# NVIDIA CUDA **URL:** https://dev-discuss.pytorch.org/c/hardware-backends/cuda/13.md [Latest](https://dev-discuss.pytorch.org/latest.md) · [Categories](https://dev-discuss.pytorch.org/categories.md) --- ## [About the NVIDIA CUDA category](https://dev-discuss.pytorch.org/t/about-the-nvidia-cuda-category/22)
**Author:** [@smth](https://dev-discuss.pytorch.org/u/smth)\ **Replies:** 0
--- ## [Race conditions in persistent kernels](https://dev-discuss.pytorch.org/t/race-conditions-in-persistent-kernels/3355)
**Author:** [@lcw](https://dev-discuss.pytorch.org/u/lcw)\ **Replies:** 0\ **Last updated:** [April 16, 2026, 1:14pm UTC](https://dev-discuss.pytorch.org/t/race-conditions-in-persistent-kernels/3355 "2026-04-16T13:14:34Z")
TL;DR: Probably no one really understands CUDA’s async proxy. We didn’t, and thus encountered race conditions when shared memory repeatedly alternates between regular threads and TMA, CLC, … We’ve since found instances o… --- ## [CuSolver DnXgeev faster CUDA Eigenvalue calculations](https://dev-discuss.pytorch.org/t/cusolver-dnxgeev-faster-cuda-eigenvalue-calculations/3248)
**Author:** [@Johannes99](https://dev-discuss.pytorch.org/u/Johannes99)\ **Replies:** 9\ **Last updated:** [November 4, 2025, 7:48pm UTC](https://dev-discuss.pytorch.org/t/cusolver-dnxgeev-faster-cuda-eigenvalue-calculations/3248 "2025-11-04T19:48:30Z")
Hello, as I am using torch as a backend for a simulation that relies heavily on solving non-hermetian eigenvalue problems I have stumbeled across the quite slow Magma geev implementation. In the process of troubleshoot… --- ## [OpenGL interoperability](https://dev-discuss.pytorch.org/t/opengl-interoperability/2696)
**Author:** [@cajoek](https://dev-discuss.pytorch.org/u/cajoek)\ **Replies:** 2\ **Last updated:** [September 30, 2025, 6:04pm UTC](https://dev-discuss.pytorch.org/t/opengl-interoperability/2696 "2025-09-30T18:04:49Z")
:rocket: The feature, motivation and pitch Zero-copy transfer of data between PyTorch and OpenGL on GPU by including “OpenGL interoperability” from CUDA in pytorch. I am working on a real-time machine learning graphics … --- ## [Using Nsight Systems to profile GPU workload](https://dev-discuss.pytorch.org/t/using-nsight-systems-to-profile-gpu-workload/59)
**Author:** [@ptrblck](https://dev-discuss.pytorch.org/u/ptrblck)\ **Replies:** 12\ **Last updated:** [April 30, 2025, 6:10am UTC](https://dev-discuss.pytorch.org/t/using-nsight-systems-to-profile-gpu-workload/59 "2025-04-30T06:10:57Z")
This topic describes a common workflow to profile workloads on the GPU using Nsight Systems. As an example, let’s profile the forward, backward, and optimizer.step() methods using the resnet18 model from torchvision. T… --- ## [Does PyTorch support RTX 5090?](https://dev-discuss.pytorch.org/t/does-pytorch-support-rtx-5090/2849)
**Author:** [@zeshengzong](https://dev-discuss.pytorch.org/u/zeshengzong)\ **Replies:** 1\ **Last updated:** [March 14, 2025, 3:51pm UTC](https://dev-discuss.pytorch.org/t/does-pytorch-support-rtx-5090/2849 "2025-03-14T15:51:27Z")
If not, is there a plan to support it? Thanks! --- ## [How profiling Pytorch Using Nsight Compute?](https://dev-discuss.pytorch.org/t/how-profiling-pytorch-using-nsight-compute/2530)
**Author:** [@woongjoonchoi](https://dev-discuss.pytorch.org/u/woongjoonchoi)\ **Replies:** 2\ **Last updated:** [October 16, 2024, 4:20am UTC](https://dev-discuss.pytorch.org/t/how-profiling-pytorch-using-nsight-compute/2530 "2024-10-16T04:20:39Z")
I am currently trying to profile the evaluation of RT-DETR github using Nsight Comput. However, the following problems occur. RT-DETR : RT-DETR/rtdetr\_pytorch at main · lyuwenyu/RT-DETR (github.com) ncu -o ./rt\_deter\_pr… --- ## [RFC Proposal for CUDA-Accelerated Dynamic Time Warping (DTW) Implementation in PyTorch](https://dev-discuss.pytorch.org/t/rfc-proposal-for-cuda-accelerated-dynamic-time-warping-dtw-implementation-in-pytorch/2292)
**Author:** [@vis7tan](https://dev-discuss.pytorch.org/u/vis7tan)\ **Replies:** 2\ **Last updated:** [September 20, 2024, 9:44pm UTC](https://dev-discuss.pytorch.org/t/rfc-proposal-for-cuda-accelerated-dynamic-time-warping-dtw-implementation-in-pytorch/2292 "2024-09-20T21:44:50Z")
Hi PyTorch Community, I’m excited to share an idea that I believe could significantly enhance the performance of time-series analysis in PyTorch. I’ve developed a CUDA-accelerated implementation of the Dynamic Time Warp… --- ## [Overlapping device to host copy with GPU collectives](https://dev-discuss.pytorch.org/t/overlapping-device-to-host-copy-with-gpu-collectives/2120)
**Author:** [@YangFei1990](https://dev-discuss.pytorch.org/u/YangFei1990)\ **Replies:** 5\ **Last updated:** [June 4, 2024, 4:30pm UTC](https://dev-discuss.pytorch.org/t/overlapping-device-to-host-copy-with-gpu-collectives/2120 "2024-06-04T16:30:48Z")
Hello, I’m running tests on AWS P5 instances and I’m trying to asynchronously offload tensors from device to host while doing GPU collectives. My impression is that DtoH is using the PCIe and GPU CC is using RDMA through… --- ## [How to share CUcontext with other application?](https://dev-discuss.pytorch.org/t/how-to-share-cucontext-with-other-application/1322)
**Author:** [@mthrok](https://dev-discuss.pytorch.org/u/mthrok)\ **Replies:** 0\ **Last updated:** [June 12, 2023, 9:09pm UTC](https://dev-discuss.pytorch.org/t/how-to-share-cucontext-with-other-application/1322 "2023-06-12T21:09:52Z")
Hi I am n00b in CUDA and have questions. Is there a way to activate CUcontext from PyTorch? (perhaps just sending a dummy Tensor to CUDA would do?) proper way to retrieve the flag used to activate CUcontext by PyTorc… --- ## [PyTorch and TensorFloat32](https://dev-discuss.pytorch.org/t/pytorch-and-tensorfloat32/504)
**Author:** [@mruberry](https://dev-discuss.pytorch.org/u/mruberry)\ **Replies:** 5\ **Last updated:** [June 28, 2022, 6:17pm UTC](https://dev-discuss.pytorch.org/t/pytorch-and-tensorfloat32/504 "2022-06-28T18:17:03Z")
Context TensorFloat32 (TF32) is a math mode introduced with NVIDIA’s Ampere GPUs. When enabled, it computes float32 GEMMs faster but with reduced numerical accuracy. For many programs this results in a significant speedu… --- ## [More In-Depth Details of Floating Point Precision](https://dev-discuss.pytorch.org/t/more-in-depth-details-of-floating-point-precision/654)
**Author:** [@syed-ahmed](https://dev-discuss.pytorch.org/u/syed-ahmed)\ **Replies:** 0\ **Last updated:** [June 21, 2022, 8:59pm UTC](https://dev-discuss.pytorch.org/t/more-in-depth-details-of-floating-point-precision/654 "2022-06-21T20:59:42Z")
Hi PyTorch Community! This post is a supplementary material to our soon to be published “What Every User Should Know About Mixed Precision Training in PyTorch” blog post. We hope this would help you use mixed precision …