China is preparing to rival Western countries in the hardware world. Moore Threads, one of Nvidia’s biggest rivals, has announced its latest graphics processing unit (GPU) focused on artificial intelligence and high-performance computing. Called Moore Threads S4000, the new GPU is almost twice as fast as the company’s previous product.
Moore Threads S4000 GPU specifications
S4000 offers significant improvements compared to the previous generation of Moore Threads, S3000. Foremost among these is FP32 performance, which is crucial for deep learning. Compared to S3000’s 15 teraflops, S4000 is said to be 25 teraflops fast.
It also offers 200 trillion operations per second for INT8 calculations, significantly surpassing Nvidia’s previous generation (Turing GPU). To fuel this power, the S4000 is equipped with 48GB of GDDR6 memory, delivering 768GB/s bandwidth. This represents a 50 percent increase in VRAM compared to the S3000.
The S4000 will be used as the core hardware for Moore Threads’ Intelligent Computing Center. According to the company, the center consists of 125 servers, each equipped with eight S4000 GPUs. That adds up to 1,000 GPUs per cluster.
Moore Threads claimed that a single cluster can train AI models like ChatGPT in about a month. It should be noted that this system was developed in collaboration with some companies such as Lenovo.
Moore Threads’ new GPU is critical for China’s local semiconductor powerhouse. Although still behind Nvidia, AMD and Intel, the domestic S4000 is attracting attention.
So, what do you think about this issue? Don’t forget to share your opinions with us in the comments section!
{{user}} {{datetime}}
{{text}}