Loong
Loong
I try to reinstall all the Ubuntu system and amdgpu driver and ROCm. It still does NOT work. ``` $ sudo amd-smi set -f 100% -g 0 ******WARNING****** Operating your...
> @Looong01 I'm not seeing this issue on the version released with ROCm version 6.1.0, can you update to the latest ROCm and see if you can reproduce? Also can...
去最新版寻空里的主页，点击复制cookie 
> python -c "import tensorflow as tf;print(tf.config.list_physical_devices('GPU'))" ``` (TensorFlow) guest@home:~$ export AMD_LOG_LEVEL=7 (TensorFlow) guest@home:~$ python -c "import tensorflow as tf;print(tf.config.list_physical_devices('GPU'))" 2024-05-01 17:11:49.651768: E external/local_xla/xla/stream_executor/plugin_registry.cc:93] Invalid plugin kind specified: DNN 2024-05-01...
> Can you please set the following env var and attach the logs? **export AMD_LOG_LEVEL=7.** > > _2024-04-29 21:45:04.020735: E external/local_xla/xla/stream_executor/rocm/rocm_driver.cc:261] failed call to hipInit: HIP_ERROR_InvalidDevice_ Based on above statement,...
> Can you share the output with `HIPCC_VERBOSE=1` > > So something like `HIPCC_VERBOSE=1 hipcc -v` (base) loong@home:~$ HIPCC_VERBOSE=1 hipcc -v hipcc-cmd: /opt/rocm-5.4.3/llvm/bin/clang++ -L"/opt/rocm-5.4.3/lib" -O3 -lgcc_s -lgcc -lpthread -lm -lrt...
> What do you see for the below command? /opt/rocm/bin/rocminfo > > Also for "uname -a". ``` $ /opt/rocm-6.1.0/bin/rocminfo ROCk module version 6.7.0 is loaded ===================== HSA System Attributes =====================...
> What do you see for the below command? /opt/rocm/bin/rocminfo > > Also for "uname -a". ``` $ uname -a Linux home 5.15.0-105-generic #115-Ubuntu SMP Mon Apr 15 09:52:04 UTC...
~$ hipcc -v AMD clang version 17.0.0 (https://github.com/RadeonOpenCompute/llvm-project roc-6.0.2 24012 af27734ed982b52a9f1be0f035ac91726fc697e4) Target: x86_64-unknown-linux-gnu Thread model: posix InstalledDir: /opt/rocm-6.0.2/llvm/bin Configuration file: /opt/rocm-6.0.2/lib/llvm/bin/clang++.cfg Found candidate GCC installation: /usr/lib/gcc/x86_64-linux-gnu/11 Found candidate GCC installation:...
> What do you see for the below command? /opt/rocm/bin/rocminfo > > Also for "uname -a". ``` $ dkms status amdgpu/6.7.0-1756574.22.04, 5.15.0-105-generic, x86_64: installed ```