From NetSysLab

Jump to: navigation, search

General-purpose computing on graphics processing units (GPU) has recently gained considerable attention in various domains such as bioinformatics, scientific computing, and climate modeling. In this model GPUs are used as co-processors/accelerators to offload computationally-intensive tasks from the CPU.

This project starts from the observation that a number of GPU features (such as overlapping communication and computation, short-lived buffer reuse, and harnessing multi-GPU systems) can be abstracted and reused across different GPU applications.

Additionally, we aim expose all computational resources available on a platform (traditional cores, GPUs, other accelerators) though a unified task-like interface.

We develop CrystalGPU, a modular framework that transparently enables applications to exploit a number of GPU optimizations. Our evaluation shows that CrystalGPU enables up to 16x speedup gains on synthetic benchmarks, while introducing negligible latency overhead.


CrystalGPU V0.1


[2] A GPU Accelerated Storage System, Abdullah Gharaibeh, Samer Al-Kiswany, Sathish Gopalakrishnan, Matei Ripeanu, IEEE/ACM International Symposium on High Performance Distributed Computing (HPDC 2010), Chicago, IL, June 2010. (acceptance rate 25%) pdf
[1] CrystalGPU: Transparent and Efficient Utilization of GPU Power, Abdullah Gharaibeh, Samer Al-Kiswany, Matei Ripeanu, Technical report, Networked Systems Lab, University of British Columbia, NetSysLab-TR-2010-01. pdf