Anyone managed to get it to run on Windows 10 or Windows 11?

[Greg7000]

Anyone managed to get it to run on Windows 10 or Windows 11 ?

(Ideally cmake build --no-container-build from Dockerfile.win10.min )

I have tried way too many combinations, patches, and various releases without success. I thought I was close many times (getting far into the build), but I always face an issue at some point. I'm starting to lose my mind a bit.

Latest try (using the provided .exe from https://github.com/triton-inference-server/server/releases/tag/v2.54.0 and the DLL(s) from the build image):

dll error: 0x0000007b ...

The goal of this discussion is to share the success story of your Windows adventure.
(SUCCESS ONLY, since Issues is more appropriate for failure stories)

Ideally (not mandatory), try to use the following as a starting point:

docker build -t win10-py3-min -f Dockerfile.win10.min .

Please provide the steps (build command), commit used for your success.
If possible the nividia-smi of the host launching the triton-server.exe would also be great.

[Greg7000]

Execution wise

I am now able to avoid the 0x0000007b issue.

I don't have much time right now so here is basically how I did it.

• Recover & extract tritonserver2.54.0-win.zip
• Build image docker build -t win10-py3-min -f Dockerfile.win10.min .
• Enter the image, mount & extract (docker run -it -v C:\your_local_path:C:\your_remote_path --entrypoint "cmd.exe" win10-py3-min:latest /k "C:\BuildTools\VC\Auxiliary\Build\vcvars64.bat")
• From the container, extract b64.dll, libcrypto-3-x64.dll, libssl-3-x64.dll, pthreadVC3.dll from your container and bring them back on your real PC. (Google: Search & copy file via commande line)
• Place all of them into tritonserver\bin (from tritonserver2.54.0-win.zip)
• Try to get a valid nvidia-smi (comes with nvidia driver), nvcc (comes with CUDA toolkit) conf (try to be above or equal what is available in Dockerfile.win10.min .) Here is what I used:
  nvdia-smi

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 573.48                 Driver Version: 573.48         CUDA Version: 12.8     |
|-----------------------------------------+------------------------+----------------------+
...

nvcc (note that I am using a conda nvcc)

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Wed_Jan_15_19:38:46_Pacific_Standard_Time_2025
Cuda compilation tools, release 12.8, V12.8.61
Build cuda_12.8.r12.8/compiler.35404655_0

• From a cmd line that provide you the right nvidia-smi, nvcc info, start tritonserver\bin\tritonserver.exe, you shoud get

+----------------------------------+----------------------------------------------------------------------------------+
| Option                           | Value                                                                            |
+----------------------------------+----------------------------------------------------------------------------------+
| server_id                        | triton                                                                           |
| server_version                   | 2.54.0                                                                           |
| server_extensions                | classification sequence model_repository model_repository(unload_dependents) sch |
|                                  | edule_policy model_configuration system_shared_memory cuda_shared_memory binary_ |
|                                  | tensor_data parameters statistics trace logging                                  |
| model_control_mode               | MODE_NONE                                                                        |
| strict_model_config              | 0                                                                                |
| model_config_name                |                                                                                  |
| rate_limit                       | OFF                                                                              |
| pinned_memory_pool_byte_size     | 268435456                                                                        |
| min_supported_compute_capability | 6.0                                                                              |
| strict_readiness                 | 1                                                                                |
| exit_timeout                     | 30                                                                               |
| cache_enabled                    | 0                                                                                |
+----------------------------------+----------------------------------------------------------------------------------+

Build wise

Still no success but now with my various attempts I get really far in the build log I am pretty confident I can find some ways to fix the issues.

• For Docker most likely an issue regarding my docker, local nvidia-smi, local nvcc versions (Getting a corrupted error at some point)
• For local cmake (VS 2022), getting even farther in the build but I eventually get some cuda toolkit mismatch issue (With warning recommending setting CudaToolkitDir which strongly suggest that my conda cudatoolkit is probably too minimal for this build and I could probably fix by simply installing the real cuda toolkit from installer.

Note that you need some patches to build.py if you want to use conda since it starts powershell in the background and messes up your activated conda env.

I now need to put this task aside for now but if I come back to it and manage to build it for windows. I will try to take the time to come update this discussion.