NVIDIA has unveiled the NVIDIA DGX GH200, a supercomputer designed to tackle the most demanding artificial intelligence workloads. The DGX GH200 pairs the NVIDIA Grace Hopper Superchip with the NVLink Switch System to provide 144 terabytes of memory accessible to the GPU shared memory programming model at high speed over NVLink. The DGX GH200 is the first supercomputer to break the 100-terabyte barrier for memory accessible to GPUs over NVLink. According to a whitepaper, the new system will also accelerate deep learning recommendation models, terabyte-scale graph neural networks, and large data analytics workloads. NVIDIA is planning to release the system by the end of 2021.
https://developer.nvidia.com/blog/announcing-nvidia-dgx-gh200-first-100-terabyte-gpu-memory-system/