DeepSeek reportedly teaching ultra-efficient inference on Huawei Ascend 950 — no first-party evidence yet
teortaxesTex · x · 2026-10-21
Kyle Chan claims DeepSeek is teaching other Chinese AI labs how to run ultra-efficient large-scale inference on Huawei Ascend 950 AI chips. teortaxesTex responds that just how "ultra-efficient" DeepInfer actually runs on Ascend 950 remains an open question — there is no first-party evidence yet. If true, it would mark a milestone for inference efficiency on domestic chips, but the claim is unverified.
More from Infra
- Modal Engineers Detail Sticky Serverless Routing for Real-Time AI Inference — blelbach · 2026-10-21
- EdgeKernelBench: LLM-Generated Kernels Hit 64.5% Correct but Only 25.5% Fast on Edge Backends — encodelab · 2026-10-21
- Meta Open-Sources Dist-MoE, Speeding MoE Training Up to 1.40x on GB300 NVL72 — ezyang · 2026-10-21
- PyTorch accelerator integration group details progress and priorities at PyTorchCon — PyTorch · 2026-10-21
- Baseten ships four major open models day one, hits 668 tok/s on DeepSeek v4.1 Flash — baseten · 2026-10-21
- Report: 55% of China's AI compute goes to producing short dramas and videos — omoril0ver · 2026-10-21