SPEC CPU®2026 System Requirements

This document describes the system requirements for the SPEC CPU®2026 Benchmark Suite, a product of the SPEC® non-profit corporation (about SPEC). The SPEC CPU 2026 benchmark product includes source code and data sets for 52 benchmarks, organized into 4 suites:

You may choose which suite(s) you would like to run (there is no requirement to run all of them), and your choice affects hardware requirements: both memory and disk space differ between SPECspeed and SPECrate.

Having chosen a suite, if you will use your results in public, then you must run all the benchmarks in the suite (exceptions) and produce at least the base metric. The peak metric is optional. If producing both base and peak, you will need more disk space.

SPEC supplies toolsets for ARM, Power ISA, RISC-V, or x86.
Limitations apply:

Although SPEC CPU suites are intended to be useful with a wide range of chip architectures, in some cases it is possible that you may find that your chip is not compatible with the available toolsets (for example, if your chip is too old to run one of the supported OS+chip combinations). See the section on Supported Tools.

You can use SPEC CPU 2026 to measure the compute performance of a system with physical or virtual CPUs/processors. You can choose to measure all of the processors on a system, or a subset.

Typical: full system, all its CPUs

Usually SPEC CPU has been used to measure entire systems, with all of the physical CPU chips.

Alternatives: VM, partition, zone,...

There is no prohibition against using SPEC CPU 2026 to measure a subset of the CPU resources on a system, for example with a Virtual Machine (VM), partition, zone, domain, container, processor set, core, hardware thread, or other subset.

If you use a subset (such as a VM), it must have enough memory and disk. For public results, follow the usual rules. (Examples: use only methods that are documented, supported, and generally available to customers; fully disclose what you do, with sufficent detail so that the result can be reproduced; and if you enhance performance by doing something outside the subset -- such as temporarily silencing all the "noisy neighbor" VMs -- fully disclose that, too.)

How many CPUs? [link].

You can use SPEC CPU 2026 to measure performance with arbitrarily large numbers of processors. Usually, if your operating system can handle it, the SPEC CPU toolset can also handle it; if not, contact SPEC. (Of course, you must meet the other requirements of this document, such as sufficient disk and memory for your intended test.)

SPECrate suites run multiple identical jobs at the same time.

Typically, your OS spreads them across CPUs using capabilities known as SMP (Symmetric MultiProcessing).

SPECspeed suites run a single job at a time.

Most of the SPECspeed benchmarks can use multiple CPUs via OpenMP or language parallel features: C++ std::thread and Fortran doconcurrent.

The above points are summarized in the table below

SPEC CPU 2026 use of multiple CPUs
	SPECspeed	SPECrate
Copies	Only one copy of each benchmark is run at a time.	Multiple copies of each benchmark run simultaneously.
Compiler Parallelism	Most may be compiled using OpenMP and/or language parallelism.	OpenMP and compiler autopar are forbidden.
How To	At build time, enable OpenMP. Use runcpu --threads=N or the corresponding config file feature.	Use runcpu --copies or the corresponding config file feature.

$ ulimit -a | grep stack
stack size   (kbytes, -s) 8192 <-- too small
$

ulimit -s unlimited

The nominal memory requirements are based on pre-release testing by SPEC using Linux with GCC and LLVM which measured the SPECrate benchmarks at 1.8 GB and SPECspeed at 56.2 GB virtual / 49.3 GB RSS, thus leaving a small margin for overhead vs. the nominal 2 GB and 64 GB.

Some SPECspeed benchmarks may need large stacks, both for the main process (typically controlled via ulimit -s) and for threads (typically controlled through your config file setting of environment variables). The SPEC CPU 2017 FAQ item for 627.cam4 may also be relevant for some testers of CPU 2026.

The disk space recommendations below are only estimates. Your environment may differ.

Installation Type	Intended Usage	Detail: observations from one user's system. Yours may differ!
Development 1 TB + 5 GB / copy	Install, compile, and run on a single system. Do tuning runs. Test compiler options. Keep 2 to 4 sets of builds and runs. Eventually you will still need clean up.	11 GB Install SPEC CPU 2026 suites 101 GB Build base+peak for all 4 suites 140 GB Reportable directories, base+peak, 4 suites, 1 copy 2.4 GB Add another copy to intrate and fprate, base+peak
Reportable Basic 250 GB + 4 GB / copy	Install binaries built elsewhere. Use --reportable for 1 suite (intrate, fprate, intspeed, or fpspeed) Requires clean up between suites. Cannot select multiple suites with one runcpu command	11 GB Install SPEC CPU 2026 suites 20 GB Install benchmark binaries (base+peak) for all 4 suites 94 GB Reportable directories, base+peak, 1 suite, 1 copy 2 GB Add another copy, base+peak, 1 suite
Minimized 70 GB + 0.2 GB / copy	Cannot do reportable runs. Cannot build a full suite (intrate, fprate, intspeed, fpspeed) Can build and run 1 benchmark (e.g. 766.femflow_r) Must use minimize_builddirs. Must runcpu --action=clobber frequently	11 GB Install SPEC CPU 2026 suites 21 GB Build and run largest single benchmark, minimize_builddirs, base only 0.1 GB Add another copy, base only, largest single benchmark
The "Detail" column on the right shows actual disk space usage on one particular system. Your usage may differ, due to hardware, operating system, disk type, file system type, compiler, and, especially, compiler tuning. The far left column is a factor of 2x to 4x the detail column. Once you know the space consumption pattern for your hw/sw, you might adjust the above calculations to be more accurate for your needs.

The SPEC CPU 2026 toolset relies on several open source components, including GNU Make, Perl, and others. SPEC supplies pre-built versions of these for particular combinations of hardware and operating system, as shown in the table below.

Supported Toolsets for SPEC CPU 2026 V1.0.0
Toolset	Intended Use
linux-aarch64	64-bit AArch64 systems running Linux.
linux-ppc64le	64-bit little-endian PowerPC systems running Linux.
linux-riscv64	64-bit RISC-V systems running Linux.
linux-x86_64	x86_64 systems running Linux.
macos-arm64	Apple Silicon-based Mac systems running macOS 13 or later.
macos-x86_64	Intel-based Mac systems running macOS 13 or later.
windows-x64	64-bit Intel x86 systems running Windows 10 22H2 or later.
Limitations apply: Although SPEC has tested the above, it is possible that you may enounter an OS+chip combination that is not compatible with its intended toolet. In such cases: Please ensure that your OS+chip meets the minimum required version. Please check the list of Frequently Asked Questions If the tools still fail, please consult Technical Support.

What about other systems? For systems that are not listed in the table of supported toolsets:

• For Android systems, please check the Android install guide
• It is possible that the tools might work, but SPEC has not tested them.

• If the tools do not work, it might or might not be possible for you to build them yourself.

  If you would like to try to build the tools, please see the document Building the SPEC CPU2026 Toolset.
  SPEC may be able to provide advice for your build; however, it will not always be practical for SPEC to do so.
  You might not succeed.

SPEC supplies the benchmarks in source code form. Therefore, you will need either:

1. A set of compilers. If you want to run all the available metrics, you will need all three:
   1. C18
   2. C++2017
   3. Fortran 2018

   --or--

2. A pre-compiled set of benchmark binaries (executables), given to you by another user of the same revision of the SPEC CPU 2026 benchmark suite.
   Note that SPEC does not supply binaries. As mentioned above, SPEC supplies source code.
   Binaries generated by a different revision of the SPEC CPU 2026 benchmark suite will not work.
   You might also need a set of run-time libraries for those executables.
   Note that even if someone else compiles the benchmarks, the tester is responsible for compliance with the rules: do not accept binaries from untrusted sources.

Config file

• To compile the benchmarks, you will need a config file: a file that defines how to build, run, and report on the SPEC CPU benchmarks in a particular environment.
• You might be able to use or modify one of the Example config files that are included with your installed copy of the SPEC CPU 2026 benchmark suite, or a config file from a result published at www.spec.org/cpu2026. The installation guides provide guidance (unix, windows).
• Some compilers might not have working examples, and you may need to write your own config file using the documentation.
• Occasionally, it might not be possible to create a working config file because a compiler does not support features that are used in the benchmarks. If that occurs, you will need to discuss such problems with the compiler vendor, not with SPEC.

You should be familiar either with basic shell commands for Unix (ls, cp, mkdir ...) or basic commands for the Windows Command Prompt (dir, copy, mkdir...).

You will need access to the SPEC CPU 2026 installation media, typically as an ISO image. The Installation Guides (Unix Windows) explain how to use it.

On some systems, the mount command may require privileges or may require additional software. In such cases, you might need to burn a physical DVD using some other system; or, you might need to use the procedure described in the appendix to the Unix installation guide to extract a tarball and use that instead.

Please note that the SPEC CPU 2026 license agreement does not allow you to post the SPEC CPU 2026 software on any public server. If your institution has a SPEC CPU 2026 license, then it's fine to post it on an internal server that is accessible only to members of your institution.