fzorb/REDACTED-rig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Tweaked autoconfig for AMD CPUs with < 2 MB L3 cache per thread, again (hopefully the last time)

Browse source
This commit is contained in:
SChernykh 2025-06-19 12:58:31 +02:00
parent d125d22d27
commit aba1ad8cfc
1 changed files with 2 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

3
src/backend/cpu/platform/HwlocCpuInfo.cpp
View file

@ -320,7 +320,8 @@ void xmrig::HwlocCpuInfo::processTopLevelCache(hwloc_obj_t cache, const Algorith
L2_associativity = l2->attr->cache.associativity;
if (L3_exclusive) {
if (vendor() == VENDOR_AMD) {
if ((vendor() == VENDOR_AMD) && ((arch() == ARCH_ZEN4) || (arch() == ARCH_ZEN5))) {
// Use extra L2 only on newer CPUs because older CPUs (Zen 3 and older) don't benefit from it.
// For some reason, AMD CPUs can use only half of the exclusive L2/L3 cache combo efficiently
extra += std::min<size_t>(l2->attr->cache.size / 2, scratchpad);
}