diff --git a/clusters/k3s-dgx/nim-service/qwen.yaml b/clusters/k3s-dgx/nim-service/qwen.yaml
index 6e655da..e8b80b0 100644
--- a/clusters/k3s-dgx/nim-service/qwen.yaml
+++ b/clusters/k3s-dgx/nim-service/qwen.yaml
@@ -45,7 +45,7 @@ spec:
     - --served-model-name
     - Qwen/Qwen3.6-27B-FP8
     - --gpu-memory-utilization
-    - "0.90"
+    - "0.85"
     - --max-model-len
     - "256000"
     - --language-model-only