Huawei adds DeepSeek's R1 to its ModelArts Studio platform, saying the free model is “Ascend-adapted”, referencing its data center GPUs, but offers few details
Nvidia has competition in China. … The tech company explicitly said this version is “Ascend-adapted,” as in Huawei's Ascend data center GPUs.
I feel this should be a much bigger story: DeepSeek has trained on Nvidia H800 but is running inference on the new home Chinese chips made by Huawei, the 910C. [image]
DeepSeek tried to do one thing: transplant its own model to Huawei Ascend 910B chip to run. Through the “dynamic precision adjustment” technology, they only lost 5% of the performance in the same task, but the cost dropped by 70%. [image]
Huawei is betting that inference will be a bigger source of AI hardware demand than training, since training only occurs a few times relative to inference. It's Ascend 910c chip is designed with exactly this in mind. And DeepSeek has already announced support for it. [image]