diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py
index 55936cf0986a..10e6070b42c7 100644
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -21,7 +21,7 @@ class EngineArgs:
     tensor_parallel_size: int = 1
     block_size: int = 16
     swap_space: int = 4  # GiB
-    gpu_memory_utilization: float = 0.95
+    gpu_memory_utilization: float = 0.90
     max_num_batched_tokens: int = 2560
     max_num_seqs: int = 256
     disable_log_stats: bool = False