problem desctiption The same code only creates Python’s shared memory. Compared with the old system, the performance of the new system is much worse. Why is this? more infos system infos jp4.6.2 [image] jp5.0.2 [image] Experimental infos jp5.0.2 [image] [image] jp 4.6 [image] [image]

Hi, Thanks for your patience. There is no change in kernel or memory between r35.1 and r32.7.1 that can cause this regression. Due to the limited resources, our internal team won’t check further for this issue. Thanks.

Hi, What is the underlying memory the API used? Is it a unified memory? Thanks.

Only python’s shared memory is used. multiprocessing.shared_memory — Shared memory for direct access across processes — Python 3.12.0 documentation Here is the relevant code from multiprocessing import shared_memory #### something else # .... # .... @dataclass class ShmNpArray: def __init__…

Hi, Could you check if the same issue also occurs on JetPack 6? Thanks.

Sorry, I can’t. The machine I use is an industrial control board, and its supplier cannot provide jp6

Hi, Would you mind sharing a runnable source so we can try if the same issue occurs in our environment? Thanks.

use python3.9. test_script.py import numpy as np import logging import time from loguru import logger from shared_memory_numpy import ShmNpArray, ShmNpArrayDisposer from contextlib import contextmanager @contextmanager def timethis(logger: logging.Logger, message: str = None) -> None: start_t…

Hi, Confirmed that we can reproduce the same issue internally. We now are discussing this with our internal team. Will share more info with you later. Thanks.

Python's shared memory performs better on jp4.6.2 than on jp5.0.2

Robotics & Edge Computing Jetson Systems Jetson Xavier NX

AastaLLL December 13, 2023, 9:19am 9

Hi,

Thanks for the sample.
We will check it internally and share more with you later.

Topic		Replies	Views
Shared memory problem CUDA Programming and Performance	10	4119	April 20, 2010
why is shared memory example not faster CUDA Programming and Performance	7	1410	May 16, 2012
Shared memory parallel processing for jetson inference Jetson Nano jetson-inference	4	1492	August 17, 2022
NVMM and IPC/Shared Memory Jetson AGX Xavier decoder	2	1235	October 18, 2021
Whether NVMM-buffers can be used across processes through shared memory like CPU-buffers? Jetson Xavier NX cuda , gstreamer	2	727	March 28, 2022
shared memory in cuda fortran and increasing time of process Legacy PGI Compilers	1	2079	September 2, 2016
GPU memory leaks using shareable handles Jetson AGX Orin cuda , nvbugs	20	820	September 17, 2025
Problem about time of copy data through shared memory CUDA Programming and Performance	4	420	March 30, 2024
Extremely slow memcpy to mmap shared region DeepStream SDK	8	587	September 26, 2023
Some confusion on using shared memory. CUDA Programming and Performance	26	9375	June 2, 2009

Python's shared memory performs better on jp4.6.2 than on jp5.0.2

Related topics