Chinese major models reportedly trained on Blackwell-class chips, though scale disappoints
teortaxesTex · x · 2026-11-13
tszzl claimed that all major Chinese models are being trained on Blackwell-class chips. teortaxesTex added that this implies they all have functioning Blackwell training stacks— unsurprising for DeepSeek given its TileLang/DeepInfer releases explicitly supporting Blackwell and Ascend but not Hopper, but new information for the rest. He still found the training scale disappointing.
More from Infra
- Nvidia Has $36B in Off-Balance-Sheet Backstops for Neoclouds, Plus $20B in Leases — SumitGup · 2026-11-13
- India's pretraining gap isn't GPU smuggling — it's underinvestment, argues researcher — teortaxesTex · 2026-11-13
- Report: Jensen Huang resigns as Nvidia CEO amid -63% cash conversion, $47B CoreWeave debt — SumitGup · 2026-11-13
- Intel partners with RadixArk to optimize SGLang for Intel CPUs and GPUs upstream — ying11231 · 2026-11-13
- Microsoft, AWS, Google and more hit by simultaneous mass outage — zainhas · 2026-11-13
- vLLM pools rack-scale KV cache on Rubin, holding 96.5-97% hit rate at high concurrency — vllm_project · 2026-11-13