Interesting but a surprising lack of detail on why Linux outperforms Windows and even less on why Zen 5 widens the gap. Does anyone know why this is happening? Is Zen5 harder to schedule for and Linux better at scheduling?
I see chips and cheese noted there's an increase in latency between ccxs in zen5, which I guess would tend to exaggerate poor scheduling decisions - could that explain it?
Anandtech mentions in their review [0] that AMD is shipping a scheduling driver for Windows, with a "game mode" that disables one of the CCDs to prevent games from experiencing high latencies due to threads being spread across dies. So, presumably: yeah, scheduling for this chip is hard and Linux is much better at heterogenous scheduling situations (having been used extensively in e.g. big.LITTLE and NUMA systems for many years certainly can't hurt)!
It's worth noting that AMD had worked with Microsoft on Zen-specific optimizations in the past, too. Ahead of the Zen 2 launch, they touted improvements of up to 15% from Windows scheduler changes for CPU topology awareness [1]. And with Zen 5 having higher cross-CCD communication latency than the previous generations [2], it probably gets an even harsher penalty from poor scheduling.
While the less-than-stellar reviews for the 9950X are already out, AMD's recent strategy of staggering releases for desktop parts (first the regular, then the X3D SKUs) can improve this generation's perception, assuming there are meaningful performance gains they can reach in software for Zen 5 prior to the X3D release.
In many of the benchmarks linked in the OP and in my comment, particularly gaming benchmarks, the high-end SKUs show results that are significantly worse than would be expected considering the raw performance of the CPU cores and the performance of lower-end or previous-generation chips on the same benchmarks. The problem appears to be due to the Windows 11 scheduler doing an especially poor job of deciding how to schedule threads onto the processor's 2 dies. This generation has particularly high latency between the dies, so if an application does a lot of inter-thread communication and Windows spreads those threads across cores on different dies, application performance suffers significantly.
It's probably something that will be fixed soon with software updates, and Linux fares much better. But the result is that launch-day benchmarks are much worse than they "should" be.
I see chips and cheese noted there's an increase in latency between ccxs in zen5, which I guess would tend to exaggerate poor scheduling decisions - could that explain it?