search

gpgpu-sim / gpgpu-sim_distribution

GPGPU-Sim provides a detailed simulation model of contemporary NVIDIA GPUs running CUDA and/or OpenCL workloads. It includes support for features such as TensorCores and CUDA Dynamic Parallelism as well as a performance visualization tool, AerialVisoin, and an integrated energy model, GPUWattch.
Other
1.02k stars 487 forks source link

RuntimeError: cublas runtime error(gpgpu-sim with pytorch) #284

Open Wen-Tian-Pineapple opened 8 months ago

Wen-Tian-Pineapple commented 8 months ago

Hi, I'm having issue with gpgpu-sim with pytorch when testing mnist program. I use shared library when building pytorch :libcudnn.so libcublas.so And when mnist program is trying to call "nn.linear" function, it gives me this error:

RuntimeError: cublas runtime error : library not initialized at /home/user/pytorch-gpgpu-sim/aten/src/THC/THCGeneral.cpp:333

Anyone have any idea ho to bypass this issue?

Set up: GPGPUSIM version 4.2.0 GCC: 5.4 CUDA: 8.0.61 CUDNN: 7.1.4 PYTORCH_BIN: /home/wentian/.local/lib/python2.7/site-packages/torch/lib/libcaffe2_gpu.so

mattsinc commented 8 months ago

I am unfamiliar with this error, but would strongly encourage you to use Accel-Sim instead, which uses GPGPU-Sim in general, but has much better support for library code like this.