Picked up an RTX 2080, though im getting CUDNN errors


(Aaron McLeod) #1

Hi there, my home setup for doing fast.ai was working okay with a GTX970. However when i installed an RTX 2080, things have not gone well. I updated the video card drivers, and have tried running cuda 10, which pytorch 0.3 doesnt seem to support very well.

I managed to run lesson 1 okay, by keeping the batch size around 160, otherwise i’d get CUDNN_STATUS_EXECUTION_FAILED.

I tried removing cuda & cudnn, re-running through the bash script on setup. So i installed cuda-9.2. But when going through the lesson4 imdb notebook, i hit an error when running learner.fit(3e-3, 4, wds=1e-6, cycle_len=1, cycle_mult=2).

A Jupyter widget could not be displayed because the widget state could not be found. This could happen if the kernel storing the widget is no longer available, or if the widget state was not saved in the notebook. You may be able to create the widget by running the appropriate cells.

  0%|          | 0/5834 [00:00<?, ?it/s]

---------------------------------------------------------------------------
CuDNNError                                Traceback (most recent call last)
<ipython-input-9-357a8890c905> in <module>
----> 1 learner.fit(3e-3, 4, wds=1e-6, cycle_len=1, cycle_mult=2)

~/src/fastai/courses/dl1/AaronsWorkbook/fastai/learner.py in fit(self, lrs, n_cycle, wds, **kwargs)
    300         self.sched = None
    301         layer_opt = self.get_layer_opt(lrs, wds)
--> 302         return self.fit_gen(self.model, self.data, layer_opt, n_cycle, **kwargs)
    303 
    304     def warm_up(self, lr, wds=None):

~/src/fastai/courses/dl1/AaronsWorkbook/fastai/learner.py in fit_gen(self, model, data, layer_opt, n_cycle, cycle_len, cycle_mult, cycle_save_name, best_save_name, use_clr, use_clr_beta, metrics, callbacks, use_wd_sched, norm_wds, wds_sched_mult, use_swa, swa_start, swa_eval_freq, **kwargs)
    247             metrics=metrics, callbacks=callbacks, reg_fn=self.reg_fn, clip=self.clip, fp16=self.fp16,
    248             swa_model=self.swa_model if use_swa else None, swa_start=swa_start,
--> 249             swa_eval_freq=swa_eval_freq, **kwargs)
    250 
    251     def get_layer_groups(self): return self.models.get_layer_groups()

~/src/fastai/courses/dl1/AaronsWorkbook/fastai/model.py in fit(model, data, n_epochs, opt, crit, metrics, callbacks, stepper, swa_model, swa_start, swa_eval_freq, visualize, **kwargs)
    139             batch_num += 1
    140             for cb in callbacks: cb.on_batch_begin()
--> 141             loss = model_stepper.step(V(x),V(y), epoch)
    142             avg_loss = avg_loss * avg_mom + loss * (1-avg_mom)
    143             debias_loss = avg_loss / (1 - avg_mom**batch_num)

~/src/fastai/courses/dl1/AaronsWorkbook/fastai/model.py in step(self, xs, y, epoch)
     48     def step(self, xs, y, epoch):
     49         xtra = []
---> 50         output = self.m(*xs)
     51         if isinstance(output,tuple): output,*xtra = output
     52         if self.fp16: self.m.zero_grad()

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    355             result = self._slow_forward(*input, **kwargs)
    356         else:
--> 357             result = self.forward(*input, **kwargs)
    358         for hook in self._forward_hooks.values():
    359             hook_result = hook(self, input, result)

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/nn/modules/container.py in forward(self, input)
     65     def forward(self, input):
     66         for module in self._modules.values():
---> 67             input = module(input)
     68         return input
     69 

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    355             result = self._slow_forward(*input, **kwargs)
    356         else:
--> 357             result = self.forward(*input, **kwargs)
    358         for hook in self._forward_hooks.values():
    359             hook_result = hook(self, input, result)

~/src/fastai/courses/dl1/AaronsWorkbook/fastai/lm_rnn.py in forward(self, input)
    104                 with warnings.catch_warnings():
    105                     warnings.simplefilter("ignore")
--> 106                     raw_output, new_h = rnn(raw_output, self.hidden[l])
    107                 new_hidden.append(new_h)
    108                 raw_outputs.append(raw_output)

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    355             result = self._slow_forward(*input, **kwargs)
    356         else:
--> 357             result = self.forward(*input, **kwargs)
    358         for hook in self._forward_hooks.values():
    359             hook_result = hook(self, input, result)

~/src/fastai/courses/dl1/AaronsWorkbook/fastai/rnn_reg.py in forward(self, *args)
    122         """
    123         self._setweights()
--> 124         return self.module.forward(*args)
    125 
    126 class EmbeddingDropout(nn.Module):

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/nn/modules/rnn.py in forward(self, input, hx)
    202             flat_weight=flat_weight
    203         )
--> 204         output, hidden = func(input, self.all_weights, hx)
    205         if is_packed:
    206             output = PackedSequence(output, batch_sizes)

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/nn/_functions/rnn.py in forward(input, *fargs, **fkwargs)
    383             return hack_onnx_rnn((input,) + fargs, output, args, kwargs)
    384         else:
--> 385             return func(input, *fargs, **fkwargs)
    386 
    387     return forward

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/autograd/function.py in _do_forward(self, *input)
    326         self._nested_input = input
    327         flat_input = tuple(_iter_variables(input))
--> 328         flat_output = super(NestedIOFunction, self)._do_forward(*flat_input)
    329         nested_output = self._nested_output
    330         nested_variables = _unflatten(flat_output, self._nested_output)

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/autograd/function.py in forward(self, *args)
    348     def forward(self, *args):
    349         nested_tensors = _map_variable_tensor(self._nested_input)
--> 350         result = self.forward_extended(*nested_tensors)
    351         del self._nested_input
    352         self._nested_output = result

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/nn/_functions/rnn.py in forward_extended(self, input, weight, hx)
    292             hy = tuple(h.new() for h in hx)
    293 
--> 294         cudnn.rnn.forward(self, input, hx, weight, output, hy)
    295 
    296         self.save_for_backward(input, hx, weight, output)

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/backends/cudnn/rnn.py in forward(fn, input, hx, weight, output, hy)
    226 
    227         # init descriptors
--> 228         fn.rnn_desc = init_rnn_descriptor(fn, handle)
    229         if is_input_packed:
    230             fn.x_descs = cudnn.descriptor_sequence(x, fn.batch_sizes)

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/backends/cudnn/rnn.py in init_rnn_descriptor(fn, handle)
     40     if (dropout_desc_name not in fn.dropout_state) or (fn.dropout_state[dropout_desc_name].get() is None):
     41         fn.dropout_state[dropout_desc_name] = Unserializable(
---> 42             cudnn.DropoutDescriptor(handle, dropout_p, fn.dropout_seed)
     43         )
     44     dropout_desc = fn.dropout_state[dropout_desc_name].get()

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/backends/cudnn/__init__.py in __init__(self, handle, dropout, seed)
    205         self.handle = handle
    206 
--> 207         self._set(dropout, seed)
    208 
    209     def set_dropout(self, dropout, seed):

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/backends/cudnn/__init__.py in _set(self, dropout, seed)
    230             ctypes.c_void_p(state_ptr),
    231             ctypes.c_size_t(state_size),
--> 232             ctypes.c_ulonglong(seed),
    233         ))
    234 

~/anaconda3/envs/fastai/lib/python3.6/site-packages/torch/backends/cudnn/__init__.py in check_error(status)
    281 def check_error(status):
    282     if status is not 0:
--> 283         raise CuDNNError(status)
    284 
    285 

CuDNNError: 8: b'CUDNN_STATUS_EXECUTION_FAILED'

Here’s my system info

aaron@aaron-MS-7751:~/src/fastai/courses/dl1$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Tue_Jun_12_23:07:04_CDT_2018
Cuda compilation tools, release 9.2, V9.2.148

Output from pytorch’s collect_env.py

Collecting environment information...
PyTorch version: 0.3.1.post2
Is debug build: No
CUDA used to build PyTorch: 9.0.176

OS: Ubuntu 16.04.5 LTS
GCC version: (Ubuntu 5.4.0-6ubuntu1~16.04.11) 5.4.0 20160609
CMake version: Could not collect

Python version: 3.6
Is CUDA available: Yes
CUDA runtime version: 9.2.148
GPU models and configuration: GPU 0: GeForce RTX 2080
Nvidia driver version: 410.79
cuDNN version: Probably one of the following:
/usr/local/cuda-9.2/targets/x86_64-linux/lib/libcudnn.so
/usr/local/cuda-9.2/targets/x86_64-linux/lib/libcudnn.so.7
/usr/local/cuda-9.2/targets/x86_64-linux/lib/libcudnn.so.7.0.5
/usr/local/cuda-9.2/targets/x86_64-linux/lib/libcudnn_static.a

Versions of relevant libraries:
[pip] Could not collect
[conda] blas                      1.0                         mkl  
[conda] cuda90                    1.0                  h6433d27_0    pytorch
[conda] mkl                       2019.1                      144  
[conda] mkl_fft                   1.0.6            py36hd81dba3_0  
[conda] mkl_random                1.0.2            py36hd81dba3_0  
[conda] pytorch                   0.3.1           py36_cuda9.0.176_cudnn7.0.5_2  [cuda90]  pytorch
[conda] torchtext                 0.2.3                     <pip>
[conda] torchvision               0.2.0                     <pip>

(Matthew Teschke) #2

Not sure if this will solve the problem, but I think you need to be using PyTorch 1.0 with fastai 1.0


(Aaron McLeod) #3

I can give that a try. I’m running through v2 of the course, so wasnt sure if i should update or not, for the fact that the lessons could be out of date.


(RobG) #4

I recommend separate conda env’s with 0.7/0.4 and 1.0/1.0 so you can continue to use old and new frameworks.


(LinhNg) #5

Same for me for fastai 0.4 and doing course work. Will update for you when I try fastai 1.0. It is quite annoying for the error when calling learn.fit, but in my case, I only need to execute the cell again and it works.


(Aaron McLeod) #6

@digitalspecialists is it possible to have conda envs for different CUDA versions? I just have cuda installed at global level under /usr/local right now.


(Sam Ariabod) #7

CUDA is compiled into pytorch. Unless you are compiling your own, the appropriate CUDA versions will already be baked in. @digitalspecialists is correct, you just should set up 2 different conda environments and each Pytorch will use its own version of CUDA as needed. You just have to make sure the NVIDIA driver is installed and working (which it looks like it is).

So for Pytorch .4, using CUDA 9.2 it would be:
conda install pytorch=0.4.1 cuda92 -c pytorch

Then Pytorch 1 using CUDA 10 would be:
conda install pytorch torchvision cuda100 -c pytorch


(Aaron McLeod) #8

Ah cheers Sam, i’ll give this a try tonight :slight_smile:


(Aaron McLeod) #9

Trying to set this up. I did a git pull. Running into some issues with python libraries not being found, but it looks like dl1 is not meant to work with 1.0 yet? https://github.com/fastai/fastai/blob/master/courses/dl1/00-DO-NOT-USE-WITH-FASTAI-1.0.x.txt Wondering if this is something I can even use at the moment :frowning:


(Sam Ariabod) #10

That is correct. The current courses (v2) are designed for fast.ai .7. The new 1.0 library will not work. To complete the v2 courses as is you will need to set up an environment with fast.ai .7 and pytorch .4. The alternative is to wait for the v3 courses that use the latest 1.x libraries for both.