Cloudian has released a new PyTorch connector with RDMA support, boosting performance for AI and machine learning workloads.
The PyTorch connector is built on Nvidia's GPUDirect Storage technology and optimized for Nvidia Spectrum-X networking infrastructure, resulting in a 74% increase in data processing performance.
Testing using TorchBench showed that the new connector processed 52,000 images per second, a significant improvement over using AWS S3 connector's 30,000 images per second.
The connector leverages RDMA to enable direct memory-to-memory data transfers between Cloudian storage systems and GPU-accelerated AI frameworks, benefiting PyTorch users working with Nvidia accelerated computing.