Are my GPU results reasonable?

2 次查看（过去 30 天）
Batuhan Hangun
Batuhan Hangun 2019-8-12
编辑： Batuhan Hangun 2019-8-15
Hello there.
I was trying to run
https://www.mathworks.com/help/parallel-computing/examples/measuring-gpu-performance.html
this example to see results of a quick benchmark on CPU and GPU. After it's completed, my CPU results at Read+Write and Double Precision Matrix Multiplication seemed weird to me.
After that, I tried the benchmark provided by MathWorks Parallel Computing Toolbox Team, which was given at following link.
https://www.mathworks.com/matlabcentral/fileexchange/34080-gpubench
Results from that benchmark is below.
Is everything alright with my CPU and GPU? It seems that my GPU is slower than CPU at double precision operations. Since my GPU(GTX 1050 Ti) is not an outdated GPU, its results at double precision seems wrong to me. Maybe I am wrong or can you provide me any better benchmark(s) to diagnose if there is a problem or not.
System specifications:
  • Operating System: Windows 10 Pro 64-bit
  • Processor: Intel(R) Core(TM) i7-7700HQ CPU @2.80GHz (8 CPUs)
  • Memory: 16384MB RAM
  • GPU: NVIDIA GeForce GTX 1050 Ti

请先登录，再回答此问题。

回答（1 个）

Jason Ross
Jason Ross 2019-8-13
编辑：Jason Ross 2019-8-13
Looking at the Wikipedia page for NVidia GeForce 10 series GPUs, the benchmark results seem to agree. The double precision GFLOPS number published there is 61.9 GFLOPS for the base clock and 66.8 GFLOPS for the boost clock. The single results are also close to your results -- 1981.4 and 2138.1 respectively -- but these are many times faster than the doubles. If you look at the other GTX cards in the table you will see that this trend continues for other cards in the GeForce 10 family -- the single performance is many multiples faster than doubles.
  1 个评论
Joss Knight
Joss Knight 2019-8-13
To elaborate on Jason's answer, this is perfectly normal. The GTX range are primarily targeted at high performance graphics and only really optimize single precision computation, which graphics uses. They are usually very slow at double precision, and are commonly out-performed by a multi-core CPU system.

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 GPU Computing in MATLAB 的更多信息

标签

产品


版本

R2018b

Community Treasure Hunt

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

Start Hunting!

Translated by