diff --git a/api/common/launch.py b/api/common/launch.py
index 623de5caf6..cdab268dac 100644
--- a/api/common/launch.py
+++ b/api/common/launch.py
@@ -23,14 +23,28 @@
 
 
 def is_ampere_gpu():
-    stdout, exit_code = system.run_command("nvidia-smi -L")
-    if exit_code == 0:
-        gpu_list = stdout.split("\n")
-        if len(gpu_list) >= 1:
+    try:
+        from pynvml import nvmlInit
+        from pynvml import nvmlDeviceGetHandleByIndex
+        from pynvml import nvmlDeviceGetCudaComputeCapability
+        nvmlInit()
+        handle = nvmlDeviceGetHandleByIndex(0)
+        cc_major,cc_minor = nvmlDeviceGetCudaComputeCapability(handle)
+        #print(str(cc_major)+"."+str(cc_minor))
+        # 8.0 or 8.6
+        return str(cc_major)+"."+str(cc_minor)>"8.0"
+    except ImportError:
+        print("Warning: pynvml package is not installed, please install it as follow \"pip install pynvml\"")
+        stdout, exit_code = system.run_command("nvidia-smi -L")
+        if exit_code == 0:
+            gpu_list = stdout.split("\n")
             #print(gpu_list[0])
             # GPU 0: NVIDIA A100-SXM4-40GB (UUID: xxxx)
-            return gpu_list[0].find("A100") > 0
-    return False
+        else:
+            print("Error: Failed to run sys command \"nvidia-smi -L\"")
+            return False
+        return gpu_list[0].find("A100") > 0
+
 
 
 class NvprofRunner(object):