NVIDIA releases liquid-cooled A100 and H100 PCIe accelerators

Among its announced tonight at Computex 2022, NVIDIA revealed that it is getting ready to launch liquid-cooled versions of its high-end PCIe acceleration cards. Offered as an alternative to traditional air-cooled dual-slot cards, liquid-cooled cards have a more compact single-slot chassis for both better cooling and better density. The liquid-cooled A100 will be available in the third quarter, and the liquid-cooled H100 will be available early next year.

While liquid cooling is not new to data centers, it is typically reserved for more customized hardware with extreme cooling and / or density requirements, such as the upcoming generation of high-end NVIDIA H100 (SMX) servers. In contrast, PCIe servers are all about standardization and compatibility. What for graphics cards / server accelerators means dual-slot cards designed for use with forced air cooling in the server chassis. This serves the market segment well, but the TDP of 300 to 350W on these cards means they cannot be thinner and will still be effectively air-cooled – which in turn creates a 4 card limit for standard rack mounted systems.

But times change, and liquid cooling is being implemented in larger capacity data centers, both to keep pace with the cooling of increasingly hotter equipment and to improve the overall energy efficiency of the data center. To this end, NVIDIA will release liquid-cooled versions of its A100 and H100 PCIe cards to provide data center customers with an easy and officially supported path to installing liquid-cooled PCIe accelerators at their facilities.

The cards (pictured above) are essentially A100 / H100 models with a traditional two-slot heat sink replaced by a full-range single-slot water block. Designed to be integrated by server vendors, they use an open-loop design to be used as part of a larger liquid cooling configuration.

But apart from changing the cooling system, the specifications of the cards remain unchanged. NVIDIA does not increase the TDP or clock speed on these cards, so their performance should be identical to traditional air-cooled cards (as long as they do not cause thermal throttling, of course). In other words, these new cards use liquid cooling to improve energy efficiency and density, not performance.

The first card from the gateway will be a liquid-cooled version of the 80 GB A100 PCIe accelerator. It will be available to customers in the third quarter of this year. Meanwhile, a liquid-cooled version of the H100 PCIe is also in development, and NVIDIA expects it to be available in early 2023.

Meanwhile, NVIDIA has partnered with Equinix to qualify a liquid-cooled A100 in its data centers, as well as getting an idea of ​​the real energy savings of the new hardware. Interestingly, NVIDIA reports a significant reduction in overall data center power consumption from the transition to liquid cooling – the 2000 server configuration (4000 A100 card) saw a 28% decrease in total energy demand. According to NVIDIA, this is due to the combination of overall data center energy savings from the switch, including everything from the improved energy efficiency of the graphics card from lower temperatures to the reduced energy requirements of the cooling water compared to working with large chillers. All of this underlines why NVIDIA promotes liquid-cooled hardware as an energy efficiency enhancer for data center operators looking to reduce energy consumption.

And while this first-generation liquid-cooled hardware focuses on performance, this won’t always be the case, according to NVIDIA. For future card generations, the company will also look at liquid cooling to improve performance at current energy levels – possibly by investing data center scale profits back into higher TDP for cards.

Finally, while most of nVIDIA’s announcements (as well as the case study) today are for PCIe cards, nVIDIA also reveals that it is working on official, liquid-cooled designs for its HGX systems as well, which are used to accommodate the company’s more powerful SMX cards. The liquid-cooled HGX A100 is now commercially available, while the liquid-cooled HGX H100 is expected to be launched in the fourth quarter.

Leave a Reply