Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore the revolutionary shift toward GPU-initiated storage access in this 22-minute conference talk from the Open Compute Project. Learn how data-intensive applications are breaking free from memory constraints by leveraging storage systems that offer access to much larger datasets and improved total cost of ownership. Discover why GPUs, with their 100x more threads and superior memory bandwidth compared to CPUs, are ideally positioned to transform data processing when they can access storage directly rather than relying on CPU intermediation. Examine SCADA (Scalable Accelerated Data Access), an innovative solution that enables up to 100,000 GPU threads to concurrently initiate their own storage requests, incorporates GPU-side caching to optimize data locality, and performs secure storage access at speeds limited only by PCIe bandwidth. Gain insights into the latest technical developments, implementation challenges, and collaborative efforts with Storage-Next partners as presented by NVIDIA's Distinguished Engineer and Senior Research Scientist.
