diff --git a/.circleci/config.yml b/.circleci/config.yml
index 8a09c45e507..bcd0f365bd0 100644
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -9,6 +9,13 @@ version: 2.1
 orbs:
   win: circleci/windows@2.0.0
 
+executors:
+  windows-gpu-prototype:
+    machine:
+      resource_class: windows.gpu.small.prototype
+      image: windows-server-2019-nvidia:201908-28
+      shell: bash.exe
+
 commands:
   checkout_merge:
     description: "checkout merge branch"
@@ -184,6 +191,20 @@ jobs:
             bash packaging/build_conda.sh
           shell: powershell.exe
 
+  binary_win_conda_cuda:
+    <<: *binary_common
+    executor: windows-gpu-prototype
+    steps:
+      - checkout_merge
+      - run:
+          command: |
+            choco install miniconda3
+            (& "C:\tools\miniconda3\Scripts\conda.exe" "shell.powershell" "hook") | Out-String | Invoke-Expression
+            conda activate base
+            conda install -yq conda-build "conda-package-handling!=1.5.0"
+            bash packaging/build_conda.sh
+          shell: powershell.exe
+
   binary_macos_wheel:
     <<: *binary_common
     macos:
@@ -485,6 +506,10 @@ workflows:
           name: torchvision_win_py3.6_cpu
           python_version: "3.6"
           cu_version: "cpu"
+      - binary_win_conda_cuda:
+          name: torchvision_win_py3.6_cu101
+          python_version: "3.6"
+          cu_version: "cu101"
 
   nightly:
     jobs:
diff --git a/.circleci/config.yml.in b/.circleci/config.yml.in
index 16687a4724f..7ed762fe5e5 100644
--- a/.circleci/config.yml.in
+++ b/.circleci/config.yml.in
@@ -9,6 +9,13 @@ version: 2.1
 orbs:
   win: circleci/windows@2.0.0
 
+executors:
+  windows-gpu-prototype:
+    machine:
+      resource_class: windows.gpu.small.prototype
+      image: windows-server-2019-nvidia:201908-28
+      shell: bash.exe
+
 commands:
   checkout_merge:
     description: "checkout merge branch"
@@ -184,6 +191,20 @@ jobs:
             bash packaging/build_conda.sh
           shell: powershell.exe
 
+  binary_win_conda_cuda:
+    <<: *binary_common
+    executor: windows-gpu-prototype
+    steps:
+      - checkout_merge
+      - run:
+          command: |
+            choco install miniconda3
+            (& "C:\tools\miniconda3\Scripts\conda.exe" "shell.powershell" "hook") | Out-String | Invoke-Expression
+            conda activate base
+            conda install -yq conda-build "conda-package-handling!=1.5.0"
+            bash packaging/build_conda.sh
+          shell: powershell.exe
+
   binary_macos_wheel:
     <<: *binary_common
     macos:
@@ -284,6 +305,10 @@ workflows:
           name: torchvision_win_py3.6_cpu
           python_version: "3.6"
           cu_version: "cpu"
+      - binary_win_conda_cuda:
+          name: torchvision_win_py3.6_cu101
+          python_version: "3.6"
+          cu_version: "cu101"
 
   nightly:
 {%- endif %}
diff --git a/packaging/pkg_helpers.bash b/packaging/pkg_helpers.bash
index 594906ddb73..b999894901a 100644
--- a/packaging/pkg_helpers.bash
+++ b/packaging/pkg_helpers.bash
@@ -50,21 +50,33 @@ setup_cuda() {
   # Now work out the CUDA settings
   case "$CU_VERSION" in
     cu101)
-      export CUDA_HOME=/usr/local/cuda-10.1/
+      if [[ "$OSTYPE" == "msys" ]]; then
+        export CUDA_HOME="C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v10.1"
+      else
+        export CUDA_HOME=/usr/local/cuda-10.1/
+      fi
       export FORCE_CUDA=1
       # Hard-coding gencode flags is temporary situation until
       # https://github.com/pytorch/pytorch/pull/23408 lands
       export NVCC_FLAGS="-gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_50,code=compute_50"
       ;;
     cu100)
-      export CUDA_HOME=/usr/local/cuda-10.0/
+      if [[ "$OSTYPE" == "msys" ]]; then
+        export CUDA_HOME="C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v10.0"
+      else
+        export CUDA_HOME=/usr/local/cuda-10.0/
+      fi
       export FORCE_CUDA=1
       # Hard-coding gencode flags is temporary situation until
       # https://github.com/pytorch/pytorch/pull/23408 lands
       export NVCC_FLAGS="-gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_50,code=compute_50"
       ;;
     cu92)
-      export CUDA_HOME=/usr/local/cuda-9.2/
+      if [[ "$OSTYPE" == "msys" ]]; then
+        export CUDA_HOME="C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v9.2"
+      else
+        export CUDA_HOME=/usr/local/cuda-9.2/
+      fi
       export FORCE_CUDA=1
       export NVCC_FLAGS="-gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_50,code=compute_50"
       ;;
diff --git a/packaging/torchvision/bld.bat b/packaging/torchvision/bld.bat
index 870759c67b7..73f217c2cf1 100644
--- a/packaging/torchvision/bld.bat
+++ b/packaging/torchvision/bld.bat
@@ -3,14 +3,14 @@
 set TORCHVISION_BUILD_VERSION=%PKG_VERSION%
 set TORCHVISION_BUILD_NUMBER=%PKG_BUILDNUM%
 
-if not "%CUDA_VERSION%" == "None" (
-    set build_with_cuda=1
-    set desired_cuda=%CUDA_VERSION:~0,-1%.%CUDA_VERSION:~-1,1%
-) else (
-    set build_with_cuda=
-)
+set build_with_cuda=
 
-if "%build_with_cuda%" == "" goto cuda_flags_end
+if "%CUDA_VERSION%" == "None" goto cuda_flags_end
+if "%CUDA_VERSION%" == "cpu" goto cuda_flags_end
+if "%CUDA_VERSION%" == "" goto cuda_flags_end
+
+set build_with_cuda=1
+set desired_cuda=%CUDA_VERSION:~0,-1%.%CUDA_VERSION:~-1,1%
 
 set CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%desired_cuda%
 set CUDA_BIN_PATH=%CUDA_PATH%\bin