DeepSeek Open-Sources Six Tools to Help Huawei Ascend Chips Replace Nvidia
pstAsiatech · x · 2026-09-30
Per SCMP, DeepSeek has released a suite of six open-source tools — including TileLang, DeepGEMM Ascend (a matrix-multiplication kernel library) and DeepEP Ascend (efficient large-scale cross-device communication) — tailored for high-throughput training and inference on Huawei Ascend chips, aiming to build an "independent" software ecosystem for AI processors and help Huawei chips supplant Nvidia.
Related event: DeepSeek Open-Sources Full Ascend Infrastructure Stack, Taking Aim at CUDA(13 posts)→
More from Infra
- Vital Optics reportedly lands 6-inch InP substrate orders from a leading global customer — zephyr_z9 · 2026-09-30
- DGX Spark Handbook: how a 128GB low-bandwidth box now matches cloud inference speed — mervenoyann · 2026-09-30
- Cloudflare rebuilds Containers for agent sandboxes: 6x faster startups, 648ms median — michellechen · 2026-09-30
- Autonomous launches 4x RTX 5090 local AI workstation at $43,900 with 128GB+ VRAM — dee_hw · 2026-09-30
- Dev warns: running agents that know your whole life on someone else's computer — SuhailKakar · 2026-09-30
- AI bots are aggressively mining library websites, straining open-access resources — _akpiper · 2026-09-30