diff --git a/src/d3d8/d3d8_device.cpp b/src/d3d8/d3d8_device.cpp
index c3ad13965589..6528b636fce8 100644
--- a/src/d3d8/d3d8_device.cpp
+++ b/src/d3d8/d3d8_device.cpp
@@ -90,9 +90,21 @@ namespace dxvk {
         return E_FAIL;
       
       case D3DDEVINFOID_VCACHE:
-        // Docs say response should be S_FALSE, but we'll let D9VK
-        // decide based on the value of supportVCache. D3DX8 calls this.
+        // The query will return D3D_OK on Nvidia and D3DERR_NOTAVAILABLE on AMD/Intel
+        // in D3D9, however in the case of the latter we'll need to return a
+        // zeroed out query result and S_FALSE. This behavior has been observed both
+        // on modern native AMD drivers and D3D8-era native ATI drivers.
         res = GetD3D9()->CreateQuery(d3d9::D3DQUERYTYPE_VCACHE, &pQuery);
+
+        if(FAILED(res)) {
+          if (DevInfoStructSize != sizeof(D3DDEVINFO_VCACHE))
+            return D3DERR_INVALIDCALL;
+
+          D3DDEVINFO_VCACHE vCacheDevInfo = {0};
+          memcpy(pDevInfoStruct, &vCacheDevInfo, DevInfoStructSize);
+          return S_FALSE;
+        }
+
         break;
       case D3DDEVINFOID_RESOURCEMANAGER:
         // May not be implemented by D9VK.
diff --git a/src/d3d8/d3d8_device.h b/src/d3d8/d3d8_device.h
index 33ea2bbc1889..22035e61a096 100644
--- a/src/d3d8/d3d8_device.h
+++ b/src/d3d8/d3d8_device.h
@@ -18,6 +18,13 @@
 
 namespace dxvk {
 
+  typedef struct D3DDEVINFO_VCACHE {
+    DWORD         Pattern;
+    DWORD         OptMethod;
+    DWORD         CacheSize;
+    DWORD         MagicNumber;
+  } D3DDEVINFO_VCACHE;
+
   class D3D8Interface;
 
   struct D3D8VertexShaderInfo;