diff --git a/tests/integration/defs/accuracy/test_llm_api_pytorch.py b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
index 2cff5f6af67..08ab7dbc83e 100644
--- a/tests/integration/defs/accuracy/test_llm_api_pytorch.py
+++ b/tests/integration/defs/accuracy/test_llm_api_pytorch.py
@@ -2024,7 +2024,6 @@ def test_bf16(self, tp_size, pp_size, ep_size, attention_dp, cuda_graph,
             task = MMLU(self.MODEL_NAME)
             task.evaluate(llm)
 
-    @pytest.mark.skip_less_device_memory(140000)  ## OOM on 80G H100
     @parametrize_with_ids("eagle3_one_model", [True, False])
     @parametrize_with_ids("enable_chunked_prefill", [False, True])
     def test_eagle3(self, enable_chunked_prefill, eagle3_one_model):
@@ -2032,7 +2031,10 @@ def test_eagle3(self, enable_chunked_prefill, eagle3_one_model):
             disable_overlap_scheduler=True,
             cuda_graph_config=CudaGraphConfig(batch_sizes=[1]),
         )
-        kv_cache_config = KvCacheConfig(enable_block_reuse=False)
+        kv_cache_config = KvCacheConfig(
+            enable_block_reuse=False,
+            free_gpu_memory_fraction=0.6,
+        )
 
         eagle_model_dir = f"{llm_models_root()}/Qwen3/qwen3_8b_eagle3"
         target_model_dir = f"{llm_models_root()}/Qwen3/Qwen3-8B"