NVIDIA B200 AI Chip: 1000W, 2025 Launch

Dell has unveiled details about the upcoming artificial intelligence (AI) GPU by NVIDIA, codenamed Blackwell. These cutting-edge chips boast a power consumption of up to 1000W, representing a 40% increase over the previous generation, necessitating Dell's innovative engineering to implement effective cooling solutions.

Jeff Clarke, Chief Operating Officer at Dell, announced that in conjunction with the launch of the B200 server GPU, Dell will introduce the flagship PowerEdge XE9680 rack server. Market reports indicate that the NVIDIA B200, compared to the current H100 product, delivers superior computational performance but comes with a staggering power consumption of up to 1000W, exceeding the H100 by more than 40%. Leveraging the Hopper architecture and paired with HBM3e high-bandwidth memory, NVIDIA's H200 chip is recognized as the industry's most powerful AI computing chip. With an estimated computational capability at least twice that of H200, or four times that of H100, the B200 is poised to offer unprecedented performance.

Jeff Clarke anticipates that the forthcoming B200 chip will showcase Dell's engineering prowess in flagship servers, particularly in the realm of liquid cooling systems. However, he notes that the adoption of the Blackwell architecture for B100 and B200 is expected no earlier than 2025, deviating from the previously anticipated timeline.

Built on a custom 4nm process, the NVIDIA H100 represents the current pinnacle of performance. The next-generation GPU from NVIDIA is likely to employ a 3nm process for enhanced performance. Considering the power consumption and cooling requirements, there is a compelling argument that NVIDIA's B100 could become the company's first dual-chip design AI GPU, offering a larger surface area to manage heat generation. Given the industry trend of multi-chip GPU architectures adopted by AMD and Intel, this aligns with the prevailing direction.

In the context of high-performance artificial intelligence and high-performance computing (HPC) applications, performance measured in FLOPS, and the power consumption necessary to achieve these FLOPS while managing heat, are crucial considerations. Dell asserts that these factors are where its technology is poised to surpass competitors, providing insight into Dell's discussion of NVIDIA's next-generation Blackwell GPU.