Just when you thought graphics cards couldn’t get more power-hungry, NVIDIA is gearing up to release a GPU that will make even the most seasoned gamers and data center operators break a sweat. According to Jeff Clarke, the Chief Operating Officer of Dell Technologies, NVIDIA has a 1000-watt GPU in the works, and it’s set to arrive later this year.

Clarke’s revelation came during Dell’s recent earnings call [PDF version], where he expressed excitement about the upcoming H200 GPU and its performance improvements. However, it was his comments about NVIDIA’s forthcoming B100 accelerator and the mysterious “B200” that raised eyebrows.

“We’re excited about what’s happening with the H200 and its performance improvement,” Clarke told investors. “But we feel the same emotion about the B100 and the B200 that happens next year with the B200.”

The H100, NVIDIA’s current flagship GPU, already consumes a staggering 700 watts of power. Its successor, the H200, is expected to debut in the first half of 2024, boasting up to 141GB of HBM3e memory and a mind-boggling 4.8TB/sec of bandwidth. NVIDIA claims the H200 can double the performance of large language models, including the impressive Llama 70B, thanks to its memory stacks.

But it’s the B100 and the enigmatic “B200” that have caught the attention of the tech community. Clarke stated that direct liquid cooling might not be necessary to tame these beasts, even with the B200 consuming an astonishing 1000 watts of power – a 42 percent increase over the H100.

While Clarke’s comments about the B200 raised eyebrows, it’s worth noting that no chip by that moniker appears on NVIDIA’s roadmap shared with investors last fall. However, it’s speculated that Clarke might have been referring to the GB200 Superchip, which combines NVIDIA’s Grace CPU with the B100 GPU.

Assuming no significant changes in power consumption from the Grace CPU in the GH200, the GB200’s thermal design power (TDP) could be in the neighborhood of 1300 watts – a 30 percent increase over its predecessor.

Alternatively, it’s possible that NVIDIA has another card up its sleeve that hasn’t been revealed yet. Details of the GPU giant’s next-gen Blackwell architecture remain scarce.

Despite the power-hungry nature of these upcoming GPUs, Clarke suggested that Dell is exploring alternative cooling solutions beyond direct liquid cooling. He mentioned the company’s expertise in “fluid chemistry and performance, our interconnect work, the telemetry we are doing, the power management work we’re doing” as steps toward managing the thermal demands of dense chips without resorting to direct liquid cooling.

While NVIDIA is set to release two new accelerators this year, analysts warn that supply constraints for GPUs will persist. This is despite reports predicting that NVIDIA could triple its GPU shipments in 2024.

Beyond the new accelerators, NVIDIA’s roadmap also calls for faster, more capable InfiniBand and Ethernet NICs and switches capable of 800Gb/sec of bandwidth per port before the end of the year.