http://www.k-wave.org/forum/forum/gpu-binaries Support for the k-Wave MATLAB toolbox en-US Fri, 06 Mar 2026 02:03:14 +0000 http://bbpress.org/?v=1.0.2 q http://www.k-wave.org/forum/search.php http://www.k-wave.org/forum/topic/axysimmetric-simulations-with-kspacefirstorder-cuda#post-9246 Mon, 09 Feb 2026 12:31:57 +0000 GioF 9246@http://www.k-wave.org/forum/ <p>I'm trying to move my simulations from MATLAB to C++ exploiting the kspaceFirstOrder executable. I managed to obtain comparable results with my axysimmetric simulation in MATLAB and the one from kspaceFirstOrder-OMP. I would like now to exploit the CUDA executable to speed up the computations, but I noticed that the axysimmetric simulation is not available for this executable. As moving to 3D will increase the computational burden, I was wondering if you are planning to maka the axysimmetric simulation available soon.<br /> Thank you for your feedback! </p> http://www.k-wave.org/forum/topic/issue-with-accessing-k-wave-gpu-code-on-ubuntu-24#post-9229 Mon, 01 Sep 2025 08:41:34 +0000 nmin 9229@http://www.k-wave.org/forum/ <p>Hi ujjal, I am experiencing the same issue. Could you kindly share how you resolved it? Thanks a lot! </p> http://www.k-wave.org/forum/topic/issue-with-accessing-k-wave-gpu-code-on-ubuntu-24#post-9207 Thu, 10 Apr 2025 11:54:55 +0000 gpr 9207@http://www.k-wave.org/forum/ <p>Hi Uijal,<br /> This makes a couple of times I hear about this issue. How did you resolve it?<br /> Bests </p> http://www.k-wave.org/forum/topic/issue-with-accessing-k-wave-gpu-code-on-ubuntu-24#post-9205 Thu, 10 Apr 2025 03:32:11 +0000 ujjal 9205@http://www.k-wave.org/forum/ <p>Hi, @everyone!<br /> Resolved, Thanks </p> http://www.k-wave.org/forum/topic/issue-with-accessing-k-wave-gpu-code-on-ubuntu-24#post-9204 Sat, 05 Apr 2025 05:40:53 +0000 ujjal 9204@http://www.k-wave.org/forum/ <p>Hi everyone,</p> <p>We are trying to access MATLAB (k-Wave) GPU code on Ubuntu 24. Our system configuration is as follows: CUDA Device: NVIDIA RTX A4000</p> <p>CUDA Toolkit Version: 12.2.0</p> <p>NVIDIA-SMI: 535.230.02</p> <p>Driver Version: 535.230.02</p> <p>CUDA Version: 12.2</p> <p>MATLAB Version: R2024b</p> <p>k-Wave Toolbox Version: 1.4</p> <p>k-Wave Toolbox Version (CUDA): 1.3 (CPP Linux Binaries)</p> <p>When we run the following command in MATLAB: sensor_data = kspaceFirstOrder2DG(kgrid, medium, source, sensor, input_args{:}, 'RecordMovie', false, 'DataCast', 'single');</p> <p>We encounter the following error message:<br /> ┌───────────────────────────────────────────────────────────────┐<br /> │ kspaceFirstOrder-CUDA v1.3 │<br /> ├───────────────────────────────────────────────────────────────┤<br /> │ Reading simulation configuration: Done │<br /> │ Selected GPU device id: Failed │<br /> └───────────────────────────────────────────────────────────────┘<br /> ┌───────────────────────────────────────────────────────────────┐<br /> │ !!! K-Wave experienced a fatal error !!! │<br /> ├───────────────────────────────────────────────────────────────┤<br /> │ Error: All CUDA-capable devices are busy or unavailable. │<br /> ├───────────────────────────────────────────────────────────────┤<br /> │ Execution terminated │<br /> └───────────────────────────────────────────────────────────────┘<br /> Error using h5readc<br /> Unable to open '/tmp/kwave_output_data05-Apr-2025-16-29-44.h5'. File or folder not found.</p> <p>Error in h5read (line 95)<br /> [data,var_class] = h5readc(Filename,Dataset,start,count,stride);<br /> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br /> Error in kspaceFirstOrder3DC (line 569)<br /> Nx = h5read(output_filename, '/Nx');<br /> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br /> Error in kspaceFirstOrder2DG (line 76)<br /> sensor_data = kspaceFirstOrder3DC(varargin{:});<br /> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br /> Error in untitled (line 51)<br /> sensor_data = kspaceFirstOrder2DG(kgrid, medium, source, sensor, input_args{:}, 'RecordMovie', false, 'DataCast', 'single');<br /> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^<br /> Is it not possible to run the k-Wave binaries (related to the Linux version) with CUDA support? If so, how can we resolve this issue?</p> <p>Thank you for your help! </p> http://www.k-wave.org/forum/topic/discrepancy-between-kspacefirstorder3dg-and-kspacefirstorder3d#post-9173 Wed, 15 Jan 2025 12:07:12 +0000 Jiri Jaros 9173@http://www.k-wave.org/forum/ <p>Hi Rehman,<br /> Would it be possible to share your simulation setup (input file) with me to investigate the problem?</p> <p>Jiri </p> http://www.k-wave.org/forum/topic/cuda-simulation-on-ada-architecture-all-cuda-capable-devices-are-busy#post-9169 Wed, 08 Jan 2025 15:04:32 +0000 gcr 9169@http://www.k-wave.org/forum/ <p>I hope this message finds you well. I am currently attempting to run simulations in CUDA on my RTX 4090, but I keep encountering the error: “All CUDA-capable devices are busy or unavailable.” Would it be possible for you to provide the re-compiled binaries to help resolve this issue?</p> <p>Thank you very much for your time and assistance. </p> http://www.k-wave.org/forum/topic/discrepancy-between-kspacefirstorder3dg-and-kspacefirstorder3d#post-9147 Thu, 03 Oct 2024 19:43:11 +0000 rali2 9147@http://www.k-wave.org/forum/ <p>Hello All,</p> <p>Long time k-Wave user here. In the past, I've mostly used the 2D simulator, but I recently started using the 3D codes to simulate data for 3D ultrasound tomography.</p> <p>I noticed that when I ran the same simulation in kspaceFirstOrder3D vs kspaceFirstOrder3DG, I would get different results. The difference between the results is not simply roundoff error. There are significant delays between the signals simulated by kspaceFirstOrder3D vs kspaceFirstOrder3DG. I suspect the discrepancy might have to do with dispersion. Are there any known (or unknown?) dispersion-related issues with kspaceFirstOrder3DG? </p> <p>Please let me know if you would like me to share some sample code with the example that produces the discrepancy.</p> <p>Thank you,<br /> Rehman Ali </p> http://www.k-wave.org/forum/topic/error-kwave-cuda-3dg-p_source_input-has-wrong-dimension-size-iostream-error#post-9104 Fri, 21 Jun 2024 15:01:26 +0000 chatillon 9104@http://www.k-wave.org/forum/ <p>Hello everyone,<br /> I encounter a problem that seems similar: I define a non-symmetrical 3D source and I do the same calculation in CPU (kspaceFirstOrder3D) and GPU (kspaceFirstOrder3DC) mode. The first gives satisfactory results while for the second I have the impression that there is an inversion of the X and Y axes of the source grid and/or of the sensor grid. Could you confirm the problem for me?<br /> thanks in advance </p> http://www.k-wave.org/forum/topic/error-kwave-cuda-3dg-p_source_input-has-wrong-dimension-size-iostream-error#post-9103 Fri, 14 Jun 2024 18:33:56 +0000 Drainville1 9103@http://www.k-wave.org/forum/ <p>Any update on the labelled source mask? Is there any way to reduce the size/dimension of source.p when using a large number of source points, that could otherwise be represented by a single signal for each unique element? </p> http://www.k-wave.org/forum/topic/c-fft-plans-creation-really-slow#post-9092 Thu, 09 May 2024 15:08:22 +0000 zak morgan 9092@http://www.k-wave.org/forum/ <p>Never mind, I was ecperimeting with PML and had it set to be outside the grid, thus growing the grid to a non-power of 2 size thus explaining the slow-down! </p> http://www.k-wave.org/forum/topic/c-fft-plans-creation-really-slow#post-9091 Wed, 08 May 2024 23:09:21 +0000 zak morgan 9091@http://www.k-wave.org/forum/ <p>I'm also seeing similar, when running on the CPU c++ code pre-pocessing is instant and then my simulation takes about 47 seconds, however when running the CUDA code, the simulation takes about 10 seconds, but 80 seconds is spent on pre-processing the FFT. It seems odd that preprocessing should be this much slower on the GPU than on the CPU? </p> http://www.k-wave.org/forum/topic/kspacefirstorder-cuda-v13-claims-it-needs-cuda-120#post-9039 Sat, 17 Feb 2024 19:27:55 +0000 pamparana 9039@http://www.k-wave.org/forum/ <p>Did you find a solution to this?</p> <p>For me, the situation is reversed. it runs fine on my local linux machine with cuda 11.7 but the docker container using the cuda 11.6 image does not work and gives an error where it says cuda 0.0 is installed. </p> http://www.k-wave.org/forum/topic/memory-requirements-for-gpu-version#post-9028 Tue, 30 Jan 2024 17:59:48 +0000 daga_pankaj 9028@http://www.k-wave.org/forum/ <p>Never mind. Should read the documentation more.<br /> For anyone else, interested, it is at the end of the user manual. </p> http://www.k-wave.org/forum/topic/memory-requirements-for-gpu-version#post-9026 Tue, 30 Jan 2024 13:55:19 +0000 daga_pankaj 9026@http://www.k-wave.org/forum/ <p>I am going to run simulations on a 256 x 256 x 128 grid. Could someone suggest what memory requirements I could expect for the GPU version? </p> http://www.k-wave.org/forum/topic/kspacefirstorder-cuda-v13-claims-it-needs-cuda-120#post-8947 Mon, 20 Nov 2023 10:46:00 +0000 BIlly 8947@http://www.k-wave.org/forum/ <p>The CUDA binary kspaceFirstOrder-CUDA v1.3 claims to require CUDA 12.0</p> <p><code></code>`<br /> ┌───────────────────────────────────────────────────────────────┐<br /> │ !!! K-Wave experienced a fatal error !!! │<br /> ├───────────────────────────────────────────────────────────────┤<br /> │ Error: Insufficient CUDA driver version. The code needs CUDA │<br /> │ version 12.0 but 11.7 is installed. │<br /> ├───────────────────────────────────────────────────────────────┤<br /> │ Execution terminated │<br /> └───────────────────────────────────────────────────────────────┘<br /> <code></code>`<br /> Although I see nothing to suggest that CUDA 12.0 or above is needed in the documentation or on this site and my department's HPC only runs up to CUDA 11.7.<br /> Stranger still the binary runs completely fine in a CUDA 11.7 Docker container on my workstation.<br /> Is this a bug with k-wave and the HPC? </p> http://www.k-wave.org/forum/topic/large-scale-simulation-on-8-a100-gpus#post-8928 Fri, 06 Oct 2023 10:51:32 +0000 Pavel 8928@http://www.k-wave.org/forum/ <p>I am not a computer guy, but it looks like GPU apps use GPU memory plus half of CPU RAM memory, and it looks like default configuration of my system. For example, having threadripper architecture of motherboard I sum up half of my 256 GB RAM with 40 GB of my 4090 by rather fast (ddr5) and cost effective way (cheaper than buying four 3090 units in parallel). Maybe 3DG version of kWave allows for such CPUtoGPU memory sharing as well? </p> http://www.k-wave.org/forum/topic/large-scale-simulation-on-8-a100-gpus#post-8916 Sat, 23 Sep 2023 13:09:52 +0000 so_dence 8916@http://www.k-wave.org/forum/ <p>@Jiri Jaros<br /> Hi Jiri Jaros</p> <p> First of all,Thanks for your reply!And I am deeply interested in the multi-GPU version,could you please send me one beta version!I am deeply grateful with that! </p> http://www.k-wave.org/forum/topic/large-scale-simulation-on-8-a100-gpus#post-8910 Thu, 14 Sep 2023 16:03:30 +0000 jamesjc 8910@http://www.k-wave.org/forum/ <p>@Jiri Jaros,</p> <p>Are you and your team still on track to release a multi-GPU version this month? Do you have an updated timeline?</p> <p>We're very keen to use it :) </p> http://www.k-wave.org/forum/topic/cuda-simulation-on-ada-architecture-all-cuda-capable-devices-are-busy#post-8888 Wed, 02 Aug 2023 15:06:38 +0000 f841r 8888@http://www.k-wave.org/forum/ <p>It was a problem with my cuda installation. I now installed the recommended drivers over apt and cuda 11.8 using the run file with the --toolkit option. Now everything works with the self-compiled binaries. </p>