diff --git a/benchmarks/kernels/benchmark_w8a8_block_fp8.py b/benchmarks/kernels/benchmark_w8a8_block_fp8.py
index c6c8e0b0b936..602fad181074 100644
--- a/benchmarks/kernels/benchmark_w8a8_block_fp8.py
+++ b/benchmarks/kernels/benchmark_w8a8_block_fp8.py
@@ -14,7 +14,7 @@
 from tqdm import tqdm
 
 from vllm.model_executor.layers.quantization.utils.fp8_utils import (
-    _w8a8_block_fp8_matmul,
+    _w8a8_triton_block_scaled_mm,
 )
 from vllm.platforms import current_platform
 from vllm.triton_utils import triton
@@ -83,7 +83,7 @@ def grid(META):
         )
 
     if A.dtype == torch.float8_e4m3fn:
-        kernel = _w8a8_block_fp8_matmul
+        kernel = _w8a8_triton_block_scaled_mm
     else:
         raise RuntimeError("Currently, only support tune w8a8 block fp8 kernel.")