given by TSCFreq = ECX*(EBX/EAX).

On some processors (e.g. Intel Skylake), CPUID_15h_ECX is zero but CPUID_16h_EAX is present and not zero. On all known processors where this is the case, the TSC frequency is equal to the Processor Base Frequency, and the Core Crystal Clock Frequency in Hz can be computed as CoreCrystalFreq = (CPUID_16h_EAX * 10000000) * (CPUID_15h_EAX/CPUID_15h_EBX).

On processors that enumerate the TSC/Core Crystal Clock ratio in CPUID leaf 15h, the APIC timer frequency will be the Core Crystal Clock frequency divided by the divisor specified by the APIC's Divide Configuration Register.

EAX=17h: SoC Vendor Attribute Enumeration

This leaf is present in systems where an x86 CPU IP core is implemented in an SoC (System on chip) from another vendor - whereas the other leaves of CPUID provide information about the x86 CPU core, this leaf provides information about the SoC. This leaf takes a sub-leaf index in ECX.

Sub-leaf 0 returns a maximum sub-leaf index in EAX (at least 3), and SoC identification information in EBX/ECX/EDX:

CPUID EAX=17h,ECX=0: SoC identification information

Bit	EBX		ECX		EDX	Bit
15:0	SoC Vendor ID		SoC Project ID		SoC Stepping ID within an SoC project	15:0
16	SoC Vendor ID scheme 0 : Vendor IDs assigned by Intel 1 : Industry standard enumeration scheme					16
31:17	(reserved)					31:17

Sub-leaves 1 to 3 return a 48-byte SoC vendor brand string in UTF-8 format. Sub-leaf 1 returns the first 16 bytes in EAX,EBX,ECX,EDX (in that order); sub-leaf 2 returns the next 16 bytes and sub-leaf 3 returns the last 16 bytes. The string is allowed but not required to be null-terminated.

EAX=19h: Intel Key Locker features

This leaf provides feature information for Intel Key Locker in EAX, EBX and ECX. EDX is reserved.

CPUID EAX=19h: Key Locker feature bits in EAX, EBX and ECX

Bit	EAX			EBX			ECX		Bit
	Short	Feature		Short	Feature		Short	Feature
0		Key Locker restriction of CPL0-only supported		aes_kle	AES "Key Locker" Instructions enabled			No-backup parameter to LOADIWKEY supported	0
1		Key Locker restriction of no-encrypt supported		(reserved)				KeySource encoding of 1 (randomization of internal wrapping key) supported	1
2		Key Locker restriction of no-decrypt supported		aes_wide_kl	AES "Wide Key Locker" Instructions		(reserved)		2
3		(Process Restriction)		(reserved)			(reserved)		3
4	(reserved)			kl_msrs	"Key Locker" MSRs		(reserved)		4
31:5	(reserved)			(reserved)			(reserved)		31:5

EAX=21h: Reserved for TDX enumeration

When Intel TDX (Trust Domain Extensions) is active, attempts to execute the CPUID instruction by a TD (Trust Domain) guest will be intercepted by the TDX module. This module will, when CPUID is invoked with EAX=21h and ECX=0 (leaf 21h, sub-leaf 0), return the index of the highest supported sub-leaf for leaf 21h in EAX and a TDX module vendor ID string as a 12-byte ASCII string in EBX,EDX,ECX (in that order). Intel's own module implementation returns the vendor ID string "IntelTDX    " (with four trailing spaces) - for this module, additional feature information is not available through CPUID and must instead be obtained through the TDX-specific TDCALL instruction.

This leaf is reserved in hardware and will (on processors whose highest basic leaf is 21h or higher) return 0 in EAX/EBX/ECX/EDX when run directly on the CPU.

EAX=24h, ECX=0: AVX10 Features

This returns a maximum supported sub-leaf in EAX and AVX10 feature information in EBX. (ECX and EDX are reserved.)

CPUID EAX=24h, ECX=0: AVX10 feature bits in EBX

Bit	EBX
	Short	Feature
7:0		AVX10 Converged Vector ISA version (≥1)
15:8	(reserved)
16	avx10-128	128-bit vector support is present
17	avx10-256	256-bit vector support is present
18	avx10-512	512-bit vector support is present
31:19	(reserved)

EAX=24h, ECX=1: Discrete AVX10 Features

Subleaf 1 is reserved for AVX10 features not bound to a version. None are currently defined.

EAX=40000000h-4FFFFFFFh: Reserved for Hypervisor use

When the CPUID instruction is executed under Intel VT-x or AMD-v virtualization, it will be intercepted by the hypervisor, enabling the hypervisor to return CPUID feature flags that differ from those of the underlying hardware. CPUID leaves 40000000h to 4FFFFFFFh are not implemented in hardware, and are reserved for use by hypervisors to provide hypervisor-specific identification and feature information through this interception mechanism.

For leaf 40000000h, the hypervisor is expected to return the index of the highest supported hypervisor CPUID leaf in EAX, and a 12-character hypervisor ID string in EBX,ECX,EDX (in that order). For leaf 40000001h, the hypervisor may return an interface identification signature in EAX - e.g. hypervisors that wish to advertise that they are Hyper-V compatible may return 0x31237648—“Hv#1” in EAX. The formats of leaves 40000001h and up to the highest supported leaf are otherwise hypervisor-specific. Hypervisors that implement these leaves will normally also set bit 31 of ECX for CPUID leaf 1 to indicate their presence.

Hypervisors that expose more than one hypervisor interface may provide additional sets of CPUID leaves for the additional interfaces, at a spacing of 100h leaves per interface. For example, when QEMU is configured to provide both Hyper-V and KVM interfaces, it will provide Hyper-V information starting from CPUID leaf 40000000h and KVM information starting from leaf 40000100h.

Some hypervisors that are known to return a hypervisor ID string in leaf 40000000h include:

CPUID EAX=40000x00h: 12-character Hypervisor ID string in EBX,ECX,EDX

Hypervisor	ID String (ASCII)	Notes
Microsoft Hyper-V, Windows Virtual PC	"Microsoft Hv"
Linux KVM	"KVMKVMKVM\0\0\0"	\0 denotes an ASCII NUL character.
	"Linux KVM Hv"	Hyper-V emulation
bhyve	"BHyVE BHyVE ", "bhyve bhyve "	ID string changed from mixed-case to lower-case in 2013. Lower-case string also used in bhyve-derived hypervisors such as xhyve and HyperKit.
Xen	"XenVMMXenVMM"	Only when using HVM (hardware virtual machine) mode.
QEMU	"TCGTCGTCGTCG"	Only when the TCG (Tiny Code Generator) is enabled.
Parallels	" lrpepyh  vr"	(it possibly should be "prl hyperv", but it is encoded as " lrpepyh vr" due to an endianness mismatch)
VMware	"VMwareVMware"
Project ACRN	"ACRNACRNACRN"
VirtualBox	"VBoxVBoxVBox"	Only when configured to use the "hyperv" paravirtualization provider.
QNX Hypervisor	"QXNQSBMV"	The QNX hypervisor detection method provided in the official QNX documentation checks only the first 8 characters of the string, as provided in EBX and ECX (including an endianness swap) - EDX is ignored and may take any value.
NetBSD NVMM	"___ NVMM ___"
OpenBSD VMM	"OpenBSDVMM58"
Intel HAXM	"HAXMHAXMHAXM"	Project discontinued.
Intel KGT (Trusty)	"EVMMEVMMEVMM"	On "trusty" branch of KGT only, which is used for the Intel x86 Architecture Distribution of Trusty OS (archive) (KGT also returns a signature in CPUID leaf 3: ECX=0x4D4D5645 "EVMM" and EDX=0x43544E49 "INTC")
Unisys s-Par	"UnisysSpar64"
Lockheed Martin LMHS	"SRESRESRESRE"

EAX=80000000h: Get Highest Extended Function Implemented

The highest calling parameter is returned in EAX.

EBX/ECX/EDX return the manufacturer ID string (same as EAX=0) on AMD but not Intel CPUs.

EAX=80000001h: Extended Processor Info and Feature Bits

This returns extended feature flags in EDX and ECX.

Many of the bits in EDX (bits 0 through 9, 12 through 17, 23, and 24) are duplicates of EDX from the EAX=1 leaf - these bits are highlighted in light yellow. (These duplicated bits are present on AMD but not Intel CPUs.)

AMD feature flags are as follows:

CPUID EAX=80000001h: Feature bits in EDX and ECX

Bit	EDX			ECX		Bit
	Short	Feature		Short	Feature
0	fpu	Onboard x87 FPU		lahf_lm	LAHF/SAHF in long mode	0
1	vme	Virtual mode extensions (VIF)		cmp_legacy	Hyperthreading not valid	1
2	de	Debugging extensions (CR4 bit 3)		svm	Secure Virtual Machine	2
3	pse	Page Size Extension		extapic	Extended APIC space	3
4	tsc	Time Stamp Counter		cr8_legacy	CR8 in 32-bit mode	4
5	msr	Model-specific registers		abm/lzcnt	Advanced bit manipulation (LZCNT and POPCNT)	5
6	pae	Physical Address Extension		sse4a	SSE4a	6
7	mce	Machine Check Exception		misalignsse	Misaligned SSE mode	7
8	cx8	CMPXCHG8B (compare-and-swap) instruction		3dnowprefetch	PREFETCH and PREFETCHW instructions	8
9	apic	Onboard Advanced Programmable Interrupt Controller		osvw	OS Visible Workaround	9
10	(syscall)	(SYSCALL/SYSRET, K6 only)		ibs	Instruction Based Sampling	10
11	syscall	SYSCALL and SYSRET instructions		xop	XOP instruction set	11
12	mtrr	Memory Type Range Registers		skinit	SKINIT/STGI instructions	12
13	pge	Page Global Enable bit in CR4		wdt	Watchdog timer	13
14	mca	Machine check architecture		(reserved)		14
15	cmov	Conditional move and FCMOV instructions		lwp	Light Weight Profiling	15
16	pat	Page Attribute Table		fma4	4-operand fused multiply-add instructions	16
17	pse36	36-bit page size extension		tce	Translation Cache Extension	17
18	(reserved)			(reserved)		18
19	ecc	"Athlon MP" / "Sempron" CPU brand identification		nodeid_msr	NodeID MSR (C001_100C)	19
20	nx	NX bit		(reserved)		20
21	(reserved)			tbm	Trailing Bit Manipulation	21
22	mmxext	Extended MMX		topoext	Topology Extensions	22
23	mmx	MMX instructions		perfctr_core	Core performance counter extensions	23
24	fxsr	FXSAVE, FXRSTOR instructions, CR4 bit 9		perfctr_nb	Northbridge performance counter extensions	24
25	fxsr_opt	FXSAVE/FXRSTOR optimizations		(StreamPerfMon)	(Streaming performance monitor architecture)	25
26	pdpe1gb	Gigabyte pages		dbx	Data breakpoint extensions	26
27	rdtscp	RDTSCP instruction		perftsc	Performance timestamp counter (PTSC)	27
28	(reserved)			pcx_l2i	L2I perf counter extensions	28
29	lm	Long mode		monitorx	MONITORX and MWAITX instructions	29
30	3dnowext	Extended 3DNow!		addr_mask_ext	Address mask extension to 32 bits for instruction breakpoints	30
31	3dnow	3DNow!		(reserved)		31

EAX=80000002h,80000003h,80000004h: Processor Brand String

These return the processor brand string in EAX, EBX, ECX and EDX. CPUID must be issued with each parameter in sequence to get the entire 48-byte ASCII processor brand string. It is necessary to check whether the feature is present in the CPU by issuing CPUID with EAX = 80000000h first and checking if the returned value is not less than 80000004h.

The string is specified in Intel/AMD documentation to be null-terminated, however this is not always the case (e.g. DM&P Vortex86DX3 and AMD Ryzen 7 6800HS are known to return non-null-terminated brand strings in leaves 80000002h-80000004h), and software should not rely on it.

#include <stdio.h>
#include <string.h>
#include <cpuid.h>

int main()
{
    unsigned int regs[12];
    char str[sizeof(regs)+1];

    __cpuid(0x80000000, regs[0], regs[1], regs[2], regs[3]);

    if (regs[0] < 0x80000004)
        return 1;

    __cpuid(0x80000002, regs[0], regs[1], regs[2], regs[3]);
    __cpuid(0x80000003, regs[4], regs[5], regs[6], regs[7]);
    __cpuid(0x80000004, regs[8], regs[9], regs[10], regs[11]);

    memcpy(str, regs, sizeof(regs));
    str[sizeof(regs)] = '\0';
    printf("%s\n", str);

    return 0;
}

On AMD processors, from 180nm Athlon onwards (AuthenticAMD Family 6 Model 2 and later), it is possible to modify the processor brand string returned by CPUID leaves 80000002h-80000004h by using the WRMSR instruction to write a 48-byte replacement string to MSRs C0010030h-C0010035h. This can also be done on AMD Geode GX/LX, albeit using MSRs 300Ah-300Fh.

In some cases, determining the CPU vendor requires examining not just the Vendor ID in CPUID leaf 0 and the CPU signature in leaf 1, but also the Processor Brand String in leaves 80000002h-80000004h. Known cases include:

• Montage Jintide CPUs can be distinguished from the Intel Xeon CPU models they're based on by the presence of the substring Montage in the brand string of the Montage CPUs (e.g. Montage Jintide C2460 and Intel Xeon Platinum 8160 - both of which identify themselves as GenuineIntel Family 6 Model 55h Stepping 4 - can be distinguished in this manner.)
• CentaurHauls Family 6 CPUs may be either VIA or Zhaoxin CPUs - these can be distinguished by the presence of the substring ZHAOXIN in the brand string of the Zhaoxin CPUs (e.g. Zhaoxin KaiXian ZX-C+ C4580 and VIA Eden X4 C4250 - both of which identify themselves as CentaurHauls Family 6 Model 0Fh Stepping 0Eh - can be distinguished in this manner.)

EAX=80000005h: L1 Cache and TLB Identifiers

This provides information about the processor's level-1 cache and TLB characteristics in EAX, EBX, ECX and EDX as follows:

• EAX: information about L1 hugepage TLBs (TLBs that hold entries corresponding to 2M/4M pages)
• EBX: information about L1 small-page TLBs (TLBs that hold entries corresponding to 4K pages)
• ECX: information about L1 data cache
• EDX: information about L1 instruction cache

CPUID EAX=80000002h: L1 Cache/TLB information in EAX,EBX,ECX,EDX

Bits	EAX	EBX		ECX	EDX	Bits
7:0	Number of instruction TLB entries			Cache line size in bytes		7:0
15:8	instruction TLB associativity			Number of cache lines per tag		15:8
23:16	Number of data TLB entries			Cache associativity		23:16
31:24	Data TLB associativity			Cache size in kilobytes		31:24

EAX=80000006h: Extended L2 Cache Features

Returns details of the L2 cache in ECX, including the line size in bytes (Bits 07 - 00), type of associativity (encoded by a 4 bits field; Bits 15 - 12) and the cache size in KB (Bits 31 - 16).

#include <stdio.h>
#include <cpuid.h>

int main()
{
    unsigned int eax, ebx, ecx, edx;
    unsigned int lsize, assoc, cache;

    __cpuid(0x80000006, eax, ebx, ecx, edx);
    
    lsize = ecx & 0xff;
    assoc = (ecx >> 12) & 0x07;
    cache = (ecx >> 16) & 0xffff;

    printf("Line size: %d B, Assoc. type: %d, Cache size: %d KB.\n", lsize, assoc, cache);

    return 0;
}

EAX=80000007h: Processor Power Management Information and RAS Capabilities

This function provides information about power management, power reporting and RAS (Reliability, availability and serviceability) capabilities of the CPU.

CPUID EAX=80000007h: RAS features in EBX and power management features in EDX

Bit	EBX			EDX		Bit
	Short	Feature		Short	Feature
0	MCAOverflowRecov	MCA (Machine Check Architecture) overflow recovery support		TS	Temperature Sensor	0
1	SUCCOR	Software uncorrectable error containment and recovery capability		FID	Frequency ID Control	1
2	HWA	Hardware assert support (MSRs C001_10C0 to C001_10DF		VID	Voltage ID Control	2
3	ScalableMca	Scalable MCA supported		TTP	THERMTRIP	3
4	(reserved)			TM	Hardware thermal control (HTC) supported	4
5	(reserved)			STC	Software thermal control (STC) supported	5
6	(reserved)			100MHzSteps	100 MHz multiplier control	6
7	(reserved)			HwPstate	Hardware P-state control (MSRs C001_0061 to C001_0063)	7
8	(reserved)			TscInvariant	Invariant TSC - TSC (Time Stamp Counter) rate is guaranteed to be invariant across all P-states, C-states and sop grant transitions.	8
9	(reserved)			CPB	Core Performance Boost	9
10	(reserved)			EffFreqRO	Read-only effective frequency interface (MSRs C000_00E7 and C000_00E8)	10
11	(reserved)			ProcFeedback­Interface	Processor Feedback Interface supported	11
12	(reserved)			ProcPower­Reporting	Processor power reporting interface supported	12
13	(reserved)			Connected­Standby	Connected Standby	13
14	(reserved)			RAPL	Running Average Power Limit	14
15	(reserved)			FastCPPC	Fast CPPC (Collaborative Processor Performance Control) supported	15
31:16	(reserved)			(reserved)		31:16

CPUID EAX=80000007h: Processor Feedback info in EAX and power monitoring interface info in ECX

Bits	EAX			ECX		Bits
	Short	Feature		Short	Feature
7:0	NumberOfMonitors	Number of Processor Feedback MSR pairs available, starting from MSR C001_0080 onwards		CpuPwrSample­TimeRatio	Ratio of compute unit power accumulator sample period to TSC counter period.	7:0
15:8	Version	Processor Feedback Capabilities version				15:8
31:16	MaxWrapTime	Maximum time between reads (in milliseconds) that software should use to avoid two wraps.				31:16

EAX=80000008h: Virtual and Physical address Sizes

CPUID EAX=80000008h: Feature bits in EBX

Bit	EBX
	Short	Feature
0	clzero	CLZERO instruction
1	retired_instr	Retired instruction count MSR (C000_00E9h) supported
2	xrstor_fp_err	XRSTOR restores FP errors
3	invlpgb	INVLPGB and TLBSYNC instructions
4	rdpru	RDPRU instruction
5	(reserved)
6	mbe	Memory Bandwidth Enforcement
7	(reserved)
8	mcommit	MCOMMIT instruction
9	wbnoinvd	WBNOINVD instruction
10	(reserved)
11	(reserved)
12	IBPB	Indirect Branch Prediction Barrier (performed by writing 1 to bit 0 of PRED_CMD (MSR 049h))
13	wbinvd_int	WBINVD and WBNOINVD are interruptible
14	IBRS	Indirect Branch Restricted Speculation
15	STIBP	Single Thread Indirect Branch Prediction mode
16	IbrsAlwaysOn	IBRS mode has enhanced performance and should be left always on
17	StibpAlwaysOn	STIBP mode has enhanced performance and should be left always on
18	ibrs_preferred	IBRS preferred over software
19	ibrs_same_mode_protection	IBRS provides Same Mode Protection
20	no_efer_lmsle	EFER.LMSLE is unsupported
21	invlpgb_nested	INVLPGB support for nested pages
22	(reserved)
23	ppin	Protected Processor Inventory Number - PPIN_CTL (C001_02F0) and PPIN (C001_02F1) MSRs are present
24	ssbd	Speculative Store Bypass Disable
25	ssbd_legacy	Speculative Store Bypass Disable Legacy
26	ssbd_no	Speculative Store Bypass Disable Not Required
27	cppc	Collaborative Processor Performance Control
28	psfd	Predictive Store Forward Disable
29	btc_no	Branch Type Confusion: Processor not affected
30	IBPB_RET	IBPB (see bit 12) also clears return address predictor
31	branch_sampling	Branch Sampling Support

CPUID EAX=80000008h: Size and range fields in EAX, ECX, EDX

Bits	EAX		ECX		EDX	Bits
7:0	Number of Physical Address Bits		Number of Physical Threads in processor (minus 1)		Maximum page count for INVLPGB instruction	7:0
11:8	Number of Linear Address Bits		(reserved)			11:8
15:12			APIC ID Size			15:12
17:16	Guest Physical Address Size		Performance Timestamp Counter size		Maximum ECX value recognized by RDPRU instruction	17:16
23:18			(reserved)			23:18
31:24	(reserved)					31:24

EAX=8000000Ah: Secure Virtual Machine features

This leaf returns information about AMD SVM (Secure Virtual Machine) features in EAX, EBX and EDX.

CPUID EAX=8000000Ah: SVM information in EAX, EBX and ECX

Bits	EAX		EBX		ECX	Bits
7:0	SVM Revision Number		Number of available ASIDs (address space identifiers)		(reserved)	7:0
8	(hypervisor)					8
31:9	(reserved)					31:9

CPUID EAX=8000000Ah: SVM feature flags in EDX

Bit	EDX
	Short	Feature
0	NP	Rapid Virtualization Indexing (Nested Paging)
1	LbrVirt	LBR (Last Branch Records) virtualization
2	SVML	SVM-Lock
3	NRIPS	nRIP (next sequential instruction pointer) save on #VMEXIT supported
4	TscRateMsr	MSR-based TSC rate control (MSR C000_0104h)
5	VmcbClean	VMCB (Virtual Machine Control Block) clean bits supported
6	FlushByAsid	TLB flush events (e.g. CR3 writes, CR4.PGE toggles) only flush the TLB entries of the current ASID (address space ID)
7	DecodeAssist	Decode assists supported
8	PmcVirt	PMC (Performance Monitoring Counters) virtualization
9	(SseIsa10Compat)	(reserved)
10	PauseFilter	PAUSE intercept filter supported
11	(reserved)
12	PauseFilter­Threshold	PAUSE filter cycle count threshold supported
13	AVIC	AMD Advanced Virtualized Interrupt Controller supported
14	(reserved)
15	VMSAVEvirt	VMSAVE and VMLOAD virtualization
16	VGIF	Global Interrupt Flag (GIF) virtualization
17	GMET	Guest Mode Execution Trap
18	x2AVIC	x2APIC mode supported for AVIC
19	SSSCheck	SVM Supervisor shadow stack restrictions
20	SpecCtrl	SPEC_CTRL (MSR 2E0h) virtualization
21	ROGPT	Read-Only Guest Page Table supported
22	(reserved)
23	HOST_MCE_­OVERRIDE	Guest mode Machine-check exceptions when host CR4.MCE=1 and guest CR4.MCE=0 cause intercepts instead of shutdowns
24	TlbiCtl	INVLPGB/TLBSYNC hypervisor enable in VMCB and TLBSYNC intercept support
25	VNMI	NMI (Non-Maskable interrupt) virtualization
26	IbsVirt	IBS (Instruction-Based Sampling) virtualization
27	ExtLvtOffset­FaultChg	Read/Write fault behavior for extended LVT offsets (APIC addresses 0x500-0x530) changed to Read Allowed, Write #VMEXIT
28	VmcbAddr­ChkChg	VMCB address check change
29	BusLock­Threshold	Bus Lock Threshold
30	IdleHlt­Intercept	Idle HLT (HLT instruction executed while no virtual interrupt is pending) intercept
31	(reserved)

EAX=8000001Fh: Encrypted Memory Capabilities

CPUID EAX=8000001Fh: Encrypted Memory feature bits in EAX

Bit	EAX
	Short	Feature
0	SME	Secure Memory Encryption
1	SEV	Secure Encrypted Virtualization
2	PageFlushMSR	Page flush MSR (C001_011Eh) supported
3	SEV-ES	SEV Encrypted State
4	SEV-SNP	SEV Secure Nested Paging
5	VMPL	VM Privilege Levels
6	RMPQUERY	RMPQUERY instruction supported
7	VmplSSS	VMPL Supervisor shadow stack supported
8	SecureTSC	Secure TSC supported
9	TscAux­Virtualization	Virtualization of TSC_AUX MSR (C000_0103) supported
10	HwEnfCacheCoh	Hardware cache coherency across encryption domains enforced
11	64BitHost	SEV Guest execution only allowed from 64-bit host
12	Restricted­Injection	SEV-ES guests can refuse all event-injections except #HV (Hypervisor Injection Exception)
13	Alternate­Injection	SEV-ES guests can use an encrypted VMCB field for event-injection
14	DebugVirt	Full debug state virtualization supported for SEV-ES and SEV-SNP guests
15	PreventHostIBS	Prevent host IBS for a SEV-ES guest
16	VTE	Virtual Transparent Encryption for SEV
17	Vmgexit­Parameter	VMGEXIT parameter is supported (using the RAX register)
18	VirtualTomMsr	Virtual TOM (top-of-memory) MSR (C001_0135) supported
19	IbsVirtGuestCtl	IBS virtualization is supported for SEV-ES and SEV-SNP guests
20	PmcVirtGuestCtl	PMC virtualization is supported for SEV-ES and SEV-SNP guests
21	RMPREAD	RMPREAD instruction supported
22	(reserved)
23	(reserved)
24	VmsaRegProt	VMSA (VM Save Area) register protection supported
25	SmtProtection	SMT Protection supported
26	SecureAvic	Secure AVIC supported
27	AllowedSEV­features	ALLOWED_SEV_FEATURES_MASK field in VMCB (offset 138h) supported
28	SVSMComm­PageMSR	SVSM (Secure VM Service Module) communication page MSR (C001_F000h) supported
29	NestedVirt­SnpMsr	VIRT_RMPUPDATE (C001_F001h) and VIRT_PSMASH (C001_F002h) MSRs supported
30	HvInUse­WrAllowed	Writes to Hypervisor-owned paged allowed when marked in-use
31	IbpbOnEntry	IBPB on entry to virtual machine supported

CPUID EAX=8000001Fh: Encrypted Memory feature information in EBX, ECX and EDX

Bits	EBX	ECX	EDX	Bits
5:0	C-bit (encryption enable bit) location in page table entry	Maximum ASID value that can be used for a SEV-enabled guest (maximum number of encrypted guests that can be supported simultaneously)	Minimum ASID value for a guest that is SEV-enabled but not SEV-ES-enabled	5:0
11:6	Physical address width reduction when memory encryption is enabled			11:6
15:12	Number of VMPLs (VM Privilege Levels) supported			15:12
31:16	(reserved)			31:16

EAX=80000021h: Extended Feature Identification 2

CPUID EAX=80000021h: Extended feature bits in EAX

Bit	EAX
	Short	Feature
0	NoNestedDataBp	Processor ignores nested data breakpoints
1	FsGsKernelGsBase­NonSerializing	WRMSR to the FS_BASE, GS_BASE and KernelGSBase MSRs is non-serializing
2	LFenceAlways­Serializing	LFENCE is always dispatch serializing
3	SmmPgCfgLock	SMM paging configuration lock supported
4	(reserved)
5	(reserved)
6	NullSelect­ClearsBase	Null segment selector loads also clear the destination segment register base and limit
7	UpperAddress­Ignore	Upper Address Ignore is supported
8	AutomaticIBRS	Automatic IBRS
9	NoSmmCtlMSR	SMM_CTL MSR (C0010116h) is not supported
10	FSRS	Fast short REP STOSB supported
11	FSRC	Fast short REPE CMPSB supported
12	(reserved)
13	PrefetchCtlMsr	PrefetchControl MSR (C0000108h) is supported
14	(reserved)
15	(reserved)
16	(reserved)
17	CpuidUserDis	CPUID disable for non-privileged software
18	EPSF	Enhanced Predictive Store Forwarding supported
19	(reserved)
20	(reserved)
21	(reserved)
22	(reserved)
23	(reserved)
24	(reserved)
25	(reserved)
26	(reserved)
27	SBPB	Selective Branch Predictor Barrier supported
28	IBPB_BRTYPE	IBPB flushes all branch type predictions
29	SRSO_NO	CPU is not subject to SRSO (Speculative Return Stack Overflow) vulnerability
30	SRSO_USER_­KERNEL_NO	CPU is not subject to SRSO vulnerability across user/kernel boundary
31	SRSO_MSR_FIX	SRSO can be mitigated by setting bit 4 of BP_CFG (MSR C001_102E)

CPUID EAX=80000021h: Extended feature information in EBX

Bit	EBX
	Short	Feature
11:0	MicrocodePatchSize	The size of the Microcode patch in 16-byte multiples. If 0, the size of the patch is at most 5568 (15C0h) bytes
31:12	(reserved)

EAX=8FFFFFFFh: AMD Easter Egg

Several AMD CPU models will, for CPUID with EAX=8FFFFFFFh, return an Easter Egg string in EAX, EBX, ECX and EDX. Known Easter Egg strings include:

Processor	String
AMD K6	NexGen‍erationAMD
AMD K8	IT'S HAMMER TIME
AMD Jaguar	HELLO KITTY! ^-^

EAX=C0000000h: Get Highest Centaur Extended Function

Returns index of highest Centaur leaf in EAX. If the returned value in EAX is less than C0000001h, then Centaur extended leaves are not supported.

Present in CPUs from VIA and Zhaoxin.

On IDT WinChip CPUs (CentaurHauls Family 5), the extended leaves C0000001h-C0000005h do not encode any Centaur-specific functionality but are instead aliases of leaves 80000001h-80000005h.

EAX=C0000001h: Centaur Feature Information

This leaf returns Centaur feature information (mainly VIA PadLock) in EDX. (EAX, EBX and ECX are reserved.)

CPUID EAX=C0000001h: Centaur feature bits in EDX

Bit	EDX
	Short	Feature
0	sm2	SM2 present
1	sm2_en	SM2 enabled
2	rng	PadLock RNG present: XSTORE and REP XSTORE instructions
3	rng_en	RNG enabled
4	ccs	PadLock SM3/SM4 instructions present: CCS_HASH and CCS_ENCRYPT
5	ccs_en	SM3/SM4 instructions enabled
6	xcrypt	PadLock Advanced Cryptographic Engine (ACE, using AES cipher) present: REP XCRYPT(ECB,CBC,CFB,OFB) instructions
7	xcrypt_en	ACE enabled
8	ace2	ACE v2 present: REP XCRYPTCTR instruction, as well as support for digest mode and misaligned data for ACE's REP XCRYPT* instructions.
9	ace2_en	ACE v2 enabled
10	phe	PadLock Hash Engine (PHE): REP XSHA1 and REP XSHA256 instructions
11	phe_en	PHE enabled
12	pmm	PadLock Montgomery Multiplier (PMM): REP MONTMUL instruction
13	pmm_en	PMM enabled
14	(reserved)
15	zx_fma	FMA supported
16	parallax	Adaptive P-state control present
17	parallax_en	Adaptive P-state control enabled
18	overstress	Overstress feature for auto overclock present
19	overstress_en	Overstress feature for auto overclock enabled
20	tm3	Thermal Monitor 3 present
21	tm3_en	Thermal Monitor 3 enabled
22	rng2	RNG v2: Second generation RNG present
23	rng2_en	RNG v2 enabled
24	sem	SME feature present
25	phe2	PHE v2: SHA384 and SHA512 present
26	phe2_en	PHE v2 enabled
27	xmodx	RSA instructions present: XMODEXP and MONTMUL2
28	xmodx_en	RSA instructions enabled
29	vex	VEX instructions present
30	vex_en	VEX instructions enabled
31	stk	STK is present

CPUID usage from high-level languages

Inline assembly

This information is easy to access from other languages as well. For instance, the C code for gcc below prints the first five values, returned by the cpuid:

#include <stdio.h>
#include <cpuid.h>

int main()
{
    unsigned int i, eax, ebx, ecx, edx;

    for (i = 0; i < 5; i++) {
        __cpuid(i, eax, ebx, ecx, edx);
        printf ("InfoType %x\nEAX: %x\nEBX: %x\nECX: %x\nEDX: %x\n", i, eax, ebx, ecx, edx);
    }

    return 0;
}

In MSVC and Borland/Embarcadero C compilers (bcc32) flavored inline assembly, the clobbering information is implicit in the instructions:

#include <stdio.h>

int main()
{
    unsigned int a, b, c, d, i = 0;

    __asm {
        /* Do the call. */
        mov EAX, i;
        cpuid;
        /* Save results. */
        mov a, EAX;
        mov b, EBX;
        mov c, ECX;
        mov d, EDX;
    }

    printf ("InfoType %x\nEAX: %x\nEBX: %x\nECX: %x\nEDX: %x\n", i, a, b, c, d);
    return 0;
}

If either version was written in plain assembly language, the programmer must manually save the results of EAX, EBX, ECX, and EDX elsewhere if they want to keep using the values.

Wrapper functions

GCC also provides a header called <cpuid.h> on systems that have CPUID. The __cpuid is a macro expanding to inline assembly. Typical usage would be:

#include <stdio.h>
#include <cpuid.h>

int main()
{
    unsigned int eax, ebx, ecx, edx;

    __cpuid(0 /* vendor string */, eax, ebx, ecx, edx);
    printf("EAX: %x\nEBX: %x\nECX: %x\nEDX: %x\n", eax, ebx, ecx, edx);

    return 0;
}

But if one requested an extended feature not present on this CPU, they would not notice and might get random, unexpected results. Safer version is also provided in <cpuid.h>. It checks for extended features and does some more safety checks. The output values are not passed using reference-like macro parameters, but more conventional pointers.

#include <stdio.h>
#include <cpuid.h>

int main()
{
    unsigned int eax, ebx, ecx, edx;

    /* 0x81234567 is nonexistent, but assume it exists */
    if (!__get_cpuid (0x81234567, &eax, &ebx, &ecx, &edx)) {
        printf("Warning: CPUID request 0x81234567 not valid!\n");
        return 1;
    }

    printf("EAX: %x\nEBX: %x\nECX: %x\nEDX: %x\n", eax, ebx, ecx, edx);

    return 0;
}

Notice the ampersands in &a, &b, &c, &d and the conditional statement. If the __get_cpuid call receives a correct request, it will return a non-zero value, if it fails, zero.

Microsoft Visual C compiler has builtin function __cpuid() so the cpuid instruction may be embedded without using inline assembly, which is handy since the x86-64 version of MSVC does not allow inline assembly at all. The same program for MSVC would be:

#include <stdio.h>
#ifdef __MSVC__
    #include <intrin.h>
#endif

int main()
{
    unsigned int regs[4];
    int i;

    for (i = 0; i < 4; i++) {
        __cpuid(regs, i);
        printf("The code %d gives %d, %d, %d, %d", regs[0], regs[1], regs[2], regs[3]);
    }

    return 0;
}

Many interpreted or compiled scripting languages are capable of using CPUID via an FFI library. One such implementation shows usage of the Ruby FFI module to execute assembly language that includes the CPUID opcode.

.NET 5 and later versions provide the System.Runtime.Intrinsics.X86.X86base.CpuId method. For instance, the C# code below prints the processor brand if it supports CPUID instruction:

using System.Runtime.InteropServices;
using System.Runtime.Intrinsics.X86;
using System.Text;

namespace X86CPUID {
    class CPUBrandString {
        public static void Main(string[] args) {
            if (!X86Base.IsSupported) {
                Console.WriteLine("Your CPU does not support CPUID instruction.");
            } else {
                Span<int> raw = stackalloc int[12];
             (raw[0], raw[1], raw[2], raw[3]) = X86Base.CpuId(unchecked((int)0x80000002), 0);
             (raw[4], raw[5], raw[6], raw[7]) = X86Base.CpuId(unchecked((int)0x80000003), 0);
             (raw[8], raw[9], raw[10], raw[11]) = X86Base.CpuId(unchecked((int)0x80000004), 0);

                Span<byte> bytes = MemoryMarshal.AsBytes(raw);
                string brand = Encoding.UTF8.GetString(bytes).Trim();
                Console.WriteLine(brand);
            }
        }
    }
}

CPU-specific information outside x86

Some of the non-x86 CPU architectures also provide certain forms of structured information about the processor's abilities, commonly as a set of special registers:

• ARM architectures have a CPUID coprocessor register which requires exception level EL1 or above to access.
• The IBM System z mainframe processors have a Store CPU ID (STIDP) instruction since the 1983 IBM 4381 for querying the processor ID.
• The IBM System z mainframe processors also have a Store Facilities List Extended (STFLE) instruction which lists the installed hardware features.
• The MIPS32/64 architecture defines a mandatory Processor Identification (PrId) and a series of daisy-chained Configuration Registers.
• The PowerPC processor has the 32-bit read-only Processor Version Register (PVR) identifying the processor model in use. The instruction requires supervisor access level.

DSP and transputer-like chip families have not taken up the instruction in any noticeable way, in spite of having (in relative terms) as many variations in design. Alternate ways of silicon identification might be present; for example, DSPs from Texas Instruments contain a memory-based register set for each functional unit that starts with identifiers determining the unit type and model, its ASIC design revision and features selected at the design phase, and continues with unit-specific control and data registers. Access to these areas is performed by simply using the existing load and store instructions; thus, for such devices, there is no need for extending the register set for device identification purposes.

See also

• CPU-Z, a Windows utility that uses CPUID to identify various system settings
• CPU-X, an alternative of CPU-Z for Linux and FreeBSD
• Spectre (security vulnerability)
• Speculative Store Bypass (SSB)
• /proc/cpuinfo, a text file generated by certain systems containing some of the CPUID information
• x86-cpuid.org, a machine-readable CPUID data repository and code generator

Further reading

• "AMD64 Technology Indirect Branch Control Extension" (PDF) (White paper). Revision 4.10.18. Advanced Micro Devices, Inc. (AMD). 2018. Archived (PDF) from the original on 2018-05-09. Retrieved 2018-05-09.

External links

• Intel Processor Identification and the CPUID Instruction (Application Note 485), last published version. Said to be incorporated into the Intel 64 and IA-32 Architectures Software Developer's Manual in 2013, but as of July 2014 the manual still directs the reader to note 485.
  • Contains some information that can be and was easily misinterpreted though, particularly with respect to processor topology identification.
  • The big Intel manuals tend to lag behind the Intel ISA document, available at the top of this page, which is updated even for processors not yet publicly available, and thus usually contains more CPUID bits. For example, as of this writing, the ISA book (at revision 19, dated May 2014) documents the CLFLUSHOPT bit in leaf 7, but the big manuals although apparently more up-to-date (at revision 51, dated June 2014) don't mention it.
• AMD64 Architecture Programmer's Manual Volume 3: General-Purpose and System Instructions
• cpuid command-line program for Linux
• cpuprint.com, cpuprint.exe, cpuprint.raw command-line programs for Windows
• instlatx64 - collection of x86/x64 Instruction Latency, Memory Latency and CPUID dumps