diff --git a/zeus/util/gpu.py b/zeus/device/gpu.py
similarity index 52%
rename from zeus/util/gpu.py
rename to zeus/device/gpu.py
index 4bba19a0..799b6a13 100644
--- a/zeus/util/gpu.py
+++ b/zeus/device/gpu.py
@@ -25,7 +25,14 @@
 
 """
 
+from __future__ import annotations
+
 import abc
+import pynvml
+import os
+
+
+from amdsmi import amdsmi_init
 
 class GPU(abc.ABC):
     def __init__(self, gpu_index: int) -> None:
@@ -39,20 +46,16 @@ def get_total_energy_consumption(self) -> float:
     def set_power_limit(self, value: int) -> None:
         pass
 
-class NVIDIALocalGPU(GPU):
-    def __init__(self, gpu_index: int) -> None:
-        super().__init__(gpu_index)
 
-    def get_total_energy_consumption(self) -> float:
-        # Call NVML
-        pass
+""" NVIDIA GPUs """
+
+class NativeNVIDIAGPU(GPU):
+    """Alternatve names: DirectNVIDIAGPU, LocalNVIDIAGPU, NativeNVIDIAGPU, IntegratedNVIDIAGPU"""
+   
 
-    def set_power_limit(self, value: int) -> None:
-        # Call NVML
-        pass
 
 
-class NVIDIARemoteGPU(GPU):
+class RemoteNVIDIAGPU(NativeNVIDIAGPU):
     def __init__(self, gpu_index: int, server_address: str) -> None:
         super().__init__(gpu_index)
         self.server_address = server_address
@@ -60,11 +63,76 @@ def __init__(self, gpu_index: int, server_address: str) -> None:
     def set_power_limit(self, value: int) -> None:
         # Call server since SYS_ADMIN is required
         pass
+    def set_freq_mem(self, value: int) -> None:
+        # Call server since SYS_ADMIN is required
+        pass
+    def set_freq_core(self, value: int) -> None:
+        # Call server since SYS_ADMIN is required
+        pass
+
+
+""" AMD GPUs """
+
+class NativeAMDGPU(GPU):
+    pass
+
+class RemoteAMDGPU(NativeAMDGPU):
+    pass
+
+
+
+# Managing all GPUs - Base Abstract Class
+class GPUs(abc.ABC):
+    # Abstract class for managing all GPUs
+    @abc.abstractmethod
+    def __init__(self) -> None:
+        pass
+
+    @abc.abstractmethod
+    def Init(self) -> None:
+        pass
+
+    @abc.abstractmethod
+    def Shutdown(self) -> None:
+        pass
+
+
+# ?: Enfore singleton pattern?
+
+# NVIDIA GPUs
+class NVIDIAGPUs(GPUs):
+    def __init__(self, ensure_homogeneuous: bool = True) -> None:
+        pynvml.nvmlInit()
+
+    def __del__(self) -> None:
+        pynvml.nvmlShutdown()
+
+
+#AMD GPUs
+
+class AMDGPUs(GPUs):
+    def __init__(self, ensure_homogeneuous: bool = True) -> None:
+        amdsmi_init()
+
+    def Init(self) -> None:
+        pass
+
+    def Shutdown(self) -> None:
+        pass
+
+    def get_gpu_count(self) -> int:
+        pass
+
+    def get_total_energy_consumption(self) -> float:
+        pass
+
+    def set_power_limit(self, value: int) -> None:
+        pass
+
+
+# Ensure only one instance of GPUs is created
+def get_gpus() -> GPUs:
+    return NVIDIAGPUs() if pynvml.is_initialized() else AMDGPUs()
 
+GPUManager = get_gpus()
 
-# Factory function to get GPU
-def get_gpu(gpu_index: int, server_address: str | None = None) -> GPU:
-    if server_address is None:
-        return NVIDIALocalGPU(gpu_index)
-    else:
-        return NVIDIARemoteGPU(gpu_index, server_address)
\ No newline at end of file