Recommendations on NVIDIA cards for CUDA GPU computing

조회 수: 9 (최근 30일)
Mike
Mike 2015년 2월 20일
편집: Matt J 2015년 3월 13일
I use a Windows 7 PC with 8 i7 cores and 16GB RAM to run MATLAB and other simulation codes (i.e. FDTD software). I want to investigate how much of a performance boost we may get using GPU computing. I was wondering if any one had a recommendation for a configuration that might be good to start with. Nvidia recommends Tesla and Quadro GPUs for Matlab. However the price range between the entry level and the high performance parts is very large. We definitely don't want to build a high performance system yet because of the expense but also don't want to buy the cheapest since I have been told that it won't really increase performance over just using the CPU. So I would like to find the best bang for the buck. Can anyone make a recommendation?Thanks.

답변 (2개)

Erik S.
Erik S. 2015년 2월 20일
I agree with the previous answer, it depends on you application of course, but in general Nvidia Geforce cards offers alot of performance for their price.
  댓글 수: 1
Mike
Mike 2015년 2월 20일
Thanks. I started looking at the GeForce cards first but when I asked Nvidia they stated that Tesla and Quadro were recommended for MATLAB. I'm not quite sure how their product families are differentiated.

댓글을 달려면 로그인하십시오.


Matt J
Matt J 2015년 2월 20일
You would need to consider how much RAM your computation consumes. You would want to have a graphics card with enough memory such that you don't have to be constantly shuttling data back and forth to the card.
Beyond that, there is no easy way to predict how well a particular card will do relative to the CPU because that depends very specifically on the structure of your computation, and how you intend to parallelize it. For the most part, though, high end cards like the Tesla are a lot more than what you need.
  댓글 수: 3
Kofi
Kofi 2015년 3월 13일
I have the same concerns. Trying to decide on the specifications for a new machine and wondering what benefits a GPU card will bring to my computations... mostly communications systems error rate Monte Carlo simulations that use few functions from the Communications Systems Toolbox and should parallelize well. I understand I'll have to get the Parallel Computing Toolbox to take advantage of the GPU.
Thanks
Matt J
Matt J 2015년 3월 13일
편집: Matt J 2015년 3월 13일
@Kofi
Here is a list of operations that are GPU-parallelizeable in MATLAB,

댓글을 달려면 로그인하십시오.

카테고리

Help CenterFile Exchange에서 Get Started with GPU Coder에 대해 자세히 알아보기

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by