Moore Threads Launches Powerful New AI GPU and Computing Center to Train Large Models, Partners with Firms to Compete Globally
-
Moore Threads announced new flagship AI GPU, the MTT S4000, with 25 TFLOPs FP32 performance and 200 TOPs INT8 performance.
-
The GPU features 48GB GDDR6 memory and 768 GB/s bandwidth. It uses Moore Threads' 3rd generation architecture.
-
Moore Threads revealed its KUAE Intelligent Computing Center, comprised of clusters with 1,000 S4000 GPUs each for large language model training.
-
KUAE supports models like GPT and frameworks like DeepSpeed. A cluster can train a 70 billion parameter model in 33 days.
-
Moore Threads partnered with companies like Lenovo to establish an alliance supporting its hardware and software ecosystem. This helps the company compete with US firms amidst export sanctions.