search

GMAP / NPB-GPU

NAS Parallel Benchmarks for GPU
Other
18 stars 7 forks source link

Verification Problem #8

Closed RicardoZhy1215 closed 3 months ago

RicardoZhy1215 commented 3 months ago

I download this repo and test many problems in A size. I found that all code can not pass verification.

dalvangriebler commented 3 months ago

Hi @RicardoZhy1215, could you please provide more details about the problem? We need information about your operating system, CUDA version, and nvidia driver.

RicardoZhy1215 commented 3 months ago

Hi @dalvangriebler , My OS is ubuntu, cuda version is 12.3 and nvidia driver version is Driver Version: 545.23.08, I use A100 nvidia GPU, thanks for your quick reply!

RicardoZhy1215 commented 3 months ago

I test all the problem, the final computation is wrong and can not pass the verification. image

dalvangriebler commented 3 months ago

Have you configured correctly the compute capability in the make.def?

RicardoZhy1215 commented 3 months ago

Hi, @dalvangriebler , no, I just use the default config in the make.def file.

gabriellaraujocoding commented 3 months ago

Hello, Ricardo.

You should update the file make.def by informing the compute capability of your GPU.

You can collect the compute capability of your GPU by running the following command:

nvidia-smi --query-gpu=compute_cap --format=csv,noheader,nounits

You must replace the compute capability values in the following line in the make.def file: (in this case, replace 61 by the value of your GPU):

COMPUTE_CAPABILITY = -gencode arch=compute_61,code=sm_61
RicardoZhy1215 commented 3 months ago

hello, Gabriell, many thanks for your help! The problem was solved!