# GPU

**URL:** https://discourse.julialang.org/c/domain/gpu/11.md?page=1

[Latest](https://discourse.julialang.org/latest.md) · [Categories](https://discourse.julialang.org/categories.md) · [Tags](https://discourse.julialang.org/tags.md)

**Page:** 2

---

## [Error in testset gpuarrays/linalg/core](https://discourse.julialang.org/t/error-in-testset-gpuarrays-linalg-core/134091)

<div class="topic-metadata">

**Author:** [@Roger\_Powell](https://discourse.julialang.org/u/Roger_Powell)\
**Replies:** 2\
**Last updated:** [November 25, 2025, 11:48am UTC](https://discourse.julialang.org/t/error-in-testset-gpuarrays-linalg-core/134091 "2025-11-25T11:48:08Z")

</div>

First time user of CUDA.jl. My Julia versioninfo() is Julia Version 1.12.2 Commit ca9b6662be4 (2025-11-20 16:25 UTC) Build Info: Official https://julialang.org release Platform Info: OS: Linux (x86\_64-linux-gnu) C…

---

## [Help with performance issue when upgrading CUDA](https://discourse.julialang.org/t/help-with-performance-issue-when-upgrading-cuda/134064)

<div class="topic-metadata">

**Author:** [@jamesr137](https://discourse.julialang.org/u/jamesr137)\
**Replies:** 5\
**Last updated:** [November 24, 2025, 7:15pm UTC](https://discourse.julialang.org/t/help-with-performance-issue-when-upgrading-cuda/134064 "2025-11-24T19:15:13Z")

</div>

Hi all. I have an issue that I’m hoping somebody on here can help me with. The issue is simple to state yet hard (at least for me) to diagnose, and it’s this: every time I try to upgrade to a version of CUDA past 5.5.2, …

---

## [InvalidIRError when running AcceleratedKernels.sum on a GPU SubArray (CuArray view)](https://discourse.julialang.org/t/invalidirerror-when-running-acceleratedkernels-sum-on-a-gpu-subarray-cuarray-view/133994)

<div class="topic-metadata">

**Author:** [@0samuraiE](https://discourse.julialang.org/u/0samuraiE)\
**Replies:** 2\
**Last updated:** [November 20, 2025, 9:20am UTC](https://discourse.julialang.org/t/invalidirerror-when-running-acceleratedkernels-sum-on-a-gpu-subarray-cuarray-view/133994 "2025-11-20T09:20:41Z")

</div>

I’m running into an issue when trying to use AcceleratedKernels.sum on a GPU SubArray (a view into a CuArray). Calling AK.sum on the full array works fine, but calling it on the view throws an InvalidIRError. Question …

---

## [Looking for Windows ARM tester](https://discourse.julialang.org/t/looking-for-windows-arm-tester/133912)

<div class="topic-metadata">

**Author:** [@GeorgeGkountouras](https://discourse.julialang.org/u/GeorgeGkountouras)\
**Replies:** 6\
**Last updated:** [November 19, 2025, 2:19pm UTC](https://discourse.julialang.org/t/looking-for-windows-arm-tester/133912 "2025-11-19T14:19:59Z")

</div>

After a successful OpenCL test on my AMD iGPU, I want to see if the same method can be used to target Snapdragon Adreno GPUs. Unfortunately, I don’t have access to one of these devices. Can someone on Windows ARM help m…

---

## [Latest CUDA.jl version 5.8.3 fails to install on NVIDIA Jetson Orin with Jetpack 6.2.1+b38](https://discourse.julialang.org/t/latest-cuda-jl-version-5-8-3-fails-to-install-on-nvidia-jetson-orin-with-jetpack-6-2-1-b38/132059)

<div class="topic-metadata">

**Author:** [@rovitotv](https://discourse.julialang.org/u/rovitotv)\
**Replies:** 6\
**Last updated:** [November 12, 2025, 10:07am UTC](https://discourse.julialang.org/t/latest-cuda-jl-version-5-8-3-fails-to-install-on-nvidia-jetson-orin-with-jetpack-6-2-1-b38/132059 "2025-11-12T10:07:59Z")

</div>

Greetings, I am trying to install CUDA.jl on a NVIDIA Jetson Orin. The installation fails with many errors captured below. I would NOT like to use CUDA via artifact and use CUDA installed on the system but when I add …

---

## [GPU memory issue on AMDGPU](https://discourse.julialang.org/t/gpu-memory-issue-on-amdgpu/133560)

<div class="topic-metadata">

**Author:** [@Alexander-Barth](https://discourse.julialang.org/u/Alexander-Barth)\
**Replies:** 4\
**Last updated:** [November 10, 2025, 7:51am UTC](https://discourse.julialang.org/t/gpu-memory-issue-on-amdgpu/133560 "2025-11-10T07:51:57Z")

</div>

I am trying to train a neural network and make inference using this neural network with Lux (1.22.1) AMDGPU (2.1.1) and julia (1.12.0). The AMDGPU version kernel driver is 6.3.6 (as reported by /sys/module/amdgpu/versio…

---

## [How to call ssyevd](https://discourse.julialang.org/t/how-to-call-ssyevd/133496)

<div class="topic-metadata">

**Author:** [@billmclean](https://discourse.julialang.org/u/billmclean)\
**Replies:** 2\
**Last updated:** [November 9, 2025, 10:21pm UTC](https://discourse.julialang.org/t/how-to-call-ssyevd/133496 "2025-11-09T22:21:54Z")

</div>

I am trying to figure out how to call rocsolver\_ssyevd to compute the eigenvalues of a symmetric matrix. If I run the code using AMDGPU import AMDGPU.rocBLAS: rocblas\_int, handle import AMDGPU.rocSOLVER: rocsolver\_ssye…

---

## [Wrapping CUDA.jl with juliacall](https://discourse.julialang.org/t/wrapping-cuda-jl-with-juliacall/133554)

<div class="topic-metadata">

**Author:** [@chrissm23](https://discourse.julialang.org/u/chrissm23)\
**Replies:** 4\
**Last updated:** [November 7, 2025, 9:09am UTC](https://discourse.julialang.org/t/wrapping-cuda-jl-with-juliacall/133554 "2025-11-07T09:09:08Z")

</div>

Hello everyone! I’m trying to make a Python wrapper using PythonCall.jl for a Julia package that uses CUDA.jl. The package initializes random integers from a UnitRange using the CPU’s Random.rand into a Matrix and then …

---

## [Q: No SubArray type required for passing partial multidimensional CuArrays?](https://discourse.julialang.org/t/q-no-subarray-type-required-for-passing-partial-multidimensional-cuarrays/133525)

<div class="topic-metadata">

**Author:** [@Audrius-St](https://discourse.julialang.org/u/Audrius-St)\
**Replies:** 4\
**Last updated:** [November 4, 2025, 6:31pm UTC](https://discourse.julialang.org/t/q-no-subarray-type-required-for-passing-partial-multidimensional-cuarrays/133525 "2025-11-04T18:31:48Z")

</div>

Hello, To pass part of a multidimensional array in Julia on the CPU on has to use the SubArray type in the function arguments. This does not seem to be the case for Julia on the GPU. Excerpted from the MWE below: …

---

## [Efficient lookup of UInt indices in large GPU arrays](https://discourse.julialang.org/t/efficient-lookup-of-uint-indices-in-large-gpu-arrays/133630)

<div class="topic-metadata">

**Author:** [@Gattu\_Mytraya](https://discourse.julialang.org/u/Gattu_Mytraya)\
**Replies:** 0\
**Last updated:** [November 3, 2025, 9:09pm UTC](https://discourse.julialang.org/t/efficient-lookup-of-uint-indices-in-large-gpu-arrays/133630 "2025-11-03T21:09:23Z")

</div>

Hi all, I have a GPU kernel where the main bottleneck is finding the index of an unsigned integer in a large constant array (lll\_basis in the code below). Typical array sizes are around \\sim 10^7, but I am hoping to wo…

---

## [Understanding and optimizing Enzyme.jl Reverse AD on CUDA](https://discourse.julialang.org/t/understanding-and-optimizing-enzyme-jl-reverse-ad-on-cuda/133357)

<div class="topic-metadata">

**Author:** [@Riccardo\_Buscicchio](https://discourse.julialang.org/u/Riccardo_Buscicchio)\
**Replies:** 5\
**Last updated:** [October 25, 2025, 7:36am UTC](https://discourse.julialang.org/t/understanding-and-optimizing-enzyme-jl-reverse-ad-on-cuda/133357 "2025-10-25T07:36:54Z")

</div>

Hi all, we are working on a nested sampling algorithm whose proposals are assisted by HMC, and we want to leverage Enzyme.jl to autodiff our loglikelihood kernel. We have a MWE on a simple 4D gaussian likelihood (4 par…

---

## [Error in oneAPI.jl tests](https://discourse.julialang.org/t/error-in-oneapi-jl-tests/133252)

<div class="topic-metadata">

**Author:** [@raman\_kumar](https://discourse.julialang.org/u/raman_kumar)\
**Replies:** 3\
**Last updated:** [October 18, 2025, 9:01pm UTC](https://discourse.julialang.org/t/error-in-oneapi-jl-tests/133252 "2025-10-18T21:01:34Z")

</div>

I want to use oneAPI.jl as i have Intel GPU integrated in CPU: 14 × Intel(R) Core™ Ultra 7 155U. I am using Manjaro Linux (x86\_64-linux-gnu). On running tests i got errors. pkg\> test oneAPI ▶ Error output Is there any…

---

## [CUDA.jl calling kernels in parallel?](https://discourse.julialang.org/t/cuda-jl-calling-kernels-in-parallel/133032)

<div class="topic-metadata">

**Author:** [@jjgarzella](https://discourse.julialang.org/u/jjgarzella)\
**Replies:** 1\
**Last updated:** [October 11, 2025, 5:52am UTC](https://discourse.julialang.org/t/cuda-jl-calling-kernels-in-parallel/133032 "2025-10-11T05:52:43Z")

</div>

Hi, I’m trying to use CUDA.jl and CUBLAS to implement a mod N matrix multiplication using Algorithm 1 of this paper (see the HAL preprint). The inner loop of the algorithm requires alternating calls to CUBLAS with a broa…

---

## [Mixing CUDA.jl with external GPU compute (OpenMM / DLPack.jl)](https://discourse.julialang.org/t/mixing-cuda-jl-with-external-gpu-compute-openmm-dlpack-jl/133015)

<div class="topic-metadata">

**Author:** [@axsk](https://discourse.julialang.org/u/axsk)\
**Replies:** 0\
**Last updated:** [October 9, 2025, 2:45pm UTC](https://discourse.julialang.org/t/mixing-cuda-jl-with-external-gpu-compute-openmm-dlpack-jl/133015 "2025-10-09T14:45:38Z")

</div>

I am using OpenMM, a molecular dynamics package, through PyCall. Now I want to combine my Julia integrator with the OpenMM force computation (on GPU). To this end I tried using DLPack.jl together with openmm-dlext (man…

---

## [Dense Matrix sparse binary vector product](https://discourse.julialang.org/t/dense-matrix-sparse-binary-vector-product/132833)

<div class="topic-metadata">

**Author:** [@Fabrice\_Rosay](https://discourse.julialang.org/u/Fabrice_Rosay)\
**Replies:** 2\
**Last updated:** [October 7, 2025, 7:28am UTC](https://discourse.julialang.org/t/dense-matrix-sparse-binary-vector-product/132833 "2025-10-07T07:28:01Z")

</div>

Hi, I need to train very shallow networks(2 dense layers) whose inputs are sparse binary vector (i.e. only 0 and 1 entries, batched, it is for NNUE). I tried to write custom kernels for forward and backward pass but the…

---

## [CUDA | Avoid divide by zero in kernel using assume()](https://discourse.julialang.org/t/cuda-avoid-divide-by-zero-in-kernel-using-assume/132750)

<div class="topic-metadata">

**Author:** [@Audrius-St](https://discourse.julialang.org/u/Audrius-St)\
**Replies:** 10\
**Last updated:** [October 7, 2025, 6:26am UTC](https://discourse.julialang.org/t/cuda-avoid-divide-by-zero-in-kernel-using-assume/132750 "2025-10-07T06:26:56Z")

</div>

Hello, To avoid branching on a divide by zero test in a CUDA kernel, I am trying to use LLVM.Interop.assume() following the one example I’ve found in the CUDA docs: Below is a MNonWE. The parametric struct initializa…

---

## [Slow matrix multiplication in CUBLAS.gemm\_strided\_batched with ComplexF64](https://discourse.julialang.org/t/slow-matrix-multiplication-in-cublas-gemm-strided-batched-with-complexf64/132826)

<div class="topic-metadata">

**Author:** [@JackC](https://discourse.julialang.org/u/JackC)\
**Replies:** 1\
**Last updated:** [October 7, 2025, 6:25am UTC](https://discourse.julialang.org/t/slow-matrix-multiplication-in-cublas-gemm-strided-batched-with-complexf64/132826 "2025-10-07T06:25:05Z")

</div>

I’ve been developing some GPU code that seems does a lot of matrix multiplications with complex numbers. I figured I’d use the CUBLAS implementations that are available in the CUDA.jl package. However, I’ve found that us…

---

## [KernelAbstractions + Enzyme - how to do GPU-side autodiff?](https://discourse.julialang.org/t/kernelabstractions-enzyme-how-to-do-gpu-side-autodiff/132643)

<div class="topic-metadata">

**Author:** [@jl\_enthusiast](https://discourse.julialang.org/u/jl_enthusiast)\
**Replies:** 1\
**Last updated:** [September 25, 2025, 6:34am UTC](https://discourse.julialang.org/t/kernelabstractions-enzyme-how-to-do-gpu-side-autodiff/132643 "2025-09-25T06:34:49Z")

</div>

Hello, I recently tried to (and failed) to auto-differentiate kernels with Enzyme.jl, written using KernelAbstractions.jl. I tried to search for examples in other packages, but could not find any concrete examples of E…

---

## [RCCL wrapping](https://discourse.julialang.org/t/rccl-wrapping/132287)

<div class="topic-metadata">

**Author:** [@fran94](https://discourse.julialang.org/u/fran94)\
**Replies:** 4\
**Last updated:** [September 20, 2025, 12:32pm UTC](https://discourse.julialang.org/t/rccl-wrapping/132287 "2025-09-20T12:32:28Z")

</div>

Hi! I’m trying to wrap the Radeon collective communication library (RCCL), the ROCm counterpart to NCCL. I’m taking heavy inspiration from NCCL.jl regarding wrapping strategy since RCCL itself closely mimics NCCL. I hav…

---

## [CUDA.jl: Unexpected \`mapreduce\` error: threads per block exceed GPU limit (640 \> 512](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778)

<div class="topic-metadata">

**Author:** [@ykkan](https://discourse.julialang.org/u/ykkan)\
**Replies:** 9\
**Last updated:** [September 18, 2025, 11:27am UTC](https://discourse.julialang.org/t/cuda-jl-unexpected-mapreduce-error-threads-per-block-exceed-gpu-limit-640-512/131778 "2025-09-18T11:27:07Z")

</div>

Hello everyone, I am using mapreduce from CUDA.jl to compute the sum of function evaluations over an array of SVector{7,T} data. The code looks roughly like this: function func(x1, x2, x3, x4, x5, x6, x7, x8, x9) …

---

## [CUDA.jl: Warning about loading library from system path](https://discourse.julialang.org/t/cuda-jl-warning-about-loading-library-from-system-path/131904)

<div class="topic-metadata">

**Author:** [@CaG21](https://discourse.julialang.org/u/CaG21)\
**Replies:** 4\
**Last updated:** [August 30, 2025, 11:42am UTC](https://discourse.julialang.org/t/cuda-jl-warning-about-loading-library-from-system-path/131904 "2025-08-30T11:42:20Z")

</div>

I am using CUDA.jl on two different computers with a completely different configuration but both equipped with Windows 11. In both cases, I got a warning when loading CUDA. The warning is ┌ Warning: CUDA runtime library…

---

## [cuSOLVER: two calls to cusolverDnDgesvdj\_bufferSize, one via Juila, the other via CUDA yield (very) different results](https://discourse.julialang.org/t/cusolver-two-calls-to-cusolverdndgesvdj-buffersize-one-via-juila-the-other-via-cuda-yield-very-different-results/125904)

<div class="topic-metadata">

**Author:** [@FHulsemann](https://discourse.julialang.org/u/FHulsemann)\
**Replies:** 2\
**Last updated:** [August 22, 2025, 5:09pm UTC](https://discourse.julialang.org/t/cusolver-two-calls-to-cusolverdndgesvdj-buffersize-one-via-juila-the-other-via-cuda-yield-very-different-results/125904 "2025-08-22T17:09:21Z")

</div>

Hello. I was wondering whether someone could explain to me why the working space for an SVD (gesvdj to be precise) differs wildly between a call to cusolverDnDgesvdj\_bufferSize via the Julia interface and a call via CU…

---

## [What is the correct way to use multiple GPUs in Slurm cluster?](https://discourse.julialang.org/t/what-is-the-correct-way-to-use-multiple-gpus-in-slurm-cluster/131740)

<div class="topic-metadata">

**Author:** [@ytdHuang](https://discourse.julialang.org/u/ytdHuang)\
**Replies:** 0\
**Last updated:** [August 20, 2025, 3:52pm UTC](https://discourse.julialang.org/t/what-is-the-correct-way-to-use-multiple-gpus-in-slurm-cluster/131740 "2025-08-20T15:52:11Z")

</div>

Hello Julia community, I am trying to distribute my computation using multiple GPUs in Slurm cluster. The code can be executed correctly while I’m using single GPU board in Slurm. However, when I try to use more than on…

---

## [Trying to parallelize using CUSOLVERRF.jl with @threads](https://discourse.julialang.org/t/trying-to-parallelize-using-cusolverrf-jl-with-threads/131457)

<div class="topic-metadata">

**Author:** [@mhunke](https://discourse.julialang.org/u/mhunke)\
**Replies:** 7\
**Last updated:** [August 19, 2025, 9:09am UTC](https://discourse.julialang.org/t/trying-to-parallelize-using-cusolverrf-jl-with-threads/131457 "2025-08-19T09:09:29Z")

</div>

Hi everyone, I’m working with CUSOLVERRF.jl to solve multiple sparse linear systems on the GPU, using the high-level RFLU API: GitHub - exanauts/CUSOLVERRF.jl: A Julia wrapper for cusolverRF All systems share the same …

---

## [Metal.jl does not speed up FFT](https://discourse.julialang.org/t/metal-jl-does-not-speed-up-fft/95528)

<div class="topic-metadata">

**Author:** [@huangc](https://discourse.julialang.org/u/huangc)\
**Replies:** 8\
**Last updated:** [August 13, 2025, 10:48pm UTC](https://discourse.julialang.org/t/metal-jl-does-not-speed-up-fft/95528 "2025-08-13T22:48:55Z")

</div>

I was hoping to explore the possibility of using Metal.jl to speed up Fourier transfer without needing to copy between CPU and GPU. However, my benchmark results showed no improvement in speed which makes me suspect that…

---

## [UndefVarError: cuda\_version in Google Colab with CUDA.jl](https://discourse.julialang.org/t/undefvarerror-cuda-version-in-google-colab-with-cuda-jl/131557)

<div class="topic-metadata">

**Author:** [@rrabadan](https://discourse.julialang.org/u/rrabadan)\
**Replies:** 2\
**Last updated:** [August 12, 2025, 11:57am UTC](https://discourse.julialang.org/t/undefvarerror-cuda-version-in-google-colab-with-cuda-jl/131557 "2025-08-12T11:57:03Z")

</div>

Hello, I’m working in a Google Colab notebook with a GPU runtime and am running into UndefVarError when trying to use CUDA.jl. For instance when I run CUDA.versioninfo(): using CUDA CUDA.versioninfo() I get the follow…

---

## [Using getrf\_batched to find matrix inverses](https://discourse.julialang.org/t/using-getrf-batched-to-find-matrix-inverses/131449)

<div class="topic-metadata">

**Author:** [@JackC](https://discourse.julialang.org/u/JackC)\
**Replies:** 2\
**Last updated:** [August 7, 2025, 3:45pm UTC](https://discourse.julialang.org/t/using-getrf-batched-to-find-matrix-inverses/131449 "2025-08-07T15:45:55Z")

</div>

Hello, I am pretty new to GPU programming in general, so apologies in advanced for sounding clueless. I have a problem where I need to find the inverses of a bunch of relatively small matrices on a GPU using LU factoriz…

---

## [Sparse matrix multiplication for Metal](https://discourse.julialang.org/t/sparse-matrix-multiplication-for-metal/131088)

<div class="topic-metadata">

**Author:** [@Marco\_Lombardi](https://discourse.julialang.org/u/Marco_Lombardi)\
**Replies:** 15\
**Last updated:** [July 31, 2025, 5:59pm UTC](https://discourse.julialang.org/t/sparse-matrix-multiplication-for-metal/131088 "2025-07-31T17:59:44Z")

</div>

I need to perform a multiplication between two fairly large sparse matrices C = A \* B and I would like to try using the GPU for that. The problem arises in a context where A represents a fixed 2D convolution kernel, whil…

---

## [DiffEqGPU Trajectory Failure Handling and Heterogeneous Trajectories](https://discourse.julialang.org/t/diffeqgpu-trajectory-failure-handling-and-heterogeneous-trajectories/129962)

<div class="topic-metadata">

**Author:** [@tosm01](https://discourse.julialang.org/u/tosm01)\
**Replies:** 4\
**Last updated:** [July 22, 2025, 12:06pm UTC](https://discourse.julialang.org/t/diffeqgpu-trajectory-failure-handling-and-heterogeneous-trajectories/129962 "2025-07-22T12:06:50Z")

</div>

Hi, I have a general question about the suitability of using GPUs for the kind of problem I’m working on. So far, I’ve had good success using DiffEqGPU for parameter sweeps, specifically in cases where the parameters d…

---

## [Does AMDGPU.jl support integrated graphics?](https://discourse.julialang.org/t/does-amdgpu-jl-support-integrated-graphics/130675)

<div class="topic-metadata">

**Author:** [@Mend-Amar\_Badral](https://discourse.julialang.org/u/Mend-Amar_Badral)\
**Replies:** 3\
**Last updated:** [July 19, 2025, 10:21pm UTC](https://discourse.julialang.org/t/does-amdgpu-jl-support-integrated-graphics/130675 "2025-07-19T22:21:44Z")

</div>

I have ubuntu 24.04 and my gpu1 is nvidia gpu and gpu0 is AMD ATI Radeon RX Vega 6. I was wondering if it is possible to use AMDGPU.jl to use integrated gpu.

[Previous page](https://discourse.julialang.org/c/domain/gpu/11.md)

[Next page](https://discourse.julialang.org/c/domain/gpu/11.md?page=2)