China Hot Tech @ChinaHotTech
← All posts
Infra · Open Source · Ascend

DeepSeek Open-Sources Ascend Infra: TileLang, DeepGEMM, DeepEP

DeepSeek open-sourced a set of foundational infrastructure components for Huawei Ascend (昇腾) NPUs — TileLang’s Ascend path, DeepGEMM-Ascend, DeepEP-Ascend, TileKernels, FlashMLA, and DeepSelect — announced ~2026-09-30, with code publicly viewable and testable by ~2026-11-01. The release is framed as the Ascend-side counterpart to DeepSeek’s earlier NVIDIA-oriented open components, plus joint work with Huawei on an Ascend 950 128-card supernode*. This is open infra for Ascend stacks — not a retail GPU SKU, and not a claim that CUDA is finished.

· ~7 min read · China Hot Tech

Cool slate, celadon green and warm paper editorial: TileLang abstraction bridging Ascend 950 128-card supernode and open DeepGEMM DeepEP FlashMLA stack — DeepSeek Ascend infra

Illustration: cool slate / celadon / warm paper — TileLang layer over Ascend 950 supernode + open kernel chips. Distinct from LongCat saffron/cream/ink, Iluvatar steel-blue/teal, MiniCPM lilac/jade. Asterisked labels are vendor / press figures.

China’s AI labs have spent years proving they can train strong models when they can get enough GPUs. The harder question — especially under export pressure — has been whether Chinese NPUs can host a software stack that developers will actually write for: languages, compilers, GEMM libraries, collective communication, attention kernels. DeepSeek’s Ascend open-source drop is aimed at that layer. It does not ship a consumer card you can order from Digi-Key; it ships the kind of low-level pieces that decide whether a card that “can run” starts to “run well.”

What shipped

Per DeepSeek’s WeChat announcement (English summary via Geopolitechs, 2026-09-30) and Chinese follow-up coverage (凤凰网风物研报, 2026-11-01), the open set covers:

DeepSeek’s own wording: each of these has a direct counterpart among the components previously open-sourced for NVIDIA platforms, and across key test cases compute and communication performance is “already approaching the limits of the underlying hardware”*. Huawei’s Ascend community post (2026-09-30) additionally cites DeepEP interconnect figures on a UBL128 supernode — Dispatch ~375 GB/s* / Combine ~347 GB/s* — as near hardware ceiling*. Treat those as vendor/community measurements.

Can overseas teams use it today?

As readable, cloneable open source — yes: the GitHub repos are public (DeepGEMM-Ascend is MIT). As something you can run on a laptop or a Western cloud GPU — generally no. Requirements are Ascend 950-class NPUs plus CANN (DeepGEMM-Ascend README: CANN 9.20*, torch_npu, Python ≥3.10). Western readers without Ascend access still get an architecture and ecosystem signal: what DeepSeek thinks it needs to train V4-family workloads on Chinese silicon, and how close the kernels claim to sit to the metal. Teams already on Ascend or evaluating China cloud / NPU stacks can treat the repos as a shorter path from “card can run” to “card runs well.”

Editorial panels: TileLang Ascend language layer, DeepGEMM compute kernels with BF16 99.8% util asterisk, DeepEP and FlashMLA communications — DeepSeek Ascend open infra caveats

Language / compute / comms split + access caveats — the three overseas checks. Asterisks mark vendor / press / Ascend community figures.

128-card Ascend 950 supernode

DeepSeek states that Huawei’s team was deeply involved, and that the two sides jointly worked on a 128-card supernode based on Ascend 950 with deep optimization of both computation and inter-chip communication*. Ascend community materials describe SuperPoD Flex / UBL128 networking (128-card scale-up*) supporting that joint work. By ~2026-11-01, Chinese coverage (ifeng) notes that code in the public projects can be viewed, tested, and iterated — the announcement is no longer vapor.

What changes vs CUDA inertia

Caveats

Bottom line: DeepSeek open-sourced the Ascend counterparts to its NVIDIA-side infra — TileLang, DeepGEMM, DeepEP, TileKernels, FlashMLA, DeepSelect — and co-optimized a 128-card Ascend 950 supernode with Huawei*. For overseas engineers, read it as an architecture and ecosystem signal plus usable code if you already have Ascend access — not as a Digi-Key board, and not as the funeral of CUDA.

Sources