AMD’s MI300X and MI350 win praise as Anthropic’s reported MI450 deal signals a CUDA challenge
tzmartin · x · 2026-07-23
A Grok-quoted comment says AMD’s MI300X and MI350 series are genuinely competitive for memory-heavy LLM inference and large-batch workloads, thanks to 192GB HBM3, higher bandwidth, and often better price/performance.
It also argues that while CUDA still wins on software ecosystem and real-world throughput for many teams, ROCm has narrowed the gap enough that Anthropic’s reported MI450 deal — up to 2GW from 2027 plus AMD’s $5B investment — shows frontier labs now view AMD as a serious alternative for scaling compute.
Related event: AMD and Anthropic Secure Multi-Billion Dollar AI Partnership(63 posts)→
More from Infra
- Anthropic's ~11 GW contracted compute estimated at ~42% NVIDIA GPUs, ~4.5 GW Google TPU — ns123abc · 2026-10-27
- EdgeRock-4B holds KV cache at 72 MiB across 256K context, 99.1% smaller than Qwen3.5-4B — alifcoder · 2026-10-27
- RTX 5080 price jumps from $1658 to $2318 in ten days on Amazon — Frizzy-MacDrizzle · 2026-10-27
- ComfyUI on 12GB VRAM: measured peaks show offloading turns 24s into 11 minutes — Impossible-Mail-4619 · 2026-10-27
- Neon rebuilds its docs search entirely on its own stack: Lakebase Search, Functions, AI Gateway — matei_zaharia · 2026-10-27
- PyTorch details 2.12x faster on-device streaming ASR on Arm SME2 via ExecuTorch — PyTorch · 2026-10-27