Evaluating Spec CPU2026

New chip test arrives, and the comments are already roasting its weird old yardstick

TLDR: SPEC CPU2026 is the new big chip test, but the loudest reaction wasn’t about who won — it was about why the scoring is based on an odd, outdated reference machine. Commenters split between distrust of the numbers, nostalgia for old-school benchmark coverage, and obsessive detective work over weird results.

A new industry test for computer chips, SPEC CPU2026, has landed with more workloads and a big promise: make performance testing feel more modern. But the real popcorn moment isn’t the benchmark itself — it’s the community side-eye over the ancient-feeling baseline machine used to score it. One commenter immediately asked the question hovering over the whole thing: if the reference computer is Ampere’s eMAG, a machine the article basically compares to an old prop plane, does that skew the final score? In other words: are these numbers meaningful, or just flattering? That’s the kind of tiny-detail drama hardware fans live for.

Then came the nostalgia wave. One reader sighed that AnandTech used to run these kinds of comparisons across Intel, AMD, and Apple, and now hardly anyone does. Translation: fans aren’t just reacting to the new test — they’re mourning a whole era of nerdy scorecards. It’s giving “benchmark culture is dying” energy.

And of course, no benchmark thread is complete without someone going full detective mode. Another commenter zoomed in on a strange outlier and basically asked whether the code is bloated, badly arranged, or just begging for better tuning tools. That’s the split-screen mood here: half the crowd is mocking the weird scoring yardstick, half is deep in the weeds trying to optimize a cactus-themed program like the fate of civilization depends on it. Somewhere in the middle sits the article’s main result: today’s top Intel and AMD chips are close in many tests, with AMD’s Zen 5 often flexing harder in math-heavy ones. But honestly? The comments turned a benchmark update into a mini soap opera about trust, nostalgia, and extremely online nitpicking.

Key Points

  • SPEC CPU2026 expands the benchmark suite to 52 workloads from 43 in SPEC CPU2017 and increases workload code size to modernize the suite.
  • The article’s tests use Linux and GCC 14.2.0 with `-O3` and native architecture targets after GCC 15.2.0 caused issues.
  • SPEC CPU2026 uses an Ampere eMAG 8180 system as its reference machine with a score of 1.0.
  • In the author’s results, current Intel and AMD desktop CPUs are close in integer performance, while AMD Zen 5 generally leads in floating-point tests.
  • Several CPU2026 workloads, including 706.stockfish, 749.fotonik3d, and 765.roms, compiled with GCC 14.2.0, make use of AVX-512 instructions.

Hottest takes

"What's the practical impact of choosing eMAG as the baseline CPU?" — fweimer
"I don't know any site does this any more" — ksec
"Does the 'cactus' program have a hot region of code that's too large to effectively cache" — jeffbee
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.