Nvidia plans to release a DGX system and a DGX SuperPOD system featuring its new Blackwell GPU architecture.

While the smaller system will remain air cooled, the SuperPOD makes the leap to liquid cooling. Both are expected to be available later this year.

The Nvidia DGX B200 system is the sixth generation of its air-cooled, traditional rack-mounted DGX design, and includes eight Nvidia B200 Tensor Core GPUs and two 5th Gen Intel Xeon CPUs.

The company claims that the DGX is capable of up to 144 petaflops of AI performance (with the newly-supported FP4 precision), and has 1.4TB of GPU memory and 64TBps of memory bandwidth. This delivers 15x faster real-time inference for trillion-parameter models over the previous generation, it said.

For those seeking more power, Nvidia has the DGX GB200, each of which features 36 GB200 Superchips - themselves each made out of one Grace CPU and two Blackwell GPUs. This system is liquid cooled.

Eight or more of these DGX GB200s form a SuperPOD with 11.5 exaflops of AI supercomputing (FP4) and 240 terabytes of fast memory.

“Nvidia DGX AI supercomputers are the factories of the AI industrial revolution,” said Jensen Huang, founder and CEO of Nvidia.

“The new DGX SuperPOD combines the latest advancements in Nvidia accelerated computing, networking, and software to enable every company, industry, and country to refine and generate their own AI.”