Remove NVIDIA-only GPU filtering, detect all working CUDA/ROCm GPUs

- Removed brand-specific filtering that only allowed NVIDIA GPUs
- Now detects any GPU that can actually perform CUDA or ROCm operations
- Functional test determines if GPU should be included, not brand
- GPUs are shown with correct system indices (Device 0, 1, etc.)
- AMD GPUs that support ROCm will be shown if functional
- CUDA GPUs from any vendor will be shown if functional

vidai/compat.py

@@ -265,14 +265,7 @@ def detect_gpu_backends() -> dict:
             working_cuda_devices = 0
             for i in range(torch.cuda.device_count()):
                 try:
-                    # Test if device is actually functional by trying a simple operation
-                    device_name = torch.cuda.get_device_name(i).lower()
-
-                    # Only consider NVIDIA GPUs
-                    if not any(keyword in device_name for keyword in ['nvidia', 'geforce', 'quadro', 'tesla', 'rtx', 'gtx']):
-                        continue
-
-                    # Test device functionality
+                    # Test if device can actually perform CUDA operations
                     try:
                         with torch.cuda.device(i):
                             test_tensor = torch.tensor([1.0], device=f'cuda:{i}')

vidai/web.py

@@ -508,16 +508,11 @@ def api_cluster_nodes():
             all_gpu_memory = []
             total_memory = 0
 
-            # Only count working NVIDIA GPUs with actual VRAM detection
+            # Count all working CUDA GPUs with actual VRAM detection
             if torch.cuda.is_available():
                 for i in range(torch.cuda.device_count()):
                     try:
-                        device_name = torch.cuda.get_device_name(i).lower()
-                        # Only consider NVIDIA GPUs
-                        if not any(keyword in device_name for keyword in ['nvidia', 'geforce', 'quadro', 'tesla', 'rtx', 'gtx']):
-                            continue
-
-                        # Test device functionality
+                        # Test device functionality - if it can perform CUDA operations, include it
                         try:
                             with torch.cuda.device(i):
                                 test_tensor = torch.tensor([1.0], device=f'cuda:{i}')
@@ -531,6 +526,7 @@ def api_cluster_nodes():
                             total_memory += vram_gb
                             total_gpus += 1
                         except:
+                            # Device not functional, skip it
                             continue
                     except:
                         continue
@@ -648,15 +644,10 @@ def detect_local_workers():
                             gpus = 0
 
                             if backend == 'cuda' and torch.cuda.is_available():
-                                # Only count working NVIDIA GPUs
+                                # Count all working CUDA GPUs (any brand that actually works)
                                 for i in range(torch.cuda.device_count()):
                                     try:
-                                        device_name = torch.cuda.get_device_name(i).lower()
-                                        # Only consider NVIDIA GPUs
-                                        if not any(keyword in device_name for keyword in ['nvidia', 'geforce', 'quadro', 'tesla', 'rtx', 'gtx']):
-                                            continue
-
-                                        # Test device functionality
+                                        # Test device functionality - if it can perform CUDA operations, include it
                                         try:
                                             with torch.cuda.device(i):
                                                 test_tensor = torch.tensor([1.0], device=f'cuda:{i}')
@@ -670,6 +661,7 @@ def detect_local_workers():
                                             total_memory += vram_gb
                                             gpus += 1
                                         except:
+                                            # Device not functional, skip it
                                             continue
                                     except:
                                         continue