diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index be82930b62..537cc04e6e 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -619,6 +619,44 @@ jobs:
         if: matrix.backend=='amd'
         run: python -m pytest -n=auto test/test_hcq.py test/test_tiny.py --durations=20
 
+  wintests:
+    strategy:
+      fail-fast: false
+      matrix:
+        backend: [llvm]
+
+    name: Tests on Windows (${{ matrix.backend }})
+    runs-on: windows-latest
+    timeout-minutes: 45
+    steps:
+      - name: Checkout Code
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 2 # NOTE: this fetches the HEAD commit of the PR
+      - name: Set up Python 3.12
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.12
+      - name: Cache python packages
+        uses: actions/cache@v4
+        with:
+          path: ${{ env.Python3_ROOT_DIR }}\Lib\site-packages
+          key: windows-${{ matrix.backend }}-packages-${{ hashFiles('**/setup.py') }}
+      - name: Install dependencies
+        run: pip install --user -e '.[testing]' --extra-index-url https://download.pytorch.org/whl/cpu
+      - name: Check Device.DEFAULT and print some source
+        env:
+          DEBUG: 5
+          LLVM: 1
+          PYTHONPATH: ${{ github.workspace }}
+        run: |
+          python3 test/test_ops.py TestOps.test_add
+      - name: Run pytest
+        env:
+          DEBUG: 5
+          LLVM: 1
+        run: python -m pytest -n=auto test/test_tiny.py --durations=20
+
   #testunicorn:
   #  name: ARM64 unicorn Test
   #  runs-on: ubuntu-latest
diff --git a/tinygrad/device.py b/tinygrad/device.py
index 7fea67aa00..04182a3c33 100644
--- a/tinygrad/device.py
+++ b/tinygrad/device.py
@@ -3,7 +3,6 @@ from dataclasses import dataclass, replace
 from collections import defaultdict
 from typing import Optional, Any, Iterator, Generator
 import multiprocessing, importlib, inspect, functools, pathlib, os, ctypes, ctypes.util, platform, contextlib, sys, re, atexit, pickle, decimal, time
-from mmap import mmap, PROT_READ, PROT_WRITE, PROT_EXEC, MAP_ANON, MAP_PRIVATE
 from tinygrad.helpers import CI, OSX, LRU, getenv, diskcache_get, diskcache_put, DEBUG, GlobalCounters, flat_mv, from_mv, PROFILE, temp, mv_address, \
                              cpu_time_execution, colored, Context
 from tinygrad.dtype import DType, ImageDType, PtrDType, dtypes
@@ -221,9 +220,10 @@ MAP_JIT = 0x0800
 
 # CPUProgram is a jit/shellcode program that can be just mmapped and jumped to
 class CPUProgram:
-  helper_handle = ctypes.CDLL(ctypes.util.find_library('System') if OSX else 'libgcc_s.so.1')
-
+  helper_handle = ctypes.CDLL(ctypes.util.find_library('System' if OSX else 'kernel32' if sys.platform == "win32" else 'gcc_s'))
   def __init__(self, name:str, lib:bytes):
+    assert sys.platform != "win32", "clang is not supported for windows yet"
+    from mmap import mmap, PROT_READ, PROT_WRITE, PROT_EXEC, MAP_ANON, MAP_PRIVATE
     # On apple silicon with SPRR enabled (it always is in macos) RWX pages are unrepresentable: https://blog.svenpeter.dev/posts/m1_sprr_gxf/
     # MAP_JIT allows us to easily flip pages from RW- to R-X and vice versa. It is a noop on intel cpus. (man pthread_jit_write_protect_np)
     self.mem = mmap(-1, len(lib), MAP_ANON | MAP_PRIVATE | (MAP_JIT if OSX else 0), PROT_READ | PROT_WRITE | PROT_EXEC)
@@ -329,4 +329,4 @@ if __name__ == "__main__":
         result = f"{colored('FAIL', 'yellow')} {e}"
     except Exception as e:
       result = f"{colored('FAIL', 'red')} {e}"
-    print(f"{'*' if device == Device.DEFAULT else ' '} {device:10s}: {result}")
\ No newline at end of file
+    print(f"{'*' if device == Device.DEFAULT else ' '} {device:10s}: {result}")