第四十六篇-Tesla P40关闭GPU的ECC释放部分显存

环境

系统：CentOS-7
CPU: 14C28T
显卡：Tesla P40 24G
驱动: 515
CUDA: 11.7
cuDNN: 8.9.2.26

开启关闭ECC优缺点​

Tesla系列GPU默认开启了ECC(error correcing code,错误检查和纠正)
开启ECC 提升数据可靠性，可用内存的减少和性能上的损失。
关闭ECC，释放完整显存，​性能优化，数据错误风险增加

Tesla P40 查看关闭之前

查看ECC命令

nvidia-smi -q -d ecc

开启ECC命令

nvidia-smi -i 0 -e 1

Timestamp                                 : Tue Mar 11 20:59:38 2025
Driver Version                            : 535.129.03
CUDA Version                              : 12.2

Attached GPUs                             : 1
GPU 00000000:03:00.0
    ECC Mode
        Current                           : Enabled
        Pending                           : Enabled

|=========================================+======================+======================|
|   0  Tesla P40                      Off | 00000000:03:00.0 Off |                    0 |
| N/A   14C    P8               8W / 250W |      2MiB / 23040MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

Tesla P40 查看关闭之后

关闭ECC命令

nvidia-smi -i 0 -e 0

GPU 00000000:03:00.0
    ECC Mode
        Current                           : Disabled
        Pending                           : Disabled

|=========================================+======================+======================|
|   0  Tesla P40                      Off | 00000000:03:00.0 Off |                  Off |
| N/A   15C    P8               8W / 250W |      2MiB / 24576MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

结论

试了试，显存多出1.5G左右。建议开启ECC，GPU可靠性高