GMAP / NPB-GPU

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

Verification Unsuccessful #12

Open 2Ho0 opened 2 months ago

2Ho0 commented 2 months ago

image

image

GPU: Jetson Nano CUDA Version: 10.2.3 Compute Capability: 5.3

For BT, EP, SP, there are Verification Unsuccessful But fortunately the other benchmarks are successful

I don't know how can i fix it

gabriellaraujocoding commented 2 months ago

Hello,

GPUs like the Jetson Nano have some differences compared to desktop GPUs, which may require modifications to the source code to work correctly.

Did the benchmarks work with classes S and W? If they did, the issue might be related to the GPU's resource limitations.

One possible solution is to try using fewer resources. For example, you can force the GPU to use fewer registers, as we did in the BT Makefile.

Unfortunately, I don't have direct experience or access to the Jetson Nano, so it’s challenging for me to provide specific advice on this issue. However, I am aware that some researchers have successfully run these benchmarks on the Jetson Nano. For example, the paper "A Taste of Scientific Computing on the GPU-Accelerated Edge Device" by Pilsung Kang and Sungmin Lim discusses such work.

I suggest reaching out to people who have conducted experiments with NPB-GPU on architectures similar to the Jetson Nano.

dalvangriebler commented 1 month ago

@2Ho0 is the issue solved?

2Ho0 commented 1 month ago

Sorry for late reply. I ran with Class S and W, but It didn't work. There is 212GB available memory though