Benchmarking GLM 5.3 Flash NVFP4 across GPU combos to probe P2P boundary effects
TheZachMueller · x · 2026-10-27
TheZachMueller is spinning up AIPerf 8k/1k 300s tests on GLM 5.3 Flash NVFP4 across all GPU combinations (except ADA), aiming to see what happens when GPUs must cross P2P boundaries.
He notes max cc tops out around 41 before running out of RAM, so tests run uncapped up to 32.
More from Infra
- Lovable invokes 753.6B Cloudflare Workers monthly, migrated from Next.js to TanStack Start — ritakozlov · 2026-10-27
- Prague lab BottleCapAI releases ThinkingCap models cutting thinking tokens up to 69% with negligible accuracy loss — JFPuget · 2026-10-27
- Local CPU vector mesh beats BM25 routing 100k tools: 70.7% top-1 at 9.6ms — Key_Coconut_9269 · 2026-10-27
- 4 coding agents on one RTX 3090: custom CUDA engine hits ~290 tok/s with Qwen3.8-27B — Adorable_Weakness_39 · 2026-10-27
- Google ships llm-d 1.0, an open Kubernetes-native inference stack, demoed on 1,000 agents — dl_weekly · 2026-10-27
- Transformers v5.21 ships FP8 KV-cache, MoE sharding, and new TTS models — LysandreJik · 2026-10-27