diff --git a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py
index 9bca2834d..2448210be 100644
--- a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py
+++ b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py
@@ -35,8 +35,8 @@ class CompressedTensorsW8A8Int8(CompressedTensorsScheme):
 
     @classmethod
     def get_min_capability(cls) -> int:
-        # lovelace and up
-        return 89
+        # ampere and up
+        return 80
 
     def process_weights_after_loading(self, layer) -> None:
         # If per tensor, when we have a fused module (e.g. QKV) with per