12:12:10 AMD lineup from Zen 2 to Zen 5 has been tested: https://gist.github.com/SChernykh/6058ecf01c929883b9d19c7eeadc8809 12:12:18 Need more Intel CPUs, preferably new ones 12:12:34 Could you stop cooking please 12:12:56 it's very embarrassing for us mere mortals 12:13:20 interesting to see even Zen2 going that well 12:14:18 Yes, I expected Zen 5 to get better improvement 12:14:28 Maybe I don't fully understand where the actual bottleneck is 12:14:41 But it has to do with memory access, because program size increase helps all Ryzens 12:22:21 AMD uProf works well to gather the statistics on memory or retired ops and delays 12:23:07 though if you want specifics you probably should just create a bunch of programs ahead of time so it can annotate them 12:26:05 there's a bunch more small samples in comments 12:28:55 my weird nerfed Zen2 with just P012 should arrive in a couple of days :) 21:52:16 ### RandomX v2 Benchmark Results 21:52:16 **AMD RYZEN AI MAX+ 395 w/ Radeon 8060S @ 99.86W** 21:52:18 | Algorithm | Hashrate | Relative Speed | Hash/Joule | VM+AES/s | VM+AES/Joule | Relative Work/Joule | 21:52:20 |-----------|----------|----------------|------------|----------|--------------|---------------------| 21:52:22 | RandomX v1 | 15073.65 | 100.0% | 150.94 | 63.22e9 | 633.12e6 | 100.0% | 21:52:24 | RandomX v2 | 14003.05 | 90.0% | 139.40 | 88.10e9 | 877.05e6 | 130.0% | 21:52:26 **Config:** threads=32, affinity=0xFFFFFFFF, init=32 21:52:28 **Stability:** V1 crashes: 0/2, V2 crashes: 0/2 21:53:20 (didnt tweak nothing, lazy, not running headless, lazy, numbers could be better probably, didnt check from the wall yet, lazy)