Huawei CEO: our AI chip is faster than Nvidia A100 GPU in some tasks

A Huawei executive claims that the company's Ascend 910B chip performs 20 percent better than Nvidia's A100 GPU in some AI tasks. The Huawei Ascend chips are now also said to be purchased by major Chinese companies such as Tencent Holdings and Baidu.

Huawei CEO Wang Tao told the Nanjing World Semiconductor conference that there is no longer much difference between the computing power of the Huawei Ascend 910B and Nvidia A100 when it comes to training AI models. In some tasks, Huawei's AI chip is said to be 80 percent as efficient as Nvidia's A100. “In other tests, the Huawei Ascend 910B appears to be twenty percent faster,” the man claims. Huawei's AI chips will be according to The South China Morning Post bought by major Chinese companies such as Tenscent Holdings, but also by start-ups from the Asian country.

The Nvidia A100 was introduced in May 2020 and is based on the Ampere architecture. The GPU has a surface area of ​​826mm², consists of 54 billion transistors and has 6912 CUDA cores. The A100 is intended for high performance computing, artificial intelligence and other data center applications. Nvidia introduced the successor to the A100 GPU in 2022: the H100. This accelerator is based on the Hopper architecture and contains up to three times more computing power than the A100 in fp16, tf32 and fp64.

Nvidia A100 GPU