What the leak shows
The result comes from the EPYC 9996, a 256-core, 512-thread server processor built on Zen 6c. Leaker HXL (@9550pro) spotted it scoring 2,847 in the Geekbench 7.1 single-core test at around 4.1 GHz. Wccftech reports the run was in a Dell PowerEdge R9815 on Ubuntu 26.04 LTS. It also lists a multi-core score of 68,427. Wccftech adds that 4.12 GHz is the chip's maximum boost frequency.
The Zen 5 comparison comes from two places:
- HXL's post lists a Zen 5 EPYC 4545P at 3,064 in the same test. That chip is a 16-core part that runs at about 5.45 GHz. At first glance it beats the Zen 6c chip, but it is running at a much higher clock.
- Enthusiast @326powah ran a Ryzen 7 9700X locked at 4.1 GHz. The setup used DDR5-6000 CL34 memory, automatic secondary timings and FCLK at 2,000 MHz, on CachyOS Linux. The chip scored 2,444 single-core, against 2,847 for the EPYC 9996.
Where the 16.5% and 24% figures come from
The fixed-clock test gives the more useful number. The EPYC 9996 scored about 16.5% higher than the 9700X at roughly the same frequency. @326powah's own post makes this claim.
The 24% figure is a different kind of estimate. Neowin says users estimate up to 24% higher per-clock performance for Zen 6c, adjusted for clock speed. That number compares two different server platforms and divides scores by clocks. It does not isolate IPC, and it won't apply equally to every workload. Treat it as a ceiling-style guess, not a measured gain.
Why it doesn't show a Windows-versus-Linux gap
Neowin's table also lists a Windows 11 score of 2,318 for the 9700X. It lists a CachyOS Linux score of 2,483 for the same chip. Two things stop this from being a useful OS comparison:
- The 2,483 figure doesn't match the 2,444 that @326powah's post reports. I couldn't reconcile the two.
- I found no primary record for the Windows number, and nothing shows matched firmware, memory, drivers or background load.
So the headline's "Windows 11 & Linux" framing is stronger than the evidence. The data supports a Zen 6c versus Zen 5 single-thread indication, not an operating-system verdict.
Caveats for anyone planning an upgrade
- It's server silicon. Zen 6c is AMD's dense-core variant. @326powah himself noted that its per-clock performance shouldn't differ much from classic Zen 6, but he said that isn't certain yet. That is an enthusiast's expectation, not an AMD statement.
- It's one synthetic benchmark. Geekbench single-core says nothing definitive about gaming, creative work or sustained loads.
- Desktop specs are unconfirmed. Neowin says the Olympic Ridge desktop family, possibly branded Ryzen 10000, is expected to pair the new cores with higher clocks and up to 24 cores. It also says final specifications remain unconfirmed.
- Mobile leaks point the same way. A separate Geekbench engineering-sample leak of the 10-core Medusa Point mobile chip scored 3,329 single-core and 16,555 multi-core. That is a different chip on a different Geekbench version, so the numbers can't be compared directly with the EPYC result. It is consistent with a generational uplift, though.
The CPPC claim
Neowin says Windows 11 26H2 and the upcoming 27H2 add a CPPC performance priority feature for Zen 6, delivered through chipset driver updates. I couldn't confirm the Windows-version or driver-timing claims with a Microsoft or AMD source. Treat them as reported, not verified.
Bottom line
The leak is promising. Even at a lower clock, Zen 6c comes close to a much faster-clocked Zen 5 server chip. At matched clocks it leads a Ryzen 7 9700X by roughly 16.5% in one test. That's worth watching, but it isn't a reason to skip or rush a Ryzen 9000 purchase. Wait for retail Ryzen silicon and independent tests under Windows 11 before drawing conclusions.
Update: OC3D breaks out Zen 6c’s Geekbench subtest gains (October 24, 2026)
OC3D has now published a per-subtest analysis of the same 4.1 GHz EPYC 9996 versus fixed-clock Ryzen 7 9700X comparison. According to OC3D, Zen 6c’s lead across Geekbench 7’s 15 single-thread tests ranges from 8.2% to 25.7%, rather than appearing as one uniform 16.5% gain.
The largest reported advantages were in HTML5 Browser, Photo Editor, Structure and Motion, and Video Player, while File Compression and Clang showed the smallest deltas. That adds useful texture to the earlier aggregate result: any architectural uplift may vary substantially by workload.
It still does not establish a desktop Ryzen 10000 result. The figures rely on an EPYC server system and a Ryzen desktop test platform, and Geekbench subtests are not substitutes for matched Windows 11 gaming, application, or compiler testing. The 25.7% result should therefore be read as the best reported gain in this limited comparison—not as an across-the-board Zen 6 IPC claim.