Moore's thread GPU performance breakthrough, comparable to international standards

Moore Threads officially announced that its GPU performance has achieved a new breakthrough and reached international standards. In cooperation with Zhipu AI, we conducted GPU large model adaptation and performance testing, including large model inference and large model pre-training for the KUAE kilo-card smart computing cluster. The test results show that Moore Thread's self-developed full-featured GPU is significantly better than the RTX 3090 and RTX 4090 in inference, and in terms of training, the performance expansion coefficient of the Kua'e Qianka intelligent computing cluster exceeds 90%, and the model computing power utilization rate (MFU) is comparable to the A100 cluster.