There shows "no suitable GPU found" on WSL2 with GPU

General
1

I’d like to use GPU for DFC, but no GPU found.

Could anybody please show me solution?
Or, one cannot use GPU in WSL2?

TIA

I run commands as below in virtual environment.

(v3.32.0) $ hailo --version
[info] No GPU chosen and no suitable GPU found, falling back to CPU.
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
E0000 00:00:1755666037.165252 2297586 cuda_dnn.cc:8310] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
E0000 00:00:1755666037.168971 2297586 cuda_blas.cc:1418] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
[info] Current Time: 14:01:12, 08/20/25
[info] CPU: Architecture: x86_64, Model: 11th Gen Intel(R) Core(TM) i7-11700KF @ 3.60GHz, Number Of Cores: 16, Utilization: 0.1%
[info] Memory: Total: 31GB, Available: 29GB
[info] System info: OS: Linux, Kernel: 6.6.87.2-microsoft-standard-WSL2
[info] Hailo DFC Version: 3.32.0
[info] HailoRT Version: Not Installed
[info] PCIe: No Hailo PCIe device was found
[info] Running `hailo --version`
Hailo Dataflow Compiler v3.32.0

(v3.32.0) $ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Thu_Jun__6_02:18:23_PDT_2024
Cuda compilation tools, release 12.5, V12.5.82
Build cuda_12.5.r12.5/compiler.34385749_0

(v3.32.0) $ nvidia-smi
Wed Aug 20 14:02:46 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 560.35.02              Driver Version: 560.94         CUDA Version: 12.6     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3060        On  |   00000000:01:00.0  On |                  N/A |
| 35%   33C    P8             17W /  170W |     951MiB /  12288MiB |      3%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A        27      G   /Xwayland                                   N/A      |
+-----------------------------------------------------------------------------------------+
3

I have been struggling with the same issues for the past few days. How are you installing into WSL, are you using Docker or direct installation via the wheel?

I have been trying to install directly using the wheel and I have come to the conclusion that DataflowCompiler 3.32.0 appears to not work at all in WSL even CPU only. After trying many configurations and still getting errors I have rolled back to 3.31.0.

I believe I have a working CPU only install of Dataflow Compiler in WSL from the following:

sudo apt-get update
sudo apt-get upgrade
sudo apt-get install -y unzip python3-dev python3-pip python3-tk graphviz graphviz-dev build-essential virtualenv
mkdir ~/hailo
cd ~/hailo
virtualenv ./hailo_venv
. hailo_venv/bin/activate
pip install hailo_dataflow_compiler-3.31.0-py3-none-linux_x86_64.whl
hailo -h
hailo tutorial &

I think Hailo have said that GPU is not working under WSL, but I do not know if that applies only to the latest release (which seems to not work regardless) or if previous releases had the same limitation.

4

After a bit more digging I think my statement that 3.32.0 has problems may be incorrect. I have also now run into a similar issue with 3.31.0 as well. I don’t fully understand the problem yet but I think my issues relate to this bug in TensorFlow

My guess is that since GPU support is broken for Dataflow Compiler in WSL which is causing it to fall back to CPU only and then hitting the above bug and causing a crash. I can replicate my issues using Tensorflow only so the issue doesnt appear to be with Dataflow Compiler itself but with the overall installed environment in WSL.

5

The Dataflow Compiler usage on Windows using WSL2, does not currently support GPU.

6

Thank you for your effort and explanation.

7

I’ll use direct Ubuntu with GPU.

Thank you.

1 Like
8

Just to update my findings, I did manage get a working compiler under WSL2 by rolling back to Nvidia Graphics Driver 572.83 and using dataflow_compiler_3.31.0

This is CPU only using “CUDA_VISIBLE_DEVICES”=“” but can do QAT although rather slowly as expected.

3.32.0 still throws a bunch of errors on my QAT training scripts but thats a separate issue that doesnt seem related to WSL.

9

@KlausK I have a question: is any of DFC version is incompatible supporting GPU using wsl or only DFC 3.32

10

Yes, the Dataflow Compiler on Windows with WSL2 does not currently support GPU in any version.

11

I’m a bit confused after reviewing the information from the link, It states that GPU utilization in WSL is supported when using DFC .whl file. However, it also mentions that GPU support is not available when using Docker Suite.

Could you please help clarify the following points?

  1. hailo_dataflow_compiler-3.30.0-py3-none-linux_x86_64.whl Python wheel file – If installed within WSL, will it be able to utilize the GPU?

  2. Docker Suite – When running in WSL, does it support GPU acceleration?

Thank you in advance for your assistance. @KlausK

12

My statement above stands. Support for Hailo means that we develop a feature, document it, and validate it. Whether someone with enough experience and determination can get it working in other environments, I cannot say. There are many smart users in our community, but sometimes fundamental issues prevent a feature from working.

WSL2 support has been developed to allow 3rd party software available only on Windows to use the Hailo Dataflow Compiler under the hood with the known limitations.

Unfortunately, we cannot provide assistance when problems occur under WSL2. I and my colleagues from the customer success team run the tools natively on Ubuntu, and we recommend you do the same so that we can support you effectively when issues arise.

We have added a statement in the Hailo Dataflow Compiler User Guide version 5.x. I asked my colleagues to also add it to the Hailo-8 version 3.x.

13

I understand your advice to use it natively @KlausK, but I’m in a situation where I can’t multi-boot my PC. Based on this statement, the limitations mean it doesn’t support GPU for the optimization process?

14

Yes (this answer requires at least 20 characters :slight_smile: )