Spun out of #1's last undelivered acceptance criterion ("AVX2 / unsupported hardware is surfaced rather than silently failing") — see the disposition comment there mapping the rest to the pithead#797 arc.
RandomX leans on two ISA extensions that old or exotic x86 parts lack, and today rigforge is silent about both:
- AES-NI: without it XMRig falls back to soft AES, roughly 4x slower. A rig on such hardware "works" at a mysteriously bad rate and nothing says why.
- AVX2: without it dataset init is slower (steady-state hashrate is unaffected).
rigforge.sh only sets init-avx2: -1 (XMRig auto) and never checks the CPU flags. Neither setup nor doctor surfaces the condition.
Fix: read the flags line from /proc/cpuinfo (overridable path, same pattern as MEMINFO/LOCKDOWN_FILE), warn from setup's CPU-detect step, and count a missing AES-NI as a doctor issue (AVX2 stays advisory). No x86 flags line (macOS, ARM, sandboxes) stays quiet — unknown never manufactures an issue, same stance as #333's lockdown probe. Never abort: mining on such hardware still works and may be intentional.
🤖 Generated with Claude Code
Spun out of #1's last undelivered acceptance criterion ("AVX2 / unsupported hardware is surfaced rather than silently failing") — see the disposition comment there mapping the rest to the pithead#797 arc.
RandomX leans on two ISA extensions that old or exotic x86 parts lack, and today rigforge is silent about both:
rigforge.shonly setsinit-avx2: -1(XMRig auto) and never checks the CPU flags. Neithersetupnordoctorsurfaces the condition.Fix: read the flags line from
/proc/cpuinfo(overridable path, same pattern asMEMINFO/LOCKDOWN_FILE), warn fromsetup's CPU-detect step, and count a missing AES-NI as adoctorissue (AVX2 stays advisory). No x86 flags line (macOS, ARM, sandboxes) stays quiet — unknown never manufactures an issue, same stance as #333's lockdown probe. Never abort: mining on such hardware still works and may be intentional.🤖 Generated with Claude Code