[Advertorial] At ISC, Gigabyte is showcasing its latest servers in collaboration with NVIDIA. The latest NVIDIA® H100 GPUs play an important role in this. The GPU builder provides the graphics cards and Gigabyte puts them in state-of-the-art servers.

Gigabyte has a key booth at this year’s annual supercomputing hardware show. This time, the manufacturer is not only focusing on its high-performance server solutions, but also on its new brand name: Giga Computing.

“Giga Computing is part of Gigabyte,” explains EMEA Sales Director Thomas Yen. “The new company gives us the opportunity to better focus on our market while continuing to operate as part of the Gigabyte group.”

Together with partners

Today, Giga Computing is responsible for the design of the latest high-performance HPC servers with the Gigabyte logo. “We work with numerous partners, including NVIDIA®, to bring powerful solutions to the market,” he continues. “We’re trying to be at the forefront of innovation there and have new servers ready before there’s a lot of customer demand.”

To achieve this, Gigabyte packs many accelerators into one server. For some time, the manufacturer has prided itself on its diverse range of HPC servers, ranging from relatively lightweight configurations to configurations with a very high density of HPC accelerators like the NVIDIA® H100. At ISC we see servers with room for four to eight such NVIDIA®A100 or NVIDIA®H100 GPUs.

“The H100 is the most powerful accelerator on the market today,” said Harun Bayraktar, director of the Engineering, Mathematics, and Quantum Computing Libraries at NVIDIA®. “Accelerator is used for many AI-related workloads and they are becoming increasingly important.” Today, you no longer need to be a government or academic institution to benefit from your own HPC servers. In fact, there is hardly a sector left in which AI and HPC cannot play a role.

Close cooperation

NVIDIA® and Gigabyte work closely together to develop modern solutions. After all, a server is more than a motherboard where you randomly click on the right components. Bayraktar: “These and other powerful chips generate a lot of heat. This is a challenge that we have to solve together with our partners.”

The H100 is currently the most powerful accelerator on the market.

Harun Bayraktar, director of the Engineering, Mathematics and Quantum Computing Libraries NVIDIA®

Traditional air cooling is still the most important method today. That’s because data centers are designed to be air-cooled. With existing techniques, it is also possible to keep gigabyte servers with a high accelerator density sufficiently fresh in an air-cooled environment.

Grace Hopper

However, good cooperation between server manufacturers and chip specialists will become even more important in the future. Finally, Gigabyte recently introduced the Superchip Grace hopper. “We also work together with Gigabyte,” says Bayraktar. The chip is an absolute powerhouse, combining an ARM-based Grace CPU with an NVIDIA® Hopper GPU and onboard HBM memory. With a TDP that can range from 450 watts to 1,000 watts, the added value of a server builder is immediately apparent.

At ISC, however, the current focus is on somewhat more traditional HPC servers. Gigabyte and NVIDIA® present various solutions with PCIe and SXM accelerators. These powerful devices are ready today for data centers that need to support the most modern workloads.

This is a commercial contribution in cooperation with Gigabyte.