09:24:56 <sech1> L0 cache, lol what? https://www.techpowerup.com/review/intel-panther-lake-technical-deep-dive/images/compute-and-software-10.jpg
18:29:10 <sech1> I did a RandomX v2 program size test on the record holding 7945HX rig (225 h/s/watt). Here are the results: https://p2pool.io/u/a52e286d757f6ef6/image.png
18:29:33 <sech1> The third column is efficiency (RandomX instructions executed per Joule of energy vs RandomX v1)
18:29:50 <sech1> I think that program size = 384 is much better. Hashrate drop is not that big, and efficiency is at +33%
18:30:10 <sech1> RandomX v1 (with program size 256) is ~20 kh/s on that rig
18:32:35 <sech1> The rig was running with a fixed power limit of 72W the whole time, so efficiency can be calculated and compared directly by comparing hashrates
18:33:20 <sech1> I suggest to use program size = 384 for v2
19:42:16 <eureka_> sech1 sometimes intel refers to uop cache as L0i, but no idea what they mean in that context when there's L0d and L1d both listed ...