[IND] 7 min readOraCore Editors

SPEC CPU 2017 is entering retirement

SPEC CPU 2017 is being retired in 2026, while V1.1.9 adds cheaper academic pricing, Linux on RISC-V support, and better config reporting.

Share LinkedIn
SPEC CPU 2017 is entering retirement

SPEC CPU 2017 is being retired in 2026 after a final overlap with SPEC CPU 2026.

SPEC’s SPEC CPU 2017 page now reads like a product update and an end-of-life notice at the same time. The suite still matters because it is one of the most widely used CPU benchmarks for comparing processor, memory, and compiler performance, but the clock is ticking: SPEC has already published the retirement schedule tied to SPEC CPU 2026.

That matters for anyone publishing results, buying licenses, or tracking server performance over time. The latest maintenance release, V1.1.9, also brings lower academic pricing, initial Linux on RISC-V support, and better reporting of system configuration.

ItemValue
SPEC CPU 2017 new-customer price$1000
Qualified nonprofit price$250
Accredited academic price$50
Benchmarks in the package43
V1.1.9 release date12/01/2022
Independent submissions allowed untilAugust 10, 2026 03:00 AM US Eastern Time

What SPEC CPU 2017 actually measures

Get the latest AI news in your inbox

Weekly picks of model releases, tools, and deep dives — no spam, unsubscribe anytime.

No spam. Unsubscribe at any time.

SPEC built CPU 2017 for one job: compare compute-heavy systems using workloads derived from real applications. The suite focuses on processor speed, memory subsystem behavior, and compiler quality, which makes it useful for vendors, cloud teams, and anyone trying to understand why one machine finishes a task faster than another.

SPEC CPU 2017 is entering retirement

The package ships as source code, so it is not a simple click-and-run benchmark. Users compile the workloads, run them through shell or command prompt commands, and then submit results in SPEC’s required formats. That design keeps the benchmark closer to how real systems are tuned in production, where compilers and flags matter almost as much as raw silicon.

  • 43 benchmarks are included in the suite.
  • SPECspeed 2017 Integer and SPECspeed 2017 Floating Point measure time to finish a single task.
  • SPECrate 2017 Integer and SPECrate 2017 Floating Point measure throughput, or work done per unit of time.
  • An optional metric tracks energy consumption.

That mix is why SPEC CPU 2017 keeps showing up in CPU reviews, server launch decks, and academic papers. It is less about synthetic peak numbers and more about how a system behaves under sustained, compiler-driven workloads.

V1.1.9 adds pricing changes and RISC-V support

The December 1, 2022 release of SPEC CPU 2017 V1.1.9 did not change the benchmark’s core purpose, but it did make the package easier to use in a few important cases. SPEC lowered academic pricing, added initial support for Linux on RISC-V, and improved reporting of system configuration.

Those changes sound modest, but they point to where the benchmark is being used. Lower academic pricing matters for universities and research labs that need repeatable performance data without paying enterprise rates. Linux on RISC-V support matters because the ecosystem is moving from research hardware into more serious deployment and validation work.

“RISC-V is a new and open instruction set architecture that is gaining momentum,” said Calista Redmond, CEO of the RISC-V International Foundation, in a 2022 organization announcement.

That quote helps explain why SPEC’s support update is worth noticing. Benchmark vendors rarely add architecture support for vanity reasons; they do it because users are asking for credible numbers on new platforms.

SPEC also improved reporting around system configuration, which sounds boring until you have tried to compare two benchmark submissions and found that one hides a detail you need. Better configuration reporting makes published results easier to audit and compare.

The retirement schedule is strict

SPEC is not just updating CPU 2017; it is actively winding it down. The retirement plan is tied to the publication of SPEC CPU 2026, and the deadlines are precise.

SPEC CPU 2017 is entering retirement

Here is the practical version:

  • Until August 10, 2026 03:00 AM US Eastern Time, SPEC CPU 2017 and SPEC CPU 2026 results can be submitted independently.
  • From August 10, 2026 03:01 AM to November 2, 2026 03:00 AM, publication of CPU 2017 results on SPEC’s site requires a matching CPU 2026 result on the same configuration.
  • On November 2, 2026 03:00 AM, SPEC stops accepting CPU 2017 results for publication.
  • By November 16, 2026, SPEC retires CPU 2017 and ends technical support.

That schedule matters because benchmark submissions often feed product launches, procurement decisions, and review cycles. If a vendor wants a CPU 2017 result on SPEC’s site late in the transition, it has to pair that result with CPU 2026 data on the same system. For SPECrate submissions, the same number of base copies must be used, even if compilers and tuning differ.

SPEC also makes one thing clear: retirement affects public publication, not internal use. Licensees can still use the benchmark privately, but any public post-retirement publication has to disclose that CPU 2017 has been retired and follow SPEC’s fair-use rules.

Why this matters for buyers and benchmark watchers

If you buy servers, GPUs, or cloud instances, benchmark retirement can change how you compare older data with fresh results. A CPU score from 2024 will not disappear, but the ecosystem around it will shift toward CPU 2026, and that changes what reviewers emphasize.

For teams tracking performance across generations, the best move is to preserve the context around each result: compiler version, tuning flags, base copy counts, and the exact system configuration. SPEC’s newer reporting improvements help with that, but only if publishers include the data cleanly.

For academic users, the pricing change lowers the barrier to entry. A $50 academic license is a very different proposition from a $1000 commercial one, especially for labs that need repeated runs across multiple machines.

The benchmark still has a long tail. SPEC says rule-compliant results can be published independently outside the SPEC website, and that means CPU 2017 will keep showing up in papers, vendor docs, and internal reports even after SPEC stops hosting new submissions. The question is whether those results will still be useful to readers who are already comparing systems against CPU 2026.

My read: if you maintain benchmark baselines, this is the year to start dual-tracking CPU 2017 and CPU 2026 on the same hardware. Waiting until the August 2026 deadline will leave too little time to fix tuning differences, rerun flaky workloads, or clean up documentation before SPEC closes the door on publication.