Memory increase of 500MB when calling torch._C.init_xpu()

[piDack]

Describe the bug

I am experiencing an issue with the IPEX (Intel Extension for PyTorch) integration. Whenever I execute the torch._C.init_xpu() function, I notice a consistent increase in memory usage of approximately 500MB. This behavior is reproducible and seems to occur every time the function is called.

Environment:

OS: Windows 11
PyTorch version: 2.13.1
Hardware: Utlra 5 288V
Steps to reproduce:

Install PyTorch and IPEX following the official instructions.
Run the following Python code:

import torch
torch._C.init_xpu()

Monitor the memory usage before and after the function call.
Expected behavior: The memory usage should remain stable or increase only minimally when calling torch._C.init_xpu().

Actual behavior: The memory usage increases by approximately 500MB.

Additional context: I have tried various memory profiling tools to track this issue and it consistently points to the torch._C.init_xpu() call as the source of the memory increase.

Is there a known issue related to this? Are there any workarounds or best practices to prevent such a significant memory allocation when initializing XPU with IPEX?

Thank you for your assistance.

Versions

Collecting environment information...
=====================================
PyTorch version:   2.3.1+cxx11.abi
PyTorch CXX11 ABI: No
IPEX version:      2.3.110+xpu
IPEX commit:       95c945927
Build type:        Release

OS:                Microsoft Windows 11 专业版 (10.0.26100 64 位)
GCC version:       N/A
Clang version:     N/A
IGC version:       N/A
CMake version:     N/A
Libc version:      N/A

Python version:    3.11.10 | packaged by Anaconda, Inc. | (main, Oct  3 2024, 07:22:26) [MSC v.1929 64 bit (AMD64)] (64-bit runtime)
Python platform:   Windows-10-10.0.26100-SP0
Is XPU available:  True
DPCPP runtime:     N/A
MKL version:       N/A

GPU models and configuration onboard:
* Intel(R) Arc(TM) 140V GPU (16GB)

GPU models and configuration detected:
* [0] _XpuDeviceProperties(name='Intel(R) Arc(TM) 140V GPU (16GB)', platform_name='Intel(R) Level-Zero', type='gpu', driver_version='1.3.30513', total_memory=16824MB, max_compute_units=64, gpu_eu_count=64, gpu_subslice_count=8, max_work_group_size=1024, max_num_sub_groups=64, sub_group_sizes=[16 32], has_fp16=1, has_fp64=1, has_atomic64=1)

Driver version:
* 32.0.101.6042 (20240927000000.******+***)

CPU:
Description: Intel64 Family 6 Model 189 Stepping 1
Manufacturer: GenuineIntel
Name: Intel(R) Core(TM) Ultra 9 288V
NumberOfCores: 8
NumberOfEnabledCore: 8
NumberOfLogicalProcessors: 8
ThreadCount: 8

Versions of relevant libraries:
[conda] dpcpp-cpp-rt              2024.2.1                 pypi_0    pypi
[conda] intel-cmplr-lib-rt        2024.2.1                 pypi_0    pypi
[conda] intel-cmplr-lib-ur        2024.2.1                 pypi_0    pypi
[conda] intel-cmplr-lic-rt        2024.2.1                 pypi_0    pypi
[conda] intel-extension-for-pytorch 2.3.110+xpu              pypi_0    pypi
[conda] intel-opencl-rt           2024.2.1                 pypi_0    pypi
[conda] intel-openmp              2024.2.1                 pypi_0    pypi
[conda] intel-sycl-rt             2024.2.1                 pypi_0    pypi
[conda] libuv                     1.48.0               h827c3e9_0
[conda] mkl                       2024.2.1                 pypi_0    pypi
[conda] mkl-dpcpp                 2024.2.1                 pypi_0    pypi
[conda] numpy                     1.26.4                   pypi_0    pypi
[conda] onemkl-sycl-blas          2024.2.1                 pypi_0    pypi
[conda] onemkl-sycl-datafitting   2024.2.1                 pypi_0    pypi
[conda] onemkl-sycl-dft           2024.2.1                 pypi_0    pypi
[conda] onemkl-sycl-lapack        2024.2.1                 pypi_0    pypi
[conda] onemkl-sycl-rng           2024.2.1                 pypi_0    pypi
[conda] onemkl-sycl-sparse        2024.2.1                 pypi_0    pypi
[conda] onemkl-sycl-stats         2024.2.1                 pypi_0    pypi
[conda] onemkl-sycl-vm            2024.2.1                 pypi_0    pypi
[conda] torch                     2.3.1+cxx11.abi          pypi_0    pypi
[conda] torchvision               0.18.1+cxx11.abi          pypi_0    pypi
[conda] transformers              4.37.0                   pypi_0    pypi

[devpramod]

Hi @piDack We are looking into this issue