GPUassert: invalid device symbol

[nvlabs]

GPUassert: invalid device symbol

[KyunghyunLee]

I figured out the issue.
When I build torchcule, I got an error at line 11 of setup.py.(https://github.com/NVlabs/cule/blob/master/setup.py#L11)
I modified it to "codes = ['70']", similar to line 14.
torchcule was built successfully, but I got the error message above.

I dig into the table.hpp and find that the code means the architecture of GPU.
I found '70' actually means 'sm_70', and it is for Tesla V100.
Other codes are listed in below link. For 1080TI, it is '61'
http://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/

I cleaned 'build' and 'dist' folder, then rebuild torchcule.
It works great now.

[ifrosio]

How many GPUs do you have on your machine and which one is GPU-0 (the one you are using since you are passing ('cuda', 0) as device)?

[KyunghyunLee]

I got the exact same error

python ./examples/ppo/ppo_main.py --use-cuda-env --use-openai-test-env  --gpu 0
{'ale_start_steps': 400,
 'alpha': 0.99,
 'batch_size': 256,
 'clip_epsilon': 0.1,
 'conf_file': None,
 'entropy_coef': 0.01,
 'env_name': 'PongNoFrameskip-v4',
 'episodic_life': False,
 'eps': 1e-05,
 'evaluation_episodes': 10,
 'evaluation_interval': 1000000,
 'gamma': 0.99,
 'gpu': 0,
 'local_rank': 0,
 'log_dir': 'runs',
 'loss_scale': None,
 'lr': 0.00065,
 'lr_scale': False,
 'max_episode_length': 18000,
 'max_grad_norm': 0.5,
 'multiprocessing_distributed': False,
 'no_cuda_train': False,
 'normalize': False,
 'num_ales': 16,
 'num_gpus_per_node': -1,
 'num_stack': 4,
 'num_steps': 5,
 'opt_level': 'O0',
 'output_filename': None,
 'plot': False,
 'ppo_epoch': 3,
 'profile': False,
 'save_interval': 0,
 'seed': 1565661279,
 't_max': 50000000,
 'tau': 1.0,
 'use_adam': False,
 'use_cuda_env': True,
 'use_gae': False,
 'use_openai': False,
 'use_openai_test_env': True,
 'value_loss_coef': 0.5,
 'verbose': False}

PyTorch  : 1.1.0
CUDA     : 10.0.130
CUDNN    : 7501
APEX     : 0.1.0

GPUassert: invalid device symbol /home/lkh/Codes/cule/cule/atari/cuda/tables.hpp 43

here is my nvidia-smi

Tue Aug 13 11:06:38 2019       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.48                 Driver Version: 410.48                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 108...  Off  | 00000000:03:00.0 Off |                  N/A |
| 33%   57C    P0    65W / 250W |     12MiB / 11178MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX 108...  Off  | 00000000:04:00.0  On |                  N/A |
|  0%   48C    P8    16W / 250W |    501MiB / 11177MiB |      8%      Default |
+-------------------------------+----------------------+----------------------+

[Shmuma]

I have the same gpu configuration: two 1080TI. вт, 13 авг. 2019 г. в 5:07, KyunghyunLee <[email protected]>:

I got the exact same error python ./examples/ppo/ppo_main.py --use-cuda-env --use-openai-test-env --gpu 0 {'ale_start_steps': 400, 'alpha': 0.99, 'batch_size': 256, 'clip_epsilon': 0.1, 'conf_file': None, 'entropy_coef': 0.01, 'env_name': 'PongNoFrameskip-v4', 'episodic_life': False, 'eps': 1e-05, 'evaluation_episodes': 10, 'evaluation_interval': 1000000, 'gamma': 0.99, 'gpu': 0, 'local_rank': 0, 'log_dir': 'runs', 'loss_scale': None, 'lr': 0.00065, 'lr_scale': False, 'max_episode_length': 18000, 'max_grad_norm': 0.5, 'multiprocessing_distributed': False, 'no_cuda_train': False, 'normalize': False, 'num_ales': 16, 'num_gpus_per_node': -1, 'num_stack': 4, 'num_steps': 5, 'opt_level': 'O0', 'output_filename': None, 'plot': False, 'ppo_epoch': 3, 'profile': False, 'save_interval': 0, 'seed': 1565661279, 't_max': 50000000, 'tau': 1.0, 'use_adam': False, 'use_cuda_env': True, 'use_gae': False, 'use_openai': False, 'use_openai_test_env': True, 'value_loss_coef': 0.5, 'verbose': False} PyTorch : 1.1.0 CUDA : 10.0.130 CUDNN : 7501 APEX : 0.1.0 GPUassert: invalid device symbol /home/lkh/Codes/cule/cule/atari/cuda/tables.hpp 43 here is my nvidia-smi Tue Aug 13 11:06:38 2019 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 410.48 Driver Version: 410.48 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX 108... Off | 00000000:03:00.0 Off | N/A | | 33% 57C P0 65W / 250W | 12MiB / 11178MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 1 GeForce GTX 108... Off | 00000000:04:00.0 On | N/A | | 0% 48C P8 16W / 250W | 501MiB / 11177MiB | 8% Default | +-------------------------------+----------------------+----------------------+ — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#5?email_source=notifications&email_token=AAAQE2SRMTQ3MWQQ43P3PA3QEIJMXA5CNFSM4II7WRDKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD4EKLII#issuecomment-520660385>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAQE2RG7OYS4QR3BVYN4NLQEIJMXANCNFSM4II7WRDA> .

-- wbr, Max Lapan

[ifrosio]

Thanks - we are modifying the code to support multiple architectures, although this may require a larger compilation time. Will close when done.