# CUDA 13.4.1 nvcc 13.4.59 rejects complete Visual Studio Build Tools 2026 with “Host compiler targets unsupported OS”

**URL:** <https://forums.developer.nvidia.com/t/cuda-13-4-1-nvcc-13-4-59-rejects-complete-visual-studio-build-tools-2026-with-host-compiler-targets-unsupported-os/383253>\
**Category:** CUDA NVCC Compiler\
**Tags:** windows, visual-studio, nvcc\
**Created:** [September 14, 2026, 3:54pm UTC](https://forums.developer.nvidia.com/t/cuda-13-4-1-nvcc-13-4-59-rejects-complete-visual-studio-build-tools-2026-with-host-compiler-targets-unsupported-os/383253 "2026-09-14T15:54:37Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![magen-danielsystems](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@magen-danielsystems](https://forums.developer.nvidia.com/u/magen-danielsystems)\
**Post date:** [September 14, 2026, 3:54pm UTC](https://forums.developer.nvidia.com/t/cuda-13-4-1-nvcc-13-4-59-rejects-complete-visual-studio-build-tools-2026-with-host-compiler-targets-unsupported-os/383253/1 "2026-09-14T15:54:37Z")

</div>

I am seeing a reproducible host-platform rejection when compiling a minimal CUDA program with CUDA Toolkit 13.4.1.

System:

```plaintext
Windows 11 Pro x64 25H2
OS build 26200.9445
GPU: NVIDIA RTX 5070 Ti
NVIDIA driver: 616.92
CUDA Toolkit: 13.4.1
nvcc: V13.4.59
nvcc --list-gpu-code includes sm_120

```

Host toolchain:

```plaintext
Visual Studio Build Tools 2026 Stable 18.10.0
Installation version: 18.10.12201.205
MSVC toolset: 14.50.35717
cl.exe: 19.50.35738, x64
Windows SDK: 10.0.26100.0
Complete=True
Launchable=True
RebootRequired=False

```

The Visual Studio environment was initialized with:

```cmd
call "<VS_BUILD_TOOLS_INSTALL_PATH>\Common7\Tools\VsDevCmd.bat" -host_arch=x64 -arch=x64 -vcvars_ver=14.50

```

Relevant environment values were:

```plaintext
VSCMD_ARG_HOST_ARCH=x64
VSCMD_ARG_TGT_ARCH=x64
VSCMD_VER=18.10.0
VisualStudioVersion=18.0
VCToolsInstallDir=<VS_BUILD_TOOLS_INSTALL_PATH>\VC\Tools\MSVC\14.50.35717\
VCToolsVersion=14.50.35717
WindowsSdkDir=C:\Program Files (x86)\Windows Kits\10\
WindowsSDKVersion=10.0.26100.0\

```

`cl.exe`, `link.exe`, and `lib.exe` all resolved beneath the `Hostx64\x64` directory for MSVC 14.50.35717.

The current NVIDIA Windows installation guide lists Windows 11 25H2, Visual Studio 2026 18.x, MSVC 195x, and native x86\_64 support:

Official NVIDIA documentation: [CUDA Installation Guide for Microsoft Windows — Installation Guide Windows 13.4 documentation](https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html)

This documents the stated compatibility combination, but it does not prove that nvcc V13.4.59 implements that support correctly.

Reproduction:

```cmd
call "<VS_BUILD_TOOLS_INSTALL_PATH>\Common7\Tools\VsDevCmd.bat" -host_arch=x64 -arch=x64 -vcvars_ver=14.50

```

```cmd
"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\bin\nvcc.exe" --version

```

```cmd
"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\bin\nvcc.exe" --list-gpu-code

```

Normal invocation:

```cmd
"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\bin\nvcc.exe" -ccbin "<VS_BUILD_TOOLS_INSTALL_PATH>\VC\Tools\MSVC\14.50.35717\bin\Hostx64\x64" -arch=sm_120 -std=c++17 -Xcompiler="/EHsc /W4 /WX" -o "cuda-smoke.exe" "cuda-smoke.cu"

```

Invocation with `--use-local-env`:

```cmd
"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\bin\nvcc.exe" --use-local-env -ccbin "<VS_BUILD_TOOLS_INSTALL_PATH>\VC\Tools\MSVC\14.50.35717\bin\Hostx64\x64" -arch=sm_120 -std=c++17 -Xcompiler="/EHsc /W4 /WX" -o "cuda-smoke.exe" "cuda-smoke.cu"

```

`--use-local-env` forces nvcc to use the already initialized local compiler environment.

Both invocations start the MSVC compiler successfully, then fail with:

```plaintext
Microsoft (R) C/C++ Optimizing Compiler Version 19.50.35738 for x64
Copyright (C) Microsoft Corporation. All rights reserved.

tmpxft_...-2.cpp
nvcc fatal : Host compiler targets unsupported OS.

```

I can provide the exact minimal `cuda-smoke.cu` used for this reproduction if needed.

The Visual Studio Build Tools installation is reported complete and launchable. Visual Studio 2026 initializes the compiler through `VsDevCmd.bat` and `vsdevcmd\ext\vcvars.bat`. The legacy `vcvarsall.bat` and `vcvars64.bat` files are absent from this installation.

Several historical community reports associate this diagnostic with nvcc probing for legacy `VC\Auxiliary\Build\vcvarsall.bat` files. That is a testable hypothesis based on community reports only, and is not being presented as NVIDIA policy or as an established requirement.

No application source code is involved. The standalone CUDA source reproduces the issue. The CUDA installation, driver, headers, libraries, x64 compiler environment, and `sm_120` listing were validated. No CUDA executable was produced.

Could NVIDIA please clarify:

1. Does nvcc 13.4.59 officially support the Visual Studio Build Tools 2026 layout?
2. Does nvcc incorrectly perform legacy host-environment probing when `--use-local-env` is supplied?
3. Is this a known CUDA 13.4.1 defect?
4. Is a corrected nvcc build or CUDA 13.4 update planned?
5. Is there a supported configuration-only workaround that does not use dummy files, `--allow-unsupported-compiler`, compiler replacement, or unsupported toolsets?
6. If Visual Studio 2022 is temporarily required, which exact supported MSVC compiler family and toolset should be used?

Expected behavior: nvcc should accept the documented Windows 11, Visual Studio, MSVC, and x64 combination and compile the program.

Actual behavior: nvcc launches `cl.exe` and then reports:

```plaintext
nvcc fatal : Host compiler targets unsupported OS.

```