DeepSeek open-sources full Ascend software stack, key benchmarks near hardware limits
lxfater · x · 2026-09-30
DeepSeek has open-sourced a full set of Ascend (Huawei) foundational components, mirroring its earlier NVIDIA stack: an Ascend version of TileLang (high-level DSL over Ascend C for kernel authoring), DeepGEMM (matmul acceleration), DeepEP (large-scale cross-card communication), TileKernels (vector/memory ops), FlashMLA (sparse attention for long context), and DeepSelect (efficient data filtering).
Officially, every TileLang op used in DeepSeek training has a high-performance Ascend implementation, with several key benchmarks approaching hardware limits. The move targets the real bottleneck of domestic AI chips — the software ecosystem, long NVIDIA's CUDA moat.
Related event: DeepSeek Open-Sources Full Ascend Toolchain, Taking Aim at CUDA(17 posts)→
More from Infra
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16
- Stratechery's Fictional Obituary: Imagining Jensen Huang's 2026 Exit From Nvidia — Gregory_C_Allen · 2026-11-16
- vLLM serves DeepSeek V4.1-Flash at 954K tokens/s per GPU, $38.96/GPU-hour on GB300 NVL72 — bookwormengr · 2026-11-16
- Colossus 2's 1.2M B200s could generate ~$400B/year in tokens at DeepSeek Flash rates — bookwormengr · 2026-11-16
- BIS issues rule blocking remote access to US AI chips via third-country data centers — teortaxesTex · 2026-11-16
- Qdrant 1.20.1: 50% fewer cold-storage reads, ~2x faster quantization, agent-ready setup — qdrant_engine · 2026-11-16