Failed to detect installed CUDA version.

[holmesconan]

Describe the bug
When doing any operation at the first time, all artifacts of CUDA are downloaded included the useless ones.

To Reproduce
The Minimal Working Example (MWE) for this bug:

# After installation
using CuArrays
x = CuArrays.fill(1.0, 10)
# This will trigger the artifacts downloading.
x *= 2

Expected behavior
Only the CUDA package that match the system installed NVidia CUDA Toolkit version should be installed, not all available versions.

Build log
Pkg.build is success.

Environment details (please complete this section)
Details on Julia:

julia> versioninfo()
Julia Version 1.4.0
Commit b8e9a9ecc6 (2020-03-21 16:36 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Xeon(R) CPU E3-1230 V2 @ 3.30GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-8.0.1 (ORCJIT, ivybridge)

Julia packages:

• CuArrays.jl:
• CUDAnative.jl:
• Flux.jl

CUDA: 10.0/440.44

[maleadt]

all artifacts of CUDA are downloaded included the useless ones.

That's not true.

Only the CUDA package that match the system installed NVidia CUDA Toolkit version should be installed, not all available versions.

The CUDA toolkit itself is downloaded, matching the installed CUDA driver. If you want to use your own toolkit, set JULIA_CUDA_USE_BINARYBUILDER to false.