diff --git a/openllm_next/accelerator_spec.py b/openllm_next/accelerator_spec.py
index fe1d6632..1168127c 100644
--- a/openllm_next/accelerator_spec.py
+++ b/openllm_next/accelerator_spec.py
@@ -78,6 +78,7 @@ def get_local_machine_spec():
 
     from pynvml import (
         nvmlDeviceGetCount,
+        nvmlDeviceGetCudaComputeCapability,
         nvmlDeviceGetHandleByIndex,
         nvmlDeviceGetMemoryInfo,
         nvmlDeviceGetName,
@@ -98,6 +99,14 @@ def get_local_machine_spec():
                     model=name, memory_size=math.ceil(int(memory_info.total) / 1024**3)
                 )
             )
+            compute_capability = nvmlDeviceGetCudaComputeCapability(handle)
+            if compute_capability < (7, 5):
+                output(
+                    f"GPU {name.decode()} with compute capability {compute_capability} "
+                    "may not be supported, 7.5 or higher is recommended. check "
+                    "https://developer.nvidia.com/cuda-gpus for more information",
+                    style="yellow",
+                )
         nvmlShutdown()
         return DeploymentTarget(
             accelerators=accelerators,