Nvidia announced its latest GPU in the data center, the Blackwell Ultra.
The chip is built on the Blackwell architecture introduced last year, NVIDIA going to an annual cadence of GPU version from the BOOM of the AI.
Blackwell Ultra is available in the NVIDIA GB300 NVL72 scale solution and the NVIDIA HGX B300 NVL16 system.
The GB300 NVL72 connects 72 GPU Blackwell Ultra and 36 CPU NVIDIA Grace, based in ARM Neoverse, in a single rack cooled by liquid, which, according to the company
The ultra combines two GPUs the size of a reticle, has 15 Petaflops of FP4 performance and 288 GB of HBM3E.
Ultra products should be delivered from the second half of 2025, although it is noted that the Blackwell GPUs were affected by delays and have missed their launch.
Cisco, Dell Technologies, Hewlett Packard Enterprise, Lenovo and Supermicro should offer a wide range of servers based on Blackwell Ultra. Servers are also expected from Avres, Asrock Rack, Asus, Eviden, Foxconn, Gigabyte, Inventec, Pegatron, Quanta Cloud Technology (QCT), Wistron and Wiwynn.
Hyperscalers and neoclouds also plan to offer ultra-fueled instances at Blackwell, including Amazon Web Services, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure, with Coreweave, Crusoe, Lambda, Nebius, Nscale, Yotta and YTL.
Nvidia himself will offer access to the GB300 NVL72 on his DGX Cloud.