VectorVisor - A Binary Translation Scheme for Throughput-Oriented GPU Acceleration

Explore a groundbreaking approach to GPU acceleration in this 25-minute conference talk from USENIX ATC '23. Delve into VectorVisor, a novel binary translation scheme that bridges the gap between low-level and high-level GPU programming models. Learn how this innovative system enables widespread GPU acceleration for server-side applications by mimicking CPU thread abstractions on GPUs. Discover the technical challenges overcome, including cross-platform system call support and efficient use of GPU resources. Examine the impressive performance gains achieved, with up to 2.9x improvement in throughput-per-dollar compared to Intel x86-64 VMs in cloud environments. Gain insights into the potential impact of VectorVisor on compute-bound workloads and its ability to match native CUDA baselines in certain scenarios.