use gfx and num_CU for device name

vllm/platforms/rocm.py

@@ -161,14 +161,11 @@ def is_full_nvlink(physical_device_ids: List[int]) -> bool:
     def get_device_name(cls, device_id: int = 0) -> str:
         physical_device_id = device_id_to_physical_device_id(device_id)
         handle = amdsmi_get_processor_handles()[physical_device_id]
-        market_name = amdsmi_get_gpu_asic_info(handle)["market_name"]
-        # Note: this may not be exactly the same as the torch device name
-        # E.g. `AMD Instinct MI300X OAM` vs `AMD Instinct MI300X`
-        if "MI308" in market_name:
-            return "AMD Instinct MI308X"
-        if "MI300" in market_name:
-            return "AMD Instinct MI300X"
-        return market_name
+        gpu_info = amdsmi_get_gpu_asic_info(handle)
+        # Using num_cu to distinguish mi300 and mi308
+        num_cu = gpu_info["num_compute_units"]
+        gpu_target_name = gpu_info["target_graphics_version"]
+        return f"AMD_{gpu_target_name}_{num_cu}CU"
 
     @classmethod
     def get_device_total_memory(cls, device_id: int = 0) -> int: