k-Wave

1. xberg
   

   Member
   Joined: Jun '21
   Posts: 4

   Hi everyone,

   I'm purchasing a laptop and I would like to run 3D k-wave simulations with it. I see that the compatible GPUs are described as " Nvidia GPU with SM architecture between 3.0 and 7.5 (Fermi-Turing)". The new laptop I would like to purchase has a NVIDIA® GeForce RTX™ 3060 6GB GDDR6 [60W] which is classified as a SM 8.6 Ampere which seems to be a tier above Turing. (https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/)

   I just wanted to confirm whether this meant the 3060 is incompatible with k-wave. Also, I since I am a novice in this computing realm, if it is compatible will that mean the GPU cannot be used whatsoever to run k-wave, or just that it would not be running to the 3060's maximum potential?

   Thanks for you help!

   Posted 3 years ago #

2. xberg
   

   Member
   Joined: Jun '21
   Posts: 4

   Hi everyone,

   Perhaps I could rephrase more succinctly to say: Will GPU Architectures higher than SM_75 (Turing) such as SM_86(Ampere) work for the GPU optimized code?

   Posted 3 years ago #

3. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,643

   Hi xberg,

   You will be able to run k-Wave on newer GPU architectures. I'll leave it to Jiri to answer whether you will need to re-compile the code, or whether the existing binary on the downloads page will work out of the box.

   Brad.

   Posted 3 years ago #

4. Jiri Jaros
   

   Developer
   Joined: Feb '12
   Posts: 115

   Hi xberg,
   you'll have to update the makefile and recompile the code from sources with the newest CUDA release (11.1+).

   1. Install CUDA and HDF5 libraries
   2. Add necessary SM models to this line in the makefile
   
   # What CUDA GPU architectures to include in the binary
   CUDA_ARCH = --generate-code arch=compute_50,code=sm_50 \
   --generate-code arch=compute_52,code=sm_52 \
   --generate-code arch=compute_53,code=sm_53 \
   --generate-code arch=compute_60,code=sm_60 \
   --generate-code arch=compute_61,code=sm_61 \
   --generate-code arch=compute_62,code=sm_62 \
   --generate-code arch=compute_70,code=sm_70 \
   --generate-code arch=compute_72,code=sm_72 \
   --generate-code arch=compute_75,code=sm_75
   
   3. Rebuild the code.

   It shouldn't take more than 30 minutes. If you're not successful, let me know and I'll send you new binaries.

   Best
   Jiri

   Posted 3 years ago #

5. xberg
   

   Member
   Joined: Jun '21
   Posts: 4

   Thank you for your reply! I will be getting the computer shortly and I will attempt to follow your instructions!

   Posted 3 years ago #

6. xberg
   

   Member
   Joined: Jun '21
   Posts: 4

   Hi Jiri,

   This is my first time updating makefiles and using the CUDA package and I am having trouble modifying the code. I would greatly appreciate if you could send me the new binaries. Thank you for you help!

   Best,

   Xander

   Posted 3 years ago #

7. Jiri Jaros
   

   Developer
   Joined: Feb '12
   Posts: 115

   Hi Xander
   I will release a new binary within a month and also add some new functionality.

   In order to support Ampere GPUs, it is necessary to install CUDA 11.4. and update the makefile by

   # What CUDA GPU architectures to include in the binary
   CUDA_ARCH = --generate-code arch=compute_50,code=sm_50 \
   --generate-code arch=compute_52,code=sm_52 \
   --generate-code arch=compute_53,code=sm_53 \
   --generate-code arch=compute_60,code=sm_60 \
   --generate-code arch=compute_61,code=sm_61 \
   --generate-code arch=compute_62,code=sm_62 \
   --generate-code arch=compute_70,code=sm_70 \
   --generate-code arch=compute_72,code=sm_72 \
   --generate-code arch=compute_75,code=sm_75 \
   --generate-code arch=compute_80,code=sm_80 \
   --generate-code arch=compute_86,code=sm_86

   Posted 3 years ago #

8. NicholasHorton
   

   Member
   Joined: Mar '16
   Posts: 2

   Hi Jiri,
   I ran the kspaceFirstOrder2DG and the code reported a problem. I think this problem has the same topic with this post.
   The computer I used is NVIDIA A100, whose GPU architechure is Ampere. When I assigned my A100 (gpu device 0) to run a simulation using kspaceFirstOrder2DG, the MATLAB reported the error message
   'cuda device id 0 is busy and unavailable'.
   I downloaded k-Wave Toolbox and compiled binary files in this page:http://www.k-wave.org/download.php. Is this error resulted by my k-wave version does not support the Ampere GPUs?
   If so, where could I download the newest k-wave which supports the Ampere GPUs?

   Posted 3 years ago #

9. paoloC
   

   Member
   Joined: Apr '20
   Posts: 1

   Hi everyone,

   I am going to buy a new desktop PC but the only configurations my university proposes are with GPUs with Ampere architecture.
   Could you confirm that all ampere GPUs are compatible (after recompiling) ? If I understood well I have to recompile updating the makefile as mentioned beforehand, is it right ?

   Dr Jaros, did you already released the binary you mentioned in the other post (to try it) ?

   Thank you very much for your help.

   Best regards,

   Paolo

   Posted 2 years ago #

10. ktritz
    

    Member
    Joined: Jul '22
    Posts: 1

    Any progress on updated binaries supporting the new Ampere architecture?

    Posted 2 years ago #

11. Ines Milheiro
    

    Member
    Joined: Sep '21
    Posts: 1

    Hello everyone,

    I am currently working with a NVIDIA GeForce RTX 3090, with Ampere architecture, which I use to run 3D k-Wave simulations.

    A collegue of mine was able to follow the steps described by Jiri Jaros (I don't really know what else was necessary to compile the codes) and obtain the new binary which enabled me to run GPU optimized code. I share the new binary in the following link: https://github.com/InesConceicao/Binary-k-Wave.git

    Hope it helps!
    Best regards.

    Posted 2 years ago #

12. Bradley Treeby
    

    Developer
    Joined: Mar '10
    Posts: 1,643

    We plan to upload a new version of the C++/CUDA binaries later this year (after supercomputing).

    Posted 2 years ago #

13. so_dence
    

    Member
    Joined: Nov '22
    Posts: 4

    hello, i was wandering when the multi-gpus can be supported,can i use multi-gpus now in the early 2023?

    Posted 1 year ago #

14. lschlieder
    

    Member
    Joined: Mar '20
    Posts: 4

    > We plan to upload a new version of the C++/CUDA binaries later this year (after supercomputing).

    Is the new version uploaded yet? I am running into the same error.

    Posted 1 year ago #

15. Jiri Jaros
    

    Developer
    Joined: Feb '12
    Posts: 115

    CUDA k-Wave is compatible with all GPUs, from Fermi to Hopper. It is only necessary to recompile the sources with a particular SM architecture.

    https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/

    I will release a new binary with the new release which was postponed by bugs in CUDA 12.x

    Posted 1 year ago #

RSS feed for this topic

Reply

You must log in to post.