AMD MI455X tops Epoch's new inference index; NVIDIA's 12x FLOPS jump decoded in 5 tricks
bookwormengr · x · 2026-11-16
Part 4 of bookwormengr's thread: Epoch AI updated its chip inference-comparison methodology, and the new index (GB300=1) puts AMD MI455X at 2.89 and NVIDIA VR200 at 2.18, while H100 sits at just 0.04 — newer chips offer up to 72x H100's inference capacity. The author explains the 5 engineering tricks replacing Moore's Law as transistors hit 10 atoms wide at 2nm: multi-die tiling, FP8→FP4 precision, NVL72 rack-scale interconnect, fused CPU+GPU superchips, and HBM4 (22 TB/s). He argues this is also Huawei's playbook for catching up while stuck at 7nm.
More from Infra
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16
- Stratechery's Fictional Obituary: Imagining Jensen Huang's 2026 Exit From Nvidia — Gregory_C_Allen · 2026-11-16
- vLLM serves DeepSeek V4.1-Flash at 954K tokens/s per GPU, $38.96/GPU-hour on GB300 NVL72 — bookwormengr · 2026-11-16
- Colossus 2's 1.2M B200s could generate ~$400B/year in tokens at DeepSeek Flash rates — bookwormengr · 2026-11-16
- BIS issues rule blocking remote access to US AI chips via third-country data centers — teortaxesTex · 2026-11-16
- Qdrant 1.20.1: 50% fewer cold-storage reads, ~2x faster quantization, agent-ready setup — qdrant_engine · 2026-11-16