diff --git a/vectordb_bench/backend/clients/milvus/config.py b/vectordb_bench/backend/clients/milvus/config.py
index 0e93e82b8..eea7b11f5 100644
--- a/vectordb_bench/backend/clients/milvus/config.py
+++ b/vectordb_bench/backend/clients/milvus/config.py
@@ -14,13 +14,17 @@ class MilvusIndexConfig(BaseModel):
 
     index: IndexType
     metric_type: MetricType | None = None
+    
+    @property
+    def is_gpu_index(self) -> bool:
+        return self.index in [IndexType.GPU_CAGRA, IndexType.GPU_IVF_FLAT, IndexType.GPU_IVF_PQ]
 
     def parse_metric(self) -> str:
         if not self.metric_type:
             return ""
 
-        # if self.metric_type == MetricType.COSINE:
-        #     return MetricType.L2.value
+        if self.is_gpu_index and self.metric_type == MetricType.COSINE:
+            return MetricType.L2.value
         return self.metric_type.value
 
 
diff --git a/vectordb_bench/backend/clients/milvus/milvus.py b/vectordb_bench/backend/clients/milvus/milvus.py
index 58334efe9..2436e2680 100644
--- a/vectordb_bench/backend/clients/milvus/milvus.py
+++ b/vectordb_bench/backend/clients/milvus/milvus.py
@@ -8,7 +8,7 @@
 from pymilvus import Collection, utility
 from pymilvus import CollectionSchema, DataType, FieldSchema, MilvusException
 
-from ..api import VectorDB, IndexType
+from ..api import VectorDB
 from .config import MilvusIndexConfig
 
 
@@ -119,7 +119,7 @@ def wait_index():
             wait_index()
 
             # Skip compaction if use GPU indexType
-            if self.case_config.index in [IndexType.GPU_CAGRA, IndexType.GPU_IVF_FLAT, IndexType.GPU_IVF_PQ]:
+            if self.case_config.is_gpu_index:
                 log.debug("skip compaction for gpu index type.")
             else :
                 self.col.compact()
@@ -157,6 +157,10 @@ def optimize(self):
 
     def need_normalize_cosine(self) -> bool:
         """Wheather this database need to normalize dataset to support COSINE"""
+        if self.case_config.is_gpu_index:
+            log.info(f"current gpu_index only supports IP / L2, cosine dataset need normalize.")
+            return True
+        
         return False
 
     def insert_embeddings(