Pavilion Data takes pole position in GPUDirect race

Pavilion Data, the high-end storage array maker, sends data to an Nvidia DGX-A100 GPU server faster than DDN, VAST Data and WekaIO, according to a Nvidia-validated test result.

Developed by Nvidia, GPUDirect is a data transmission scheme that provides a direct path between storage and GPUs. This eliminates or reduces I/O bottlenecks for data intensive AI applications.

The Pavilion-Nvidia performance is revealed in a video. [Find it on a Pavilion partner resource webpage entitled “Nvidia Pavilion Summary of GPU Direct Testing”.] Pavilion sent block data to the DGX-A100 at 182GB/sec read and 149GB/sec write. File data transmission speeds were 191GB/sec read and 118GB/sec write. 

Pavilion Nvidia video screen grab.

Blocks & Files now has four performance results for storage systems using GPU Direct to send data to an Nvidia DGX-A100 and they are:

Pavilion is clearly top of the DGX-A100 charts, coming in 10 per cent faster than DDN and VAST, and 26 per cent faster than WekaIO.

A little bit more about Pavilion

Pavilion Data’s 4RU HyperParallel Flash Array (HFA) has up to 20 controllers linked across NVMe to up to 72 SSDs, with NVMe-oF external access, providing end-to-end NVMe performance.

Blocks & Files schematic of Pavilion system.

The company’s HyperOS 3.0 provides the HyperParallel File System with multi-chassis, global namespace support for NFS and S3. It exhibits linear scaling across an unlimited number of clustered HyperParallel Flash Array systems.