11:50:57 <\x> sech1 but as you see, we have new cpu instructions right
11:51:17 <\x> maybe you can make an optimized code path for zen4 and mix in some avx512 if its worth the trouble
12:01:50 <sech1> there are 500 hundred different instruction in avx512, I need to find useful ones
12:04:00 <sech1> and RandomX uses 128-bit vectors, it doesn't even use avx256
12:04:52 <\x> ah
12:05:00 <\x> theres some cpu flags already on like phoronix
12:05:33 <\x> https://openbenchmarking.org/s/AMD+Ryzen+9+7950X+16-Core
12:05:42 <\x> atleast the ones on /proc/cpuinfo
12:07:47 <sech1> argon2 has avx512 implementation, so RandomX dataset will be generated 0.1 seconds faster :D