Merge #1835 #1836

1835: Document disabling GPUs r=DhairyaLGandhi a=DhairyaLGandhi From the discussion in #1834 1836: Try using latest cu(DNN) binaries r=DhairyaLGandhi a=ToucheSir Possible alternative to #1822. Co-authored-by: Dhairya Gandhi <[email protected]> Co-authored-by: Brian Chen <[email protected]>
FluxML · Jan 15, 2022 · 9c789dc · 9c789dc
3 parents 79dbbd6 + 8608107 + 6e3ca67
commit 9c789dc
Show file tree

Hide file tree

Showing 3 changed files with 19 additions and 5 deletions.
diff --git a/.buildkite/pipeline.yml b/.buildkite/pipeline.yml
@@ -17,6 +17,8 @@ steps:
     agents:
       queue: "juliagpu"
       cuda: "*"
+    env:
+      JULIA_CUDA_USE_BINARYBUILDER: "true"
     timeout_in_minutes: 60
 
   # - label: "GPU nightly"

diff --git a/docs/src/gpu.md b/docs/src/gpu.md
@@ -85,3 +85,19 @@ julia> x |> cpu
  ⋮
  0.192538
 ```
+
+## Disabling CUDA or choosing which GPUs are visible to Flux
+
+Sometimes it is required to control which GPUs are visible to `julia` on a system with multiple GPUs or disable GPUs entirely. This can be achieved with an environment variable `CUDA_VISIBLE_DEVICES`.
+
+```
+$ export CUDA_VISIBLE_DEVICES=''
+
+# or
+
+$ export CUDA_VISIBLE_DEVICES='0,1'
+```
+
+The environment variable `CUDA_VISIBLE_DEVICES` can also take a comma separated list of device IDs (`0,1,...`) which would be visible to CUDA related processes.
+
+More information for conditional use of GPUs in CUDA.jl can be found in its [documentation](https://cuda.juliagpu.org/stable/installation/conditional/#Conditional-use), and information about the specific use of the variable is described in the [Nvidia CUDA blogpost](https://developer.nvidia.com/blog/cuda-pro-tip-control-gpu-visibility-cuda_visible_devices/).
diff --git a/test/cuda/layers.jl b/test/cuda/layers.jl
@@ -45,11 +45,7 @@ function gpu_gradtest(name::String, layers::Vector, x_cpu = nothing, args...; te
 
           # test
           if test_cpu
-            if VERSION >= v"1.7" && layer === GroupedConvTranspose && args[end] == selu
-              @test_broken y_gpu ≈ y_cpu rtol=1f-3 atol=1f-3
-            else
-              @test y_gpu ≈ y_cpu rtol=1f-3 atol=1f-3
-            end
+            @test y_gpu ≈ y_cpu rtol=1f-3 atol=1f-3
             if isnothing(xg_cpu)
               @test isnothing(xg_gpu)
             else