Beff Jezos doubles down on Cerebras: GPU backends feel 'stupid' for frontier-scale fast inference
beffjezos · x · 2026-10-31
Responding to critics who accused him of shilling Cerebras without technical depth, Beff Jezos says he uses Astra Ultrafast daily and argues any GPU backend feels "stupid in terms of frontier thoughts per second." He is very bullish on Cerebras-like accelerators and the post-GPU era, claiming "once you go ultrafast you don't go back." The quoted skeptic notes he never explained why Nvidia or Google couldn't match fast-inference positioning with their own products.
More from Infra
- Lawyer runs DeepSeek-V4.1-Flash on 2x DGX Spark + 3090s: 2,618 tok/s prefill at 256K, 95.9% KLD accuracy, all data open — TumbleweedNew6515 · 2026-10-31
- Koboldcpp v1.124 "aesthetic edition" released for local LLM inference — Fcking_Chuck · 2026-10-31
- CXMT validates G6 DRAM engineering samples in Q3, mass production expected within a year — zephyr_z9 · 2026-10-31
- AI scarcity moved to atoms: memory, GPU and wafer margins hit 58%-80% — sujingshen · 2026-10-31
- Anthropic's S-1 risk factors: compute tops the list at 13.4%, edging out regulation — ayushtweetshere · 2026-10-31
- TU Eindhoven shows glass interposer with BGA below and MPO optical port on top — jwt0625 · 2026-10-31