Troubleshooting fixes landed: HBM bad-block masking at boot (cmp-badblk, 8 MB cost), 1300 MHz SM cap for the MoE-row-corruption card, 180 W cap for the 250 W drop-offs, reflow for x8-afte...

  • Swimlane: Default swimlane
  • Column: Done
  • Position: 23
  • Assignee: not assigned
  • Creator: Ivan
  • Started:
  • Created: 10/30/2026 09:15
  • Modified: 10/30/2026 09:15
  • Moved: 10/30/2026 09:15
  • 8 GB
Description

Full title: Troubleshooting fixes landed: HBM bad-block masking at boot (cmp-badblk, 8 MB cost), 1300 MHz SM cap for the MoE-row-corruption card, 180 W cap for the 250 W drop-offs, reflow for x8-after-cap-mod, libglvnd ICD fix, TP>1 pipeline hand-off fix (vLLM PR #60477), custom all-reduce lock-hang fix (Morrowmake f5ba3755), guest BAR1 repair hook rules

Reported by / on it: @zombielurker, @reala7096, @moliminous, @anon33anon, @alanoo, @darkmaniac7, Morrowmake

Source: https://github.com/vllm-project/vllm/pull/60477

Wiki page: Known bad combinations; Recovery

Cards: 8GB

Card D-023, seeded from the cmpunlocker Discord with credit (reference github:vllm-project/vllm).

Sub-Tasks
Internal links
Comments