Intel集成显卡如何玩转AI 神器pytorch

月光技术杂谈

已于 2025-03-04 16:52:57 修改

阅读量1.2k

点赞数 48

分类专栏：大模型初探文章标签：人工智能 pytorch python intel集成显卡 Iris Xe conda intel GPU

于 2025-03-04 16:52:23 首次发布

本文链接：https://blog.youkuaiyun.com/seaneer/article/details/146020106

版权

大模型初探专栏收录该内容

23 篇文章

订阅专栏

背景

最近AI太火了，老技术人备感压力，各种恶补知识。手上有一台性能还不错的老笔记本：配置是cpu i7 cpu + Iris Xe集成显卡，能否用来搭建一套GPU环境来进行一些AI模型的训练和学习呢。答案是：yes。

本文先重点关注基于Intel集成显卡 Iris Xe 显卡的Pytorch 开发环境搭建，后续逐步分享一些AI小项目进行实操。欢迎底部扫码关注公众号

Intel GPU 支持（原型）已在 PyTorch* 2.6 中为 Linux 和 Windows 上的 Intel® 客户端 GPU 和 Intel® 数据中心 GPU Max 系列准备就绪，这会将 Intel GPU 和 SYCL* 软件堆栈引入官方 PyTorch 堆栈，并提供一致的用户体验，以拥抱更多的 AI 应用场景。

如下图所示，白色为PyTorch组件，蓝色为intel 针对pytorch的扩展组件。

Introducing the Intel® Extension for PyTorch* for GPUs

其中， XPU代表的是CPU, GPU, FPGA, 和其他加速器。

软件依赖

要在 Intel GPU 上使用 PyTorch，您需要首先安装 Intel GPU 驱动程序。有关安装指南，请访问Intel GPU 驱动程序安装。

Intel GPU 驱动程序足以进行二进制安装，而从源代码构建则需要 Intel GPU 驱动程序和 Intel® Deep Learning Essentials。请参阅Intel GPU 的 PyTorch 安装先决条件以获取更多信息。

另外还需要安装oneapi, oneAPI 工具包为 PyTorch 在 Intel GPU 上运行提供必要的编译器和库。

以下分别介绍。

安装 oneAPI 工具包

下载安装

下载安装程序：访问英特尔 oneAPI 工具包下载页面，根据你的操作系统（如 Windows、Linux）选择合适的版本进行下载。
运行安装程序：下载完成后，运行安装程序，按照向导提示进行安装。在安装过程中，建议选择默认的安装选项，以确保安装所有必要的组件。安装完成后，可能需要重启计算机。

配置环境变量

安装完成 oneAPI 后，需要配置相应的环境变量，以便系统能够找到 oneAPI 的工具和库。

Windows 系统：打开命令提示符或 PowerShell，运行以下命令：

call "D:\Program Files (x86)\Intel\oneAPI\setvars.bat"

路径可能因安装位置不同而有所差异，你可以根据实际安装路径进行调整。

结果如下：

C:\Windows\System32>call "D:\Program Files (x86)\Intel\oneAPI\setvars.bat"
:: initializing oneAPI environment...
   Initializing Visual Studio command-line environment...
:: WARNING: Visual Studio was not found in a standard install location:
     "C:\Program Files\Microsoft Visual Studio\<Year>\<Edition>" or
     "C:\Program Files (x86)\Microsoft Visual Studio\<Year>\<Edition>"
   Set the VS2019INSTALLDIR or VS2022INSTALLDIR environment variable to
   point to your install location and try again. For example, if you
   are using the VS 2022 Build Tools, do the following:
     set "VS2022INSTALLDIR=C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools"
     "C:\Program Files (x86)\Intel\oneAPI\setvars.bat"
   Visual Studio command-line environment was not configured.
:  advisor -- latest
:  compiler -- latest
:  dal -- latest
:  debugger -- latest
:  dev-utilities -- latest
:  dnnl -- latest
:  dpcpp-ct -- latest
:: WARNING: Visual Studio was not found in the standard installation location:
   "C:\Program Files\Microsoft Visual Studio\<Year>\<Edition>" or
   "C:\Program Files (x86)\Microsoft Visual Studio\<Year>\<Edition>"
   Set the VS2017INSTALLDIR or VS2019INSTALLDIR or VS2022INSTALLDIR
   environment variable to point to your install location and try again.
:  dpl -- latest
:  ipp -- latest
:  ippcp -- latest
:  mkl -- latest
:  ocloc -- latest
:  pti -- latest
:  tbb -- latest
:  umf -- latest
:  vtune -- latest
:: oneAPI environment initialized ::

intel显卡驱动安装

下载驱动： Intel® Arc™ & Iris® Xe Graphics - Windows*

支持以下芯片平台

#Platform  

11th Gen Intel® Core™ processor family (Codename Tiger Lake, Rocket Lake, Tiger Lake-H)
12th Gen Intel® Core™ processor family (Codename Alder Lake-S, Alder Lake-H, Alder Lake-P, Alder Lake-U, Alder Lake-HX, Alder Lake-N, Twin Lake)
13th Gen Intel® Core™ processor family (Codename Raptor Lake-S, Raptor Lake-HX, Raptor Lake-H, Raptor Lake-P, Raptor Lake-U)
14th Gen Intel® Core™ processor family (Codename Raptor Lake Refresh)
Intel® Core™ Ultra processor family (Codename Meteor Lake, Lunar Lake, Arrow Lake)
Intel® Iris® Xe Dedicated Graphics family (Codename DG1)
Intel® Arc™ Graphics family (Codename Alchemist, Battlemage)

Intel® Deep Learning Essentials Installation（可选）

如何使用二进制方式安装pytorch, 需要安装Intel GPU 驱动程序，而从源代码构建pytorch则不仅需要安装 Intel GPU 驱动程序和还要安装 Intel® Deep Learning Essentials。

Intel® Deep Learning Essentials安装程序如下：

下载win11安装程序,https://registrationcenter-download.intel.com/akdlm/IRC_NAS/efc86abd-cb77-452e-a03f-a741895b8ece/intel-deep-learning-essentials-2025.0.0.336_offline.exe

创建python conda虚拟环境

(C:\Users\seane\anaconda3) d:\code>conda create -n pytorch251 python=3.12
(C:\Users\seane\anaconda3) d:\code>conda activate pytorch251

(pytorch251) d:\code>
(pytorch251) d:\code>
(pytorch251) d:\code>pip list
Package    Version
---------- -------
pip        25.0
setuptools 75.8.0
wheel      0.45.1

pytorch安装

使用pip安装 torch

以虚拟环境中，执行下命令安装 PyTorch ，建议基于 anaconda3安装 PyTorch。anaconda3安装可参考专栏内相关文章。

(pytorch_env) d:\code>conda install libuv
(pytorch251) d:\code>conda list
# packages in environment at D:\AppGallery\Anaconda3\envs\pytorch251:
#
# Name                    Version                   Build  Channel
bzip2                     1.0.8                h2bbff1b_6    defaults
ca-certificates           2024.12.31           haa95532_0    defaults
expat                     2.6.4                h8ddb27b_0    defaults
libffi                    3.4.4                hd77b12b_1    defaults
libuv                     1.48.0               h827c3e9_0    defaults
openssl                   3.0.15               h827c3e9_0    defaults
pip                       25.0            py312haa95532_0    defaults
python                    3.12.9               h14ffc60_0    defaults
setuptools                75.8.0          py312haa95532_0    defaults
sqlite                    3.45.3               h2bbff1b_0    defaults
tk                        8.6.14               h0416ee5_0    defaults
tzdata                    2025a                h04d1e81_0    defaults
vc                        14.42                haa95532_4    defaults
vs2015_runtime            14.42.34433          he0abc0d_4    defaults
wheel                     0.45.1          py312haa95532_0    defaults
xz                        5.6.4                h4754444_1    defaults
zlib                      1.2.13               h8cc25b3_1    defaults

以下第二种验证成功。


# For Intel® Arc™ B-Series Graphics, use the commands below:
conda install libuv
python -m pip install torch==2.5.1+cxx11.abi torchvision==0.20.1+cxx11.abi torchaudio==2.5.1+cxx11.abi intel-extension-for-pytorch==2.5.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/bmg/us/

# For Intel® Arc™ A-Series Graphics, use the commands below:【验证成功】
conda install libuv
python -m pip install torch==2.5.1+cxx11.abi torchvision==0.20.1+cxx11.abi torchaudio==2.5.1+cxx11.abi intel-extension-for-pytorch==2.5.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/

# For Intel® Core™ Ultra Series 2 Mobile Processors (Arrow Lake-H), use the commands below:
conda install libuv
python -m pip install torch==2.5.1.post0+cxx11.abi torchvision==0.20.1.post0+cxx11.abi torchaudio==2.5.1.post0+cxx11.abi intel-extension-for-pytorch==2.5.10.post0+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/arl/us/

# For Intel® Core™ Ultra Series 2 with Intel® Arc™ Graphics, use the commands below:
conda install libuv
python -m pip install torch==2.5.1+cxx11.abi torchvision==0.20.1+cxx11.abi torchaudio==2.5.1+cxx11.abi intel-extension-for-pytorch==2.5.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/lnl/us/

# For Intel® Core™ Ultra Processors with Intel® Arc™ Graphics (Meteor Lake-H), use the commands below:
conda install libuv
python -m pip install torch==2.5.1+cxx11.abi torchvision==0.20.1+cxx11.abi torchaudio==2.5.1+cxx11.abi intel-extension-for-pytorch==2.5.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/mtl/us/

验证安装

验证xpu是否可用。注：intel 集成显卡GPU区分于常规的英伟达显卡， torch调用接口要使用xpu成员类。


python -c "import torch; import intel_extension_for_pytorch as ipex; print(torch.__version__); print(ipex.__version__); [print(f'[{i}]: {torch.xpu.get_device_properties(i)}') for i in range(torch.xpu.device_count())];"

另外可以编写一个简单的 Python 脚本来验证 PyTorch 是否能够正常使用 Intel GPU进行张量计算。

(pytorch251) d:\code>

(pytorch251) d:\code>python
Python 3.12.9 | packaged by Anaconda, Inc. | (main, Feb  6 2025, 18:49:16) [MSC v.1929 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.__version__)
2.5.1+cxx11.abi
>>>
>>> print(torch.xpu.is_available())
True


>>> device = torch.device("xpu")
>>> x = torch.tensor([1.0, 2.0, 3.0], device=device)
>>> print(x)
tensor([1., 2., 3.], device='xpu:0')
>>> y = torch.tensor([4.0, 5.0, 6.0], device=device)
>>> z = x + y
>>> print(z)
tensor([5., 7., 9.], device='xpu:0')

张量运算正常，说明 PyTorch 已成功安装并能够使用 Intel Iris Xe 显卡进行计算。

常见问题解决

Overriding a previously registered kernel

>>> import intel_extension_for_pytorch
[W222 18:14:14.000000000 OperatorEntry.cpp:162] Warning: Warning only once for all operators,  other operators may also be overridden.
  Overriding a previously registered kernel for the same operator and the same dispatch key
  operator: aten::_cummax_helper(Tensor self, Tensor(a!) values, Tensor(b!) indices, int dim) -> ()
    registered at C:\Jenkins\workspace\IPEX-WW-BUILDS\private-gpu\build\aten\src\ATen\RegisterSchema.cpp:6
  dispatch key: XPU
  previous kernel: registered at C:\Jenkins\workspace\IPEX-WW-BUILDS\private-gpu\build\aten\src\ATen\RegisterCPU.cpp:30476
       new kernel: registered at C:\Jenkins\workspace\IPEX-WW-BUILDS\ipex-gpu\build\Release\csrc\gpu\csrc\aten\generated\ATen\RegisterXPU.cpp:2971 (function operator ())

解决方法是卸载后重新安装：

pip uninstall torch==2.5.1+cxx11.abi torchvision==0.20.1+cxx11.abi torchaudio==2.5.1+cxx11.abi intel-extension-for-pytorch==2.5.10+xpu

Getting requirements to build wheel did not run successfully.

解决：下载Microsoft C++ 生成工具，独立 MSVC 编译器、库和脚本。

安装 Microsoft C++ Build Tools

https://aka.ms/vs/17/release/vs_BuildTools.exe

下载安装

安装选件：

确保勾选「C++桌面开发」和「Windows 10 SDK」

网络问题

不稳定的网络连接可能导致无法正确下载包的依赖项。更换源：

pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/

安装包冲突

外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传

解决：单独安装。

pip install faster-whisper==1.0.1 -i https://pypi.tuna.tsinghua.edu.cn/simple  --trusted-host pypi.tuna.tsinghua.edu.cn

关键似乎是faster-whisper的依赖ctranslate2包，重新安装。

ERROR: Could not find a version that satisfies the requirement ctranslate2<5,>=4.0 (from faster-whisper) (from versions: none)

CTranslate2 is a C++ and Python library for efficient inference with Transformer models.The project implements a custom runtime that applies many performance optimization techniques such as weights quantization, layers fusion, batch reordering, etc., to accelerate and reduce the memory usage of Transformer models on CPU and GPU.
The following model types are currently supported:
Encoder-decoder models: Transformer base/big, M2M-100, NLLB, BART, mBART, Pegasus, T5, Whisper
Decoder-only models: GPT-2, GPT-J, GPT-NeoX, OPT, BLOOM, MPT, Llama, Mistral, Gemma, CodeGen, GPTBigCode, Falcon, Qwen2
Encoder-only models: BERT, DistilBERT, XLM-RoBERTa

TimeoutError: The read operation timed out

pip install -r requirements.txt安装依赖出现以下错误，

解决方法是：

pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/

ERROR: Exception:
Traceback (most recent call last):
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_vendor\urllib3\response.py", line 438, in _error_catcher
    yield
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_vendor\urllib3\response.py", line 561, in read
    data = self._fp_read(amt) if not fp_closed else b""
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_vendor\urllib3\response.py", line 527, in _fp_read
    return self._fp.read(amt) if amt is not None else self._fp.read()
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_vendor\cachecontrol\filewrapper.py", line 98, in read
    data: bytes = self.__fp.read(amt)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\http\client.py", line 466, in read
    s = self.fp.read(amt)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\socket.py", line 717, in readinto
    return self._sock.recv_into(b)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\ssl.py", line 1307, in recv_into
    return self.read(nbytes, buffer)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\ssl.py", line 1163, in read
    return self._sslobj.read(len, buffer)
TimeoutError: The read operation timed out

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\cli\base_command.py", line 106, in _run_wrapper
    status = _inner_run()
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\cli\base_command.py", line 97, in _inner_run
    return self.run(options, args)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\cli\req_command.py", line 67, in wrapper
    return func(self, options, args)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\commands\install.py", line 386, in run
    requirement_set = resolver.resolve(
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\resolution\resolvelib\resolver.py", line 179, in resolve
    self.factory.preparer.prepare_linked_requirements_more(reqs)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\operations\prepare.py", line 554, in prepare_linked_requirements_more
    self._complete_partial_requirements(
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\operations\prepare.py", line 469, in _complete_partial_requirements
    for link, (filepath, _) in batch_download:
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\network\download.py", line 184, in __call__
    for chunk in chunks:
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\cli\progress_bars.py", line 55, in _rich_progress_bar
    for chunk in iterable:
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_internal\network\utils.py", line 65, in response_chunks
    for chunk in response.raw.stream(
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_vendor\urllib3\response.py", line 622, in stream
    data = self.read(amt=amt, decode_content=decode_content)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_vendor\urllib3\response.py", line 560, in read
    with self._error_catcher():
  File "C:\Users\seane\anaconda3\envs\env_name\lib\contextlib.py", line 153, in __exit__
    self.gen.throw(typ, value, traceback)
  File "C:\Users\seane\anaconda3\envs\env_name\lib\site-packages\pip\_vendor\urllib3\response.py", line 443, in _error_catcher
    raise ReadTimeoutError(self._pool, None, "Read timed out.")
pip._vendor.urllib3.exceptions.ReadTimeoutError: HTTPSConnectionPool(host='files.pythonhosted.org', port=443): Read timed out.

ref

在 Intel GPU 上入门 — PyTorch 2.6 文档 - PyTorch 深度学习库

Intel® Extension for PyTorch* Installation Guide

https://blog.youkuaiyun.com/lggirls/article/details/131878810)

(host=‘files.pythonhosted.org’, port=443): Read timed out.




# ref

[在 Intel GPU 上入门 — PyTorch 2.6 文档 - PyTorch 深度学习库](https://pytorch.ac.cn/docs/stable/notes/get_start_xpu.html)

[Intel® Extension for PyTorch* Installation Guide](https://pytorch-extension.intel.com/installation?platform=gpu&version=v2.5.10%2Bxpu&os=windows&package=pip)

https://blog.youkuaiyun.com/lggirls/article/details/131878810)