CANN
The software stack accompanying Huawei's Ascend chips, with the full name Compute Architecture for Neural Networks. Analogous in positioning to NVIDIA's CUDA, it provides operator optimization and AI computation acceleration capabilities.
Timeline (last 90 days)
China is restricted in its access to advanced-process chips, so companies such as Huawei have turned to self-developed chip architectures and software stacks (such as CANN), and are exploring clustered approaches to make up for the per-card performance gap.
GPUs have mature software stacks such as CUDA, cuDNN and NCCL, while NPUs mostly rely on vendor-proprietary SDKs (such as Ascend's CANN or the TPU's XLA path), and their cross-platform portability and operator coverage are far inferior to those of GPUs.
Where Huawei Ascend hardware truly lags behind NVIDIA is in software ecosystem; Huawei has built its own CANN ecosystem but still has gaps when it comes to supporting mainstream large models.
CANN includes the operator development framework (TBE/DSL), graph compilation engine (ATC/GE), runtime (ACL), communication library (HCCL, corresponding to NCCL), and model conversion and tuning toolchain.
Huawei Ascend has its own software stack, CANN, but its ecosystem richness is far inferior to that of CUDA.
DeepGEMM Ascend requires the Ascend 950 series, CANN 9.20 toolkit, torch_npu, Python 3.10 or above, and C++20 format support.
The Ascend version of DeepGEMM retains the original API, is developed and validated on the Ascend 950 series, and requires the corresponding CANN toolchain
The gap between Huawei Ascend and NVIDIA is mainly reflected in two dimensions: interconnect bandwidth and software ecosystem, with Ascend's corresponding CANN framework having started relatively late.
DeepSeek-R1's training was completed on Huawei Ascend clusters, and Ascend's CANN platform is growing rapidly in the Chinese market.
Huawei Ascend's software stack CANN/MindSpore still lags behind NVIDIA's CUDA ecosystem in areas such as toolchain maturity.
All Facts (11)
China is restricted in its access to advanced-process chips, so companies such as Huawei have turned to self-developed chip architectures and software stacks (such as CANN), and are exploring clustered approaches to make up for the per-card performance gap.
50%UnverifiedGPUs have mature software stacks such as CUDA, cuDNN and NCCL, while NPUs mostly rely on vendor-proprietary SDKs (such as Ascend's CANN or the TPU's XLA path), and their cross-platform portability and operator coverage are far inferior to those of GPUs.
50%UnverifiedWhere Huawei Ascend hardware truly lags behind NVIDIA is in software ecosystem; Huawei has built its own CANN ecosystem but still has gaps when it comes to supporting mainstream large models.
50%UnverifiedCANN includes the operator development framework (TBE/DSL), graph compilation engine (ATC/GE), runtime (ACL), communication library (HCCL, corresponding to NCCL), and model conversion and tuning toolchain.
50%UnverifiedHuawei Ascend has its own software stack, CANN, but its ecosystem richness is far inferior to that of CUDA.
50%UnverifiedDeepGEMM Ascend requires the Ascend 950 series, CANN 9.20 toolkit, torch_npu, Python 3.10 or above, and C++20 format support.
50%UnverifiedThe Ascend version of DeepGEMM retains the original API, is developed and validated on the Ascend 950 series, and requires the corresponding CANN toolchain
50%UnverifiedThe gap between Huawei Ascend and NVIDIA is mainly reflected in two dimensions: interconnect bandwidth and software ecosystem, with Ascend's corresponding CANN framework having started relatively late.
50%UnverifiedDeepSeek-R1's training was completed on Huawei Ascend clusters, and Ascend's CANN platform is growing rapidly in the Chinese market.
50%UnverifiedHuawei Ascend's software stack CANN/MindSpore still lags behind NVIDIA's CUDA ecosystem in areas such as toolchain maturity.
50%Unverified昇腾配套的软件栈为CANN（Compute Architecture for Neural Networks），提供与PyTorch、MindSpore等框架的适配层
50%