Nvidia shorted over cuBLAS not using asymptotically optimal algorithms
basedjensen · x · 2026-08-21
A humorous tweet claiming to short Nvidia stock after discovering that the cuBLAS matrix multiplication algorithm isn't asymptotically optimal. While a joke, it reflects the intense market scrutiny on even minute details of Nvidia's software stack.
More from Infra
- LLMRouter 2.0: Unified Infrastructure for LLM Routing Dev and Eval — youjiaxuan · 2026-08-21
- llama.cpp adds tensor split for LFM2/MoE, boosting inference performance significantly — pmttyji · 2026-08-21
- Used RTX 3090 purchase review: Local AI performance crushes 3060, Qwen 35B 20x faster — Yanzihko · 2026-08-21
- 456 tok/s Qwen 3.8 on Modded RTX 2080 Ti via NInfer Port — xrailgun · 2026-08-21
- Helium Mirrors Removed from Codeberg Amid Criticism of Policy — uwukko · 2026-08-21
- Optimizing Qwen3.8-27B: From 9.5 to 153 Tokens/Second — TrifleHopeful5418 · 2026-08-21