diff --git a/vllm/engine/async_llm_engine.py b/vllm/engine/async_llm_engine.py
index 7dcd2eb6..4afb96ec 100644
--- a/vllm/engine/async_llm_engine.py
+++ b/vllm/engine/async_llm_engine.py
@@ -301,7 +301,16 @@ class AsyncLLMEngine:
         elif self.worker_use_ray:
             engine_class = ray.remote(num_cpus=0)(self._engine_class).remote
         else:
-            engine_class = ray.remote(num_gpus=1)(self._engine_class).remote
+            # FIXME(woosuk): This is a bit hacky. Be careful when changing the
+            # order of the arguments.
+            cache_config = args[1]
+            parallel_config = args[2]
+            if parallel_config.tensor_parallel_size == 1:
+                num_gpus = cache_config.gpu_memory_utilization
+            else:
+                num_gpus = 1
+            engine_class = ray.remote(num_gpus=num_gpus)(
+                self._engine_class).remote
         return engine_class(*args, **kwargs)
 
     async def engine_step(self) -> bool:
diff --git a/vllm/engine/llm_engine.py b/vllm/engine/llm_engine.py
index 2400dd53..a1acdfde 100644
--- a/vllm/engine/llm_engine.py
+++ b/vllm/engine/llm_engine.py
@@ -159,9 +159,13 @@ class LLMEngine:
         for bundle in placement_group.bundle_specs:
             if not bundle.get("GPU", 0):
                 continue
+            if self.parallel_config.tensor_parallel_size == 1:
+                num_gpus = self.cache_config.gpu_memory_utilization
+            else:
+                num_gpus = 1
             worker = ray.remote(
                 num_cpus=0,
-                num_gpus=self.cache_config.gpu_memory_utilization,
+                num_gpus=num_gpus,
                 scheduling_strategy=PlacementGroupSchedulingStrategy(
                     placement_group=placement_group,
                     placement_group_capture_child_tasks=True),