diff --git a/tinygrad/lazy.py b/tinygrad/lazy.py
index 5da21069b3..d5b994f844 100644
--- a/tinygrad/lazy.py
+++ b/tinygrad/lazy.py
@@ -6,7 +6,7 @@ from tinygrad.helpers import prod, getenv, all_int, all_same
 from tinygrad.ops import LoadOps, UnaryOps, BinaryOps, TernaryOps, ReduceOps, Op, exec_alu, python_alu
 from tinygrad.shape.symbolic import sint
 from tinygrad.shape.shapetracker import ShapeTracker
-from tinygrad.device import Buffer
+from tinygrad.buffer import Buffer
 from weakref import ref, ReferenceType, WeakValueDictionary
 
 lazycache: WeakValueDictionary[Any, LazyBuffer] = WeakValueDictionary()
diff --git a/tinygrad/runtime/graph/hsa.py b/tinygrad/runtime/graph/hsa.py
index c5cc90f32d..5161b9100e 100644
--- a/tinygrad/runtime/graph/hsa.py
+++ b/tinygrad/runtime/graph/hsa.py
@@ -1,7 +1,8 @@
 import ctypes, collections, time, itertools
 from typing import List, Any, Dict, cast, Optional, Union, Tuple
 from tinygrad.helpers import GraphException, init_c_var, round_up
-from tinygrad.device import Compiled, Buffer, BufferOptions, CompiledASTRunner, BufferXfer, MultiDeviceJITGraph, update_stats, Device
+from tinygrad.buffer import Buffer, BufferOptions
+from tinygrad.device import Compiled, CompiledASTRunner, BufferXfer, MultiDeviceJITGraph, update_stats, Device
 from tinygrad.shape.symbolic import Variable
 from tinygrad.runtime.ops_hsa import HSADevice, PROFILE, Profiler
 from tinygrad.engine.jit import JitItem, get_input_replace, get_jit_stats, \
diff --git a/tinygrad/runtime/ops_cuda.py b/tinygrad/runtime/ops_cuda.py
index ed6eb3d388..f1016582ef 100644
--- a/tinygrad/runtime/ops_cuda.py
+++ b/tinygrad/runtime/ops_cuda.py
@@ -4,7 +4,8 @@ from pathlib import Path
 from typing import Tuple, Optional, List
 import tinygrad.runtime.autogen.cuda as cuda
 from tinygrad.helpers import DEBUG, getenv, from_mv, to_char_p_p, init_c_var, init_c_struct_t, colored, cpu_time_execution
-from tinygrad.device import Compiled, LRUAllocator, MallocAllocator, Compiler, BufferOptions, CompilerOptions
+from tinygrad.device import Compiled, LRUAllocator, MallocAllocator, Compiler, CompilerOptions
+from tinygrad.buffer import BufferOptions
 from tinygrad.renderer.cstyle import CUDARenderer
 from tinygrad.renderer.assembly import PTXRenderer
 if getenv("IOCTL"): import extra.nv_gpu_driver.nv_ioctl  # noqa: F401
diff --git a/tinygrad/runtime/ops_gpu.py b/tinygrad/runtime/ops_gpu.py
index f0449990db..da6332d039 100644
--- a/tinygrad/runtime/ops_gpu.py
+++ b/tinygrad/runtime/ops_gpu.py
@@ -4,7 +4,8 @@ import ctypes, functools, hashlib
 import tinygrad.runtime.autogen.opencl as cl
 from tinygrad.helpers import init_c_var, to_char_p_p, from_mv, OSX, DEBUG
 from tinygrad.renderer.cstyle import OpenCLRenderer
-from tinygrad.device import Compiled, LRUAllocator, BufferOptions, Compiler, CompilerOptions
+from tinygrad.buffer import BufferOptions
+from tinygrad.device import Compiled, LRUAllocator, Compiler, CompilerOptions
 
 # see test/external/external_osx_profiling.py to determine this ratio. it's in like GPU clocks or something
 OSX_TIMING_RATIO = (125/3) if OSX else 1.0
diff --git a/tinygrad/runtime/ops_hsa.py b/tinygrad/runtime/ops_hsa.py
index 5f95236291..60e97c6e0a 100644
--- a/tinygrad/runtime/ops_hsa.py
+++ b/tinygrad/runtime/ops_hsa.py
@@ -3,9 +3,10 @@ import ctypes, functools, subprocess, io, atexit, collections, json
 from typing import Tuple, TypeVar, List, Dict, Any
 import tinygrad.runtime.autogen.hsa as hsa
 from tinygrad.helpers import DEBUG, init_c_var, from_mv, round_up, to_mv, init_c_struct_t, getenv
-from tinygrad.device import Compiled, LRUAllocator, BufferOptions, Compiler, CompilerOptions
-from tinygrad.runtime.driver.hsa import check, scan_agents, find_memory_pool, AQLQueue
+from tinygrad.device import Compiled, LRUAllocator, Compiler, CompilerOptions
+from tinygrad.buffer import BufferOptions
 from tinygrad.renderer.cstyle import HIPRenderer
+from tinygrad.runtime.driver.hsa import check, scan_agents, find_memory_pool, AQLQueue
 from tinygrad.runtime.driver.hip_comgr import compile_hip
 if getenv("IOCTL"): import extra.hip_gpu_driver.hip_ioctl  # noqa: F401
 
diff --git a/tinygrad/runtime/ops_kfd.py b/tinygrad/runtime/ops_kfd.py
index 5f66f9794d..79e9ea6151 100644
--- a/tinygrad/runtime/ops_kfd.py
+++ b/tinygrad/runtime/ops_kfd.py
@@ -1,7 +1,8 @@
 from __future__ import annotations
 from typing import Tuple, Any
 import os, fcntl, ctypes, functools, re, pathlib, mmap, struct, errno, io
-from tinygrad.device import Compiled, LRUAllocator, Compiler, BufferOptions, CompilerOptions
+from tinygrad.device import Compiled, LRUAllocator, Compiler, CompilerOptions
+from tinygrad.buffer import BufferOptions
 from tinygrad.helpers import getenv, from_mv, init_c_struct_t, to_mv, round_up
 from tinygrad.renderer.cstyle import HIPRenderer
 from tinygrad.runtime.driver.hip_comgr import compile_hip