[windows] Improve GPU detection (#9300)

Co-authored-by: Mehrdad <noreply@github.com>

python/ray/resource_spec.py

@@ -3,6 +3,8 @@ from collections import namedtuple
 import logging
 import multiprocessing
 import os
+import subprocess
+import sys
 
 import ray
 import ray.ray_constants as ray_constants
@@ -229,12 +231,23 @@ class ResourceSpec(
 def _autodetect_num_gpus():
     """Attempt to detect the number of GPUs on this machine.
 
-    TODO(rkn): This currently assumes Nvidia GPUs and Linux.
+    TODO(rkn): This currently assumes NVIDIA GPUs on Linux.
+    TODO(mehrdadn): This currently does not work on macOS.
+    TODO(mehrdadn): Use a better mechanism for Windows.
+
+    Possibly useful: tensorflow.config.list_physical_devices()
 
     Returns:
         The number of GPUs if any were detected, otherwise 0.
     """
-    proc_gpus_path = "/proc/driver/nvidia/gpus"
-    if os.path.isdir(proc_gpus_path):
-        return len(os.listdir(proc_gpus_path))
-    return 0
+    result = 0
+    if sys.platform.startswith("linux"):
+        proc_gpus_path = "/proc/driver/nvidia/gpus"
+        if os.path.isdir(proc_gpus_path):
+            result = len(os.listdir(proc_gpus_path))
+    elif sys.platform == "win32":
+        props = "AdapterCompatibility"
+        cmdargs = ["WMIC", "PATH", "Win32_VideoController", "GET", props]
+        lines = subprocess.check_output(cmdargs).splitlines()[1:]
+        result = len([l.rstrip() for l in lines if l.startswith(b"NVIDIA")])
+    return result