DeepGEMM Ascend
DeepGEMM Ascend is a port of DeepGEMM for Huawei Ascend NPUs, with an API compatible with DeepGEMM. It provides kernels for BF16, FP8, and FP4 GEMM, MQA logits, and MegaMoE.
What it does
Developers can use it to build and run optimized GEMM and related kernels on Ascend NPUs using DeepGEMM-compatible APIs.
- Availability
- Initial release; supports Ascend 950 devices
- Pricing model
- MIT License
- Deployment
- HUAWEI Ascend NPU; developed and validated on Ascend 950 series
- Privacy
- Not established
- Licence
- MIT License
Capability
Modality
Access
Deployment
Pricing
Target user
Evidence
6 claims- DeepGEMM Ascend is fully API-compatible with DeepGEMM. [1]
It is fully API-compatible with DeepGEMM and supports BF16, FP8, FP4 GEMM, MQA logits, and MegaMoE.
GitHub new repositories. Quoted text is outside CC BY 4.0 and keeps its original rights. - It supports BF16, FP8, and FP4 GEMM, MQA logits, and MegaMoE. [1]
It is fully API-compatible with DeepGEMM and supports BF16, FP8, FP4 GEMM, MQA logits, and MegaMoE.
GitHub new repositories. Quoted text is outside CC BY 4.0 and keeps its original rights. - The initial release supports Ascend 950 devices. [1]
- 2026.09.30: Initial release of DeepGEMM Ascend, with support for Ascend 950 devices.
GitHub new repositories. Quoted text is outside CC BY 4.0 and keeps its original rights. - The project requires a HUAWEI Ascend NPU and CANN 9.20 toolkit. [1]
- HUAWEI Ascend NPU (developed and validated on the Ascend 950 series)
GitHub new repositories. Quoted text is outside CC BY 4.0 and keeps its original rights. - Dense GEMM reaches up to 99.8% of the hardware limit in the stated benchmarks. [1]
Dense GEMM reaches up to **99.8%** of the hardware limit across types:
GitHub new repositories. Quoted text is outside CC BY 4.0 and keeps its original rights. - The repository is released under the MIT License. [1]
This code repository is released under the [MIT License](LICENSE).
GitHub new repositories. Quoted text is outside CC BY 4.0 and keeps its original rights.
Sources
1
1
GitHub new repositoriesOwner-controlled
https://github.com/deepseek-ai/DeepGEMM-Ascend
Retrieved 2026-10-27 01:06 UTC
Verified
Versions
3| Version | Effective | Change | Changed fields | State |
|---|---|---|---|---|
| 3 | 2026-10-14 18:49 UTC | material change | access, deployment | Current |
| 2 | 2026-10-01 07:31 UTC | material change | capability, access, availability, deployment | Current |
| 1 | 2026-09-30 08:24 UTC | publication | None | Current |
In briefs
1- GitHub expands Copilot controls and adds local sandboxing Daily brief 2026-09-30
Original catalog content is licensed under CC BY 4.0, attributed to wowsuchai. Source quotations and linked content keep their original rights.