NetApp and SK Telecom said on June 24 that they had completed a proof of concept demonstrating that AI workloads on virtual machines could achieve at least 99% of bare-metal performance.

The test used NetApp AFX systems, SK Telecom’s Petasus AI Cloud, and NVIDIA GPUDirect Storage.

The companies said the proof of concept achieved throughput of 32.7 GB/s in both virtualized and bare-metal environments.

It also reduced CPU usage by 40% to 50%, helping address the performance overhead typically associated with running AI workloads on virtual machines.

SK Telecom described Petasus AI Cloud as a cloud platform for building and operating AI data centers.

The platform uses virtualization and GPU cluster virtualization to improve resource use and reduce operating costs.

NVIDIA said GPUDirect Storage reduces CPU load by moving data directly between storage and GPU memory instead of routing it through CPU memory.

🔗 Source: NetApp