-1

Help me please. Is there any way to start a project with 2.1 cuda capability?
Trying to launch a project https://github.com/CorentinJ/Real-Time-Voice-Cloning/ 

Found GPU0 GeForce GT 630M which is of cuda capability 2.1.
    PyTorch no longer supports this GPU because it is too old.
    The minimum cuda capability that we support is 3.5.

  warnings.warn(old_gpu_warn % (d, name, major, capability[1]))
Found 1 GPUs available. Using GPU 0 (GeForce GT 630M) of compute capability 2.1 with 2.1Gb total memory.

Preparing the encoder, the synthesizer and the vocoder...
Traceback (most recent call last):
  File "demo_cli.py", line 61, in <module>
    encoder.load_model(args.enc_model_fpath)
  File "C:\Users\kisel\Desktop\Real-Time-Voice-Cloning-master\encoder\inference.py", line 32, in load_model
    _model = SpeakerEncoder(_device, torch.device("cpu"))
  File "C:\Users\kisel\Desktop\Real-Time-Voice-Cloning-master\encoder\model.py", line 21, in __init__
    batch_first=True).to(device)
  File "C:\Users\kisel\anaconda3\envs\TestVoice\lib\site-packages\torch\nn\modules\module.py", line 386, in to
    return self._apply(convert)
  File "C:\Users\kisel\anaconda3\envs\TestVoice\lib\site-packages\torch\nn\modules\rnn.py", line 127, in _apply
    self.flatten_parameters()
  File "C:\Users\kisel\anaconda3\envs\TestVoice\lib\site-packages\torch\nn\modules\rnn.py", line 123, in flatten_parameters
    self.batch_first, bool(self.bidirectional))
RuntimeError: cuDNN error: CUDNN_STATUS_ARCH_MISMATCH
talonmies
  • 70,661
  • 34
  • 192
  • 269
Александр Киселев
  • 1
  • 1
  • 1
  • See the answer [here](https://stackoverflow.com/a/61034368/2790047) which describes the difference between CUDA and compute capability. FYI compute capability 2.1 does support CUDA 8 (but not CUDA 9 or 10). Pytorch binaries that you install with pip or conda are **not** compiled with support for compute capability 2.1. This means you need to build PyTorch 1.0 or earlier from source (since support for CUDA 8 was dropped in PyTorch 1.1). See the link for information. You can't use PyTorch 1.1 or newer at all with CUDA support since your hardware doesn't support CUDA 9 or 10. – jodag Apr 09 '20 at 08:14
  • @jodag And how can I solve the problem on Windows? – Александр Киселев Apr 09 '20 at 10:57
  • Build a version of pytorch which supports CUDA 8 from source. A link to the instructions on the official pytorch website is provided in the answer I linked earlier. Build instructions are available for Windows, Linux, and MacOS. – jodag Apr 09 '20 at 18:05

1 Answers1

0

as far as I know, NVIDIA has different technology starting CUDA 3.5. So CUDA 2.1 and CUDA 3.5 is totally different things of architecture.

CUDA 2.1 is released more than ten years ago, mainly released for Tesla architecture, but CUDA 3.5 mainly released for Kepler architecture.

Biggest difference of both feature is dynamic parallelism.

Source: https://en.wikipedia.org/wiki/CUDA

aviezab
  • 1
  • 1