CUDA版本、显卡算力、Pytorch对应关系

Chenfei Ye
Jun 11, 2022 1 min read 踩坑

背景

试用合作医院提供的堡垒机服务器，自带的NVIDIA-A6000显卡不支持结构像自动分割的pipeline镜像，运行时报错capability sm_86 is not compatible。发现是由于显卡的架构比较新，旧版pytorch库不支持。同时根据输出可以看到 The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70 sm_75当前pytorch只能支持上面几种（显卡算力）架构。

解决方案

把基础镜像直接更新到pytorch/pytorch:1.7.0-cuda11.0-cudnn8-runtime，解决。 NOTE: CUDA10.x最高支持算力7.x; CUDA11.0最高支持算力8.x

png

png

pipeline AI
Chenfei Ye
Chenfei Ye
A connectome weaver

The brain’s billions of neurons resemble trees of many species and come in many fantastic shapes. Only the most determined explorers can hope to capture a glimpse of this forest’s interior, and even they see little, and see it poorly.