Dell confirms Nvidia Blackwell B100 GPUs shipping in 2024, B200 upgraded variant arriving in 2025.

NVIDIA's next-generation Blackwell GPU lineup will consist of two major accelerators: the B100 scheduled for 2024 and the B200 for 2025, according to information revealed by Dell during its recent earnings call. The company disclosed that NVIDIA has already shared details of this roadmap with its largest partners, who will be operating servers and data centers equipped with this latest hardware. NVIDIA's Blackwell GPUs are expected to launch this year with a formal debut at GTC 2024, with current roadmaps pointing to the B100 (the GPU codename), GB200 (the Superchip platform), and GB200NVL (the interconnected supercomputing platform).

The B100 is anticipated to leverage HBM3e memory technology similar to the Hopper H200 GPUs, while the upgraded B200 variant is expected to feature an even faster version of HBM memory along with higher memory capacities, upgraded specifications, and enhanced features. Samsung is noted as a major memory provider for the Blackwell GPUs. Reports suggest that NVIDIA's Blackwell GPUs might utilize a monolithic design, though this remains unconfirmed.

According to Robert L. Williams, SVP of IR at Dell Technologies, the power density of the Blackwell B100 and B200 GPUs will reach 1,000 watts per GPU. Williams emphasized that direct liquid cooling would not be necessary to achieve this energy density at the B100 level, with the higher power density materializing at the B200 in the following year. He highlighted Dell's engineering capabilities in liquid cooling performance, including advances in fluid chemistry, interconnect work, telemetry, and power management that will enable scaling of these capabilities to handle the significant computational capacity that Blackwell will offer.

The 1,000-watt power specification represents a 200-watt increase over current-generation accelerators, which is a modest increment relative to the expected performance gains of Blackwell. The NVIDIA Hopper H200 and AMD Instinct MI300X currently consume up to 800 watts at peak power. As power requirements for both CPUs and GPUs continue to increase, companies are relying on advanced process and packaging technologies to optimize chip development costs while maintaining higher power efficiency.