Couldn't detect CUDA/GPU with ERROR: No GPU client found

[XLin0mu]

Reactant.jl couldn't detect my GPU:

julia> using Reactant

julia> Reactant.set_default_backend("gpu")
ERROR: No GPU client found
Stacktrace:
 [1] error(s::String)
   @ Base .\error.jl:35
 [2] client(backend::String)
   @ Reactant.XLA C:\Users\Xlin0\.julia\packages\Reactant\8rzTQ\src\xla\XLA.jl:87
 [3] set_default_backend
   @ C:\Users\Xlin0\.julia\packages\Reactant\8rzTQ\src\xla\XLA.jl:109 [inlined]
 [4] set_default_backend(backend::String)
   @ Reactant C:\Users\Xlin0\.julia\packages\Reactant\8rzTQ\src\Reactant.jl:350
 [5] top-level scope
   @ REPL[6]:1

Progress in Configuration | Reactant.jl doesn't work, and no any "verbose information from Reactant during its installation process" occurred:

julia> rm(joinpath(Base.DEPOT_PATH[1], "compiled", "v$(VERSION.major).$(VERSION.minor)", "Reactant_jll"); recursive=true, force=true)

julia> ENV["JULIA_DEBUG"] = "Reactant_jll";

julia> import Pkg

julia> Pkg.add("Reactant_jll")
   Resolving package versions...
  No Changes to `C:\Users\Xlin0\.julia\environments\v1.10\Project.toml`
  No Changes to `C:\Users\Xlin0\.julia\environments\v1.10\Manifest.toml`
Precompiling packages finished.
  22 dependencies successfully precompiled in 164 seconds. 799 already precompiled.

julia> using Reactant_jll

julia> Reactant_jll.is_available()
true

julia> Reactant_jll.host_platform
Windows x86_64 {cxxstring_abi=cxx11, gpu=none, gpu_version=none, julia_version=1.10.10, libgfortran_version=5.0.0, mode=opt}

I've tried to delete Reactant_jll manually:

(@v1.10) pkg> rm Reactant, Reactant_jll
    Updating `C:\Users\Xlin0\.julia\environments\v1.10\Project.toml`
  [3c362404] - Reactant v0.2.178
  [0192cb87] - Reactant_jll v0.0.261+0
    Updating `C:\Users\Xlin0\.julia\environments\v1.10\Manifest.toml`
  [3c362404] - Reactant v0.2.178
  [0192cb87] - Reactant_jll v0.0.261+0

(@v1.10) pkg> gc
      Active manifest files: 4 found
      Active artifact files: 123 found
      Active scratchspaces: 5 found
     Deleted no artifacts, repos, packages or scratchspaces

julia> rm(joinpath(Base.DEPOT_PATH[1], "compiled", "v$(VERSION.major).$(VERSION.minor)", "Reactant_jll"); recursive=true, force=true)

julia> ENV["JULIA_DEBUG"] = "Reactant_jll";

julia> import Pkg

julia> Pkg.add("Reactant_jll")
   Resolving package versions...
    Updating `C:\Users\Xlin0\.julia\environments\v1.10\Project.toml`
  [0192cb87] + Reactant_jll v0.0.264+0
    Updating `C:\Users\Xlin0\.julia\environments\v1.10\Manifest.toml`
  [0192cb87] + Reactant_jll v0.0.264+0
Precompiling packages finished.
  1 dependency successfully precompiled in 9 seconds. 799 already precompiled.

julia> using Reactant_jll

julia> Reactant_jll.is_available()
true

julia> Reactant_jll.host_platform
Windows x86_64 {cxxstring_abi=cxx11, gpu=none, gpu_version=none, julia_version=1.10.10, libgfortran_version=5.0.0, mode=opt}

And tried to restart julia session after deleting, which is not working with me (different from this post, libcuda was not founded correctly in my case).

I've reinstalled CUDA:

shell> nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Wed_Aug_20_13:58:20_Pacific_Daylight_Time_2025
Cuda compilation tools, release 13.0, V13.0.88
Build cuda_13.0.r13.0/compiler.36424714_0

And CUDA.jl could detect it:

julia> using CUDA

julia> CUDA.functional()
true

julia> devices()
CUDA.DeviceIterator() for 1 devices:
0. NVIDIA GeForce RTX 3060 Ti

julia> CUDA.versioninfo()
CUDA toolchain:
- runtime 13.0, artifact installation
- driver 581.15.0 for 13.0
- compiler 13.0

CUDA libraries:
- CUBLAS: 13.1.0
- CURAND: 10.4.0
- CUFFT: 12.0.0
- CUSOLVER: 12.0.4
- CUSPARSE: 12.6.3
- CUPTI: 2025.3.1 (API 13.0.1)
- NVML: 13.0.0+581.15

Julia packages:
- CUDA: 5.9.4
- CUDA_Driver_jll: 13.0.2+0
- CUDA_Compiler_jll: 0.3.0+0
- CUDA_Runtime_jll: 0.19.2+0

Toolchain:
- Julia: 1.10.10
- LLVM: 15.0.7

1 device:
  0: NVIDIA GeForce RTX 3060 Ti (sm_86, 2.484 GiB / 8.000 GiB available)

But nothing changed.

I've tried julia 1.11.7 and 1.12.1, both of them can reproduce the appearance:

julia> versioninfo()
Julia Version 1.12.1
Commit ba1e628ee4 (2025-10-17 13:02 UTC)
Build Info:
  Official https://julialang.org release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 24 × 13th Gen Intel(R) Core(TM) i7-13700F
  WORD_SIZE: 64
  LLVM: libLLVM-18.1.7 (ORCJIT, alderlake)
  GC: Built with stock GC
Threads: 1 default, 1 interactive, 1 GC (on 24 virtual cores)

julia> rm(joinpath(Base.DEPOT_PATH[1], "compiled", "v$(VERSION.major).$(VERSION.minor)", "Reactant_jll"); recursive=true, force=true)

julia> ENV["JULIA_DEBUG"] = "Reactant_jll";

julia> import Pkg

julia> Pkg.add("Reactant_jll")
   Resolving package versions...
     Project No packages added to or removed from `C:\Users\Xlin0\.julia\environments\v1.12\Project.toml`
    Manifest No packages added to or removed from `C:\Users\Xlin0\.julia\environments\v1.12\Manifest.toml`
Precompiling packages finished.
  1 dependency successfully precompiled in 3 seconds. 29 already precompiled.

julia> using Reactant_jll

julia> Reactant_jll.host_platform
Windows x86_64 {cxxstring_abi=cxx11, gpu=none, gpu_version=none, julia_version=1.12.1, libgfortran_version=5.0.0, mode=opt}

and 1.11.7:

julia> versioninfo()
Julia Version 1.11.7
Commit f2b3dbda30 (2025-09-08 12:10 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 24 × 13th Gen Intel(R) Core(TM) i7-13700F
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, alderlake)
Threads: 1 default, 0 interactive, 1 GC (on 24 virtual cores)

julia> rm(joinpath(Base.DEPOT_PATH[1], "compiled", "v$(VERSION.major).$(VERSION.minor)", "Reactant_jll"); recursive=true, force=true)

julia>

julia> ENV["JULIA_DEBUG"] = "Reactant_jll";

julia> import Pkg

julia> Pkg.add("Reactant_jll")
   Resolving package versions...
    Updating `C:\Users\Xlin0\.julia\environments\v1.11\Project.toml`
  [0192cb87] + Reactant_jll v0.0.264+0
    Updating `C:\Users\Xlin0\.julia\environments\v1.11\Manifest.toml`
  [692b3bcd] + JLLWrappers v1.7.1
  [21216c6a] + Preferences v1.5.0
  [0192cb87] + Reactant_jll v0.0.264+0
  [0dad84c5] + ArgTools v1.1.2
  [56f22d72] + Artifacts v1.11.0
  [2a0f44e3] + Base64 v1.11.0
  [ade2ca70] + Dates v1.11.0
  [f43a241f] + Downloads v1.6.0
  [7b1f6079] + FileWatching v1.11.0
  [4af54fe1] + LazyArtifacts v1.11.0
  [b27032c2] + LibCURL v0.6.4
  [76f85450] + LibGit2 v1.11.0
  [8f399da3] + Libdl v1.11.0
  [56ddb016] + Logging v1.11.0
  [d6f4376e] + Markdown v1.11.0
  [ca575930] + NetworkOptions v1.2.0
  [44cfe95a] + Pkg v1.11.0
  [de0858da] + Printf v1.11.0
  [9a3f8284] + Random v1.11.0
  [ea8e919c] + SHA v0.7.0
  [fa267f1f] + TOML v1.0.3
  [a4e569a6] + Tar v1.10.0
  [cf7118a7] + UUIDs v1.11.0
  [4ec0a83e] + Unicode v1.11.0
  [e66e0078] + CompilerSupportLibraries_jll v1.1.1+0
  [deac9b47] + LibCURL_jll v8.6.0+0
  [e37daf67] + LibGit2_jll v1.7.2+0
  [29816b5a] + LibSSH2_jll v1.11.0+1
  [c8ffd9c3] + MbedTLS_jll v2.28.6+0
  [14a3606d] + MozillaCACerts_jll v2023.12.12
  [83775a58] + Zlib_jll v1.2.13+1
  [8e850ede] + nghttp2_jll v1.59.0+0
  [3f19e933] + p7zip_jll v17.4.0+2
Precompiling project...
  3 dependencies successfully precompiled in 3 seconds. 25 already precompiled.

julia> using Reactant_jll

julia> Reactant_jll.host_platform
Windows x86_64 {cxxstring_abi=cxx11, gpu=none, gpu_version=none, julia_version=1.11.7, libgfortran_version=5.0.0, mode=opt}

It's looks almost same with this issue, which may also not be fully solved.

[wsmoses]

ah yeah we don't build a JLL atm with native windows + cuda, but you can use WSL and CUDA should be supported.

[wsmoses]

cc @giordano could we add this as another debug warning (e.g. cuda gpu detected, but on windows which doesn't have jll)

[giordano]

Should that be done in Reactant.set_default_backend("gpu")?

[wsmoses]

We already have the warning where we detect cuda but an unsupported version in jll, that I feel like it's reasonable to put this there as well.

The early warn may be useful if people mistakenly assume it runs on GPU but we didn't find and it fell back to cpu if they didn't explicitly set device

[XLin0mu]

For supplement to this issue, I just configured a WSL with CUDA Toolkit on it, everything works fine so far:

shell> nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Wed_Aug_20_01:58:59_PM_PDT_2025
Cuda compilation tools, release 13.0, V13.0.88
Build cuda_13.0.r13.0/compiler.36424714_0

julia> using Reactant_jll

julia> Reactant_jll.host_platform
Linux x86_64 {cxxstring_abi=cxx11, gpu=cuda, gpu_version=13.0, julia_version=1.10.10, libc=glibc, libgfortran_version=5.0.0, libstdcxx_version=3.4.30, mode=opt}

julia> using Reactant

julia> Reactant.set_default_backend("gpu")
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
W0000 00:00:1763777614.857984    9395 cuda_executor.cc:1802] GPU interconnect information not available: INTERNAL: NVML doesn't support extracting fabric info or NVLink is not used by the device.
I0000 00:00:1763777614.858224    7546 service.cc:158] XLA service 0x2548a650 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1763777614.858253    7546 service.cc:166]   StreamExecutor device (0): NVIDIA GeForce RTX 3060 Ti, Compute Capability 8.6
I0000 00:00:1763777614.859631    7546 se_gpu_pjrt_client.cc:1039] Using BFC allocator.
I0000 00:00:1763777614.859882    7546 gpu_helpers.cc:136] XLA backend allocating 6442057728 bytes on device 0 for BFCAllocator.
I0000 00:00:1763777614.860021    7546 gpu_helpers.cc:177] XLA backend will use up to 2147352576 bytes on device 0 for CollectiveBFCAllocator.
W0000 00:00:1763777614.860430    7546 cuda_executor.cc:1802] GPU interconnect information not available: INTERNAL: NVML doesn't support extracting fabric info or NVLink is not used by the device.
I0000 00:00:1763777614.867420    7546 cuda_dnn.cc:463] Loaded cuDNN version 91400

[XLin0mu]

May Reactant.jl support with native windows in future?

[wsmoses]

it's definitely possible, but honestly not high on the priority list right now since we have cuda + WSL working, and CPU native windows working