RTX 2070 GPU Not Supported

Hello, I am a complete beginner to all of this and have been working on getting Anaconda installed and running with GPU support when I stumbled across this platform and decided to give it a try. I am running Windows 10 and have a RTX 2070 GPU. My GPU driver version is 417.71. I have tried installing DLS version 2.5.0 and the log says that my GPU is not supported. I have read some of the other topics and have uninstalled CUDA 10 & 9 so there was nothing and then reinstalled DLS w/ CUDA 9.0.

Logs
" Starting Deep Learning Studio…
Checking GPU support…
GPU not supported
skipping cifar dataset download!
skipping IMDB dataset download!
skipping MNIST dataset download!
skipping reuters dataset download!
No changes detected
Using MXNet backend."

Any help would be greatly appreciated! Thank you.

Hi you are getting this error because currently DLS doesn’t support RTC cards we are planning to do so in next release.

Regards
Rajat

Can you tell me when the next release is coming out? Currently, I can’t even do the Udemy Training as my computer freezes up on the first training run. My CPU goes up to 97% and then all functions seem to freeze…mouse won’t move, computer time doesn’t change, etc. I left my computer run like this overnight and still no change so I had to hard reboot. Maybe you can provide some tips to make this run on my CPU only without freezing? I have a AMD FX-8320 8 core processor and 32 GB ram.

Thank you,
Curtis

For the next release I can’t tell you the exact time right now but if you want to train your models on GPU in this case you can try Cloud version. There you will get 2hrs of free GPU with your Id.
Other than this you can use AMI of DLS on AWS if you have free credits over there.

For CPU you can try reducing the batch size and also don’t load the whole dataset at once.
Regards
Rajat

Hi Rajat,
First of all, thank you very much for the development of deep learning studio that allows non computer scientists like me to learn deep learning.
I’m like Curtis, I bought a RTX 2080TI card (opportunity to buy it for $ 750 …). But the installation of the card and the reinstallation of DLS made me discover CUDA and its versions 9 and 10 …
Now, I just have to wait for the next version of DLS hoping that it will be compatible with this type of graphics card.
To help me to wait, do you have an idea of the time that will take the development of this new version (weeks? Months? Years? …)

Thank you again for your amazing work on DLS
Phil

One more question.
Can you give me the name of supported graphic cards ? (All the GTX ?)
Thks a lot
Phil

Good evening everyone
I think I found a totally unconventional solution that worked for me:

  1. install the latest drivers: https://www.nvidia.com/download/driverResults.aspx/138070/en-us
    and visual studio 2017 (not 2019) https://visualstudio.microsoft.com/fr/downloads/
  2. install CUDA9.2
    https://developer.nvidia.com/cuda-92-download-archive
  3. install the patch for CUDA9.2
  4. install cuDNN 7.5
    https://developer.nvidia.com/rdp/cudnn-archive
  5. copy the directory where CUDA 9.2 is located (C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ v9.2) and paste the copy wherever you want.
  6. open the “bin” directory of the v9.2 directory that you just pasted and replace “92” with “90” in the name of all the dll’s whose name contains 92
  7. Remove CUDA9.2 with the application manager
  8. install deep learning studio and check GPU support. (you will get a message saying that the GPU is not supported but it does not matter)
  9. go to the CUDA 9.0 directory that deep learning has just created.
    C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ v9.0
  10. open in another window the directory with the modified version of CUDA 9.2 (step 6, let’s call it “v9.2mod”)
  11. copy all the files of each subdirectories of “v9.2mod” in the subdirectory of the same name of the directory v9.0
  12. also copy in v9.0 the subdirectories of v9.2mod which do not exist in v9.0 (in short, transfer all the files of v9.2mod in v9.0.
    extinguish and light up the PC … launch DLS and cross your fingers
    This is a “very dirty” method that must horrify computer scientists but it worked for me (win 10 + RTX2080TI)
    Hope it will work for you, DLS + RTX 2080TI … wouha, such a powerfull team!
4 Likes

Hi Phil,
I just followed your steps and can confirm that DLS is now utilizing my RTX 2070! Thank you for the follow up post!

Curtis

Hi,

For fun, I also tried with CUDA10.1 + cudnn7.5 for CUDA10.1 … it works perfectly:

  • install CUDA10.1 and cudnn 7.5 for CUDA10.1,
  • duplicate CUDA10 / bin directory (let’s call it bin_dup),
  • rename all dll whose name ends with 10 or 101 in this bin_dup directory using the same name but ending with “90” instead of “10” or “101”,
  • paste all the dll whose name you have just modified in the directory CUDA10 / bin
  • save the directory CUDA10 containing all the dll (original and renamed)
  • uninstall CUDA10
  • install CUDA9.0
  • install DSL with GPU support
  • copy all the files and subdirectories of CUDA10 with the modified DLLs into the CUDA9 directory (replace existing files)
    turn off and on the PC
    launch DLS … wait (it’s a bit long …) it should support the GPU …:crossed_fingers:

Best regards

Phil

Hi Phil,

I have tried this setup without being able to get my GPU supported in DLS. Is there anything that you can think of that might be missing from the instructions? I have tried this a few times now. The 9.2 version worked well. Note that I have the latest GPU driver installed (430.86), which appears to be newer than what comes with CUDA 10.1.
CUDA 10.1.168_425.25 Win10
cuDNN 10.1 win x64 v.7.6.0.64

Thank you,
Curtis

Hi Curtis,

Sorry for the delay of this post … I had to deal with a tricky ImageJ macro !!!
I am sorry but I am not a computer scientist … and I do not know how and why it worked for me and not for you.
:thinking:
If it does the trick for you with cuda 9.2, keep this version … waiting for the next DSL version and hoping it will natively support RTX cards …:crossed_fingers:
best
Phil

Thank you! It works with RTX2070 Super too

Thanks Phil2toul, still an issue in Jan/2020
Followed your instructions exactly and its working

i installed Cuda 9.0, my GPU(970 gtx) is supported, but its not used for the calculations :frowning:

anyone knows a solution for rtx sli setup?
is there any beta dls version available? this really sucks that rtx is not supported yet… :frowning:

1 Like

Hi Curtis,

I am also using an RTX2070 - I followed Phil’s steps there and it got the Log to say that the GPU is supported, however, upon launching DLS and training, it does not utilize my GPU and instead defaults to the CPU - any idea how to fix this?

Thanks

Hi Curtis

Regarding the CUDA 9 installation, did you use Express installation or custom installation (and unticking the driver install to maintain a newer driver) ?

Thanks