[parsec-users] Porting PARSEC to GPUs (CUDA)
msinclair at wisc.edu
Fri Feb 11 20:18:25 EST 2011
1. John - thanks for the information! Yeah, vips didn't seem like
the best choice to me either. Are you planning on using Orc for any
of the PARSEC benchmarks, or just things like convolution separate
from the PARSEC benchmarks?
2. Raghu - Yeah, I'm aware of the Rodinia suite. They do use both of
those, as you mention. I should have been more specific about what
had already been done. I've also done those benchmarks + fluidanimate
+ swaptions in CUDA. Additionally, as you mention, the Nvidia CUDA
SDK has blackscholes, but it's a very different blackscholes.
2011/2/11 Raghu <web.raghu.p at gmail.com>:
> Hi Matt,
> You can probably look at the Rodinia benchmark suite.
> They have accelerated PARSEC's Streamcluster using CUDA. I had worked on
> PARSEC's Streamcluster and Blackscholes myself about a year back as a small
> project, and you can achieve appreciable speedups for these (or any of the
> data-parallel benchmarks atleast) for large data sets. There is a CUDA
> implementation of Blackscholes in the Nvidia CUDA SDK - although this has
> got nothing to do with PARSEC.
> Hope this helps.
> On Tue, Feb 8, 2011 at 11:48 AM, Matt Sinclair <msinclair at wisc.edu> wrote:
>> Hi everyone,
>> I've been doing some work on porting the PARSEC benchmarks to GPUs,
>> specifically CUDA, and I was wondering if anyone else is working on
>> porting PARSEC to GPUs? I haven't seen any traffic on this on the
>> mailing list, but I thought it was worth asking? The reason I ask is
>> that I have done some work porting some of the PARSEC benchmarks to
>> CUDA myself, and I'm interested to see if anyone else has done some
>> work with different benchmarks that I could leverage?
>> Matt Sinclair
>> University of Wisconsin-Madison
>> College of Engineering, ECE Dept.
>> Vertical Research Group
>> parsec-users mailing list
>> parsec-users at lists.cs.princeton.edu
> parsec-users mailing list
> parsec-users at lists.cs.princeton.edu
More information about the parsec-users