YoVDO

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

Offered By: USENIX via YouTube

Tags

USENIX Annual Technical Conference Courses Recursion Courses System Calls Courses GPU Acceleration Courses

Course Description

Overview

Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
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.

Syllabus

USENIX ATC '23 - VectorVisor: A Binary Translation Scheme for Throughput-Oriented GPU Acceleration


Taught by

USENIX

Related Courses

Fundamentals of Accelerated Computing with CUDA C/C++
Nvidia via Independent
Using GPUs to Scale and Speed-up Deep Learning
IBM via edX
Deep Learning
IBM via edX
Deep Learning with IBM
IBM via edX
Accelerating Deep Learning with GPUs
IBM via Cognitive Class