diff --git a/vllm/platforms/cuda.py b/vllm/platforms/cuda.py
index bffa113cab89..00bbfec1ef7c 100644
--- a/vllm/platforms/cuda.py
+++ b/vllm/platforms/cuda.py
@@ -4,7 +4,7 @@
 """
 
 import os
-from functools import lru_cache, wraps
+from functools import wraps
 from typing import (TYPE_CHECKING, Callable, List, Optional, Tuple, TypeVar,
                     Union)
 
@@ -284,7 +284,6 @@ def get_device_communicator_cls(cls) -> str:
 class NvmlCudaPlatform(CudaPlatformBase):
 
     @classmethod
-    @lru_cache(maxsize=8)
     @with_nvml_context
     def get_device_capability(cls,
                               device_id: int = 0
@@ -298,7 +297,6 @@ def get_device_capability(cls,
             return None
 
     @classmethod
-    @lru_cache(maxsize=8)
     @with_nvml_context
     def has_device_capability(
         cls,
@@ -311,14 +309,12 @@ def has_device_capability(
             return False
 
     @classmethod
-    @lru_cache(maxsize=8)
     @with_nvml_context
     def get_device_name(cls, device_id: int = 0) -> str:
         physical_device_id = device_id_to_physical_device_id(device_id)
         return cls._get_physical_device_name(physical_device_id)
 
     @classmethod
-    @lru_cache(maxsize=8)
     @with_nvml_context
     def get_device_uuid(cls, device_id: int = 0) -> str:
         physical_device_id = device_id_to_physical_device_id(device_id)
@@ -326,7 +322,6 @@ def get_device_uuid(cls, device_id: int = 0) -> str:
         return pynvml.nvmlDeviceGetUUID(handle)
 
     @classmethod
-    @lru_cache(maxsize=8)
     @with_nvml_context
     def get_device_total_memory(cls, device_id: int = 0) -> int:
         physical_device_id = device_id_to_physical_device_id(device_id)