You might know Nvidia as the name behind the GPUs you can’t buy right now but frankly GPUs are perhaps the most boring thing coming out of the company.
What many folks aren’t aware of is that Nvidia is entrenched in the data centre and high performance computing sectors as well.
But the latest announcement from the firm had us doing a double take because Nvidia is making a CPU.
The firm has taken a stab at making a CPU before but the Tegra K1 for mobile devices didn’t exactly set the world on fire.
This time around Nvidia has created an Arm-based CPU it is calling Grace which is set to power the Swiss National Computing Center’s new supercomputer, Alps.
“Using licensed Arm IP, Nvidia has designed Grace as a CPU specifically for giant-scale AI and HPC. Coupled with the GPU and DPU, Grace gives us the third foundational technology for computing, and the ability to re-architect the data center to advance AI. Nvidia is now a three-chip company,” chief executive officer and founder at Nvidia, Jensen Huang said during Nvidia GTC21.
Alps is being built by Hewlett Packard Enterprise using the HPE Cray EX product line. The supercomputer will also make use of Nvidia HGX, Nvidia GPUs, the Nvidia HPC SDK and, of course, the Grace CPU.
The CPU, Nvidia says, will deliver ten times the performance of “today’s fastest servers on the most complex AI and high-performance computing workloads”.
Building on that, Nvidia claims that the combination of its CPU and GPUs will enable Alps to train GPT-3, a natural language processing model, in just two days. This is seven times faster than Nvidia’s 2.8-AI exaflops Selene supercomputer.
The Alps supercomputer is expected to come online in 2023 and it will replace the existing Piz Daint supercomputer. Alps will play a role in advancing research in quantum chemistry, molecular dynamics and particle physics among many others. The supercomputer also uses a newly designed software-defined infrastructure in order to support a wide range of projects.
“One of that team’s goals is to run global climate models with a spatial resolution of 1km that can map convective clouds such as thunderclouds,” wrote Nvidia.
The firm says that Grace arrives at a time where data and AI models are growing exponentially. We’re curious to see how well this CPU performs and whether Nvidia’s CPU play will find its way into more data centres and supercomputers.
[Source – Nvidia]