GTX1060的绝唱

写在前面

GTX 1060 作为一款 2016 年发布的显卡，到 2026 年已经经过了 10 年的历程。NVIDIA 已宣布 Pascal（包括 GTX 1060）在 2025 年 10 月停止 Game Ready 驱动支持(NVIDIA announces end-of-life support for many of its GTX cards — but it’s not all bad news)。但是生活还得继续，我还希望笔记本再战 5 年，因此总结一些驱动和 CUDA 相关。

驱动限制

某日升级系统（Gentoo）后再启动后无法进入桌面系统。startx报错：No devices detected，查看xorg的日志发现：

 98 [   709.244] (II) modeset(G0): using drv /dev/dri/card0
 99 [   709.244] (WW) Falling back to old probe method for modesetting
100 [   709.244] (WW) NVIDIA(0): The NVIDIA NVIDIA GeForce GTX 1060 with Max-Q Design GPU
101 [   709.244] (WW) NVIDIA(0):     installed in this system is supported through the NVIDIA
102 [   709.244] (WW) NVIDIA(0):     580.xx Legacy drivers. Please visit
103 [   709.244] (WW) NVIDIA(0):     http://www.nvidia.com/object/unix.html for more
104 [   709.244] (WW) NVIDIA(0):     information.  The 590.48.01 NVIDIA driver will ignore this
105 [   709.244] (WW) NVIDIA(0):     GPU.  Continuing probe...
106 [   709.244] (WW) Falling back to old probe method for modesetting
107 [   709.244] (EE) No devices detected.
108 [   709.244] (EE)
109 Fatal server error:
110 [   709.244] (EE) no screens found(EE)
111 [   709.244] (EE)

原因：错误安装了驱动，不支持590及以上驱动

sudo vim /etc/portage/package.mask/nvidia-drivers

设置屏蔽

>=x11-drivers/nvidia-drivers-581

重启后正常进入系统

PyTorch限制

Python3.12在使用yolo12的过程中，尽管pip安装没有任何报错，但不能使用GPU进行训练。

CUDA available: True
/home/vv/tmp/py_video/venv/lib/python3.12/site-packages/torch/cuda/**init**.py:283: UserWarning:
    Found GPU0 NVIDIA GeForce GTX 1060 which is of cuda capability 6.1.
    Minimum and Maximum cuda capability supported by this version of PyTorch is
    (7.0) - (12.0)
  warnings.warn(
/home/vv/tmp/py_video/venv/lib/python3.12/site-packages/torch/cuda/**init**.py:304: UserWarning:
    Please install PyTorch with a following CUDA
    configurations: 12.6 following instructions at
    https://pytorch.org/get-started/locally/
  warnings.warn(matched_cuda_warn.format(matched_arches))
/home/vv/tmp/py_video/venv/lib/python3.12/site-packages/torch/cuda/**init**.py:326: UserWarning:
NVIDIA GeForce GTX 1060 with CUDA capability sm_61 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_70 sm_75 sm_80 sm_86 sm_90 sm_100 sm_120.
If you want to use the NVIDIA GeForce GTX 1060 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
  warnings.warn(
Device: NVIDIA GeForce GTX 1060
CUDA runtime: 12.8

主要是因为2024 年底～2025 年开始，PyTorch 官方的新版本已经逐步放弃了对 sm_61 的支持，目前（2026年1月）主流版本最低要求通常是 sm_70 或 sm_75，所以才会看到这样的警告：

NVIDIA GeForce GTX 1060 with CUDA capability sm_61 is not compatible with the current PyTorch installation.

一些显卡与架构如下：

显卡 架构 SM
GTX 750 / 750Ti Maxwell sm_50
GTX 9xx Maxwell sm_52
GTX 10xx Pascal sm_61
GTX 16xx Turing sm_75
RTX 20xx Turing sm_75
RTX 30xx Ampere sm_86
RTX 40xx Ada sm_89

测试脚本：

import torch

print("PyTorch version:", torch.__version__)
print("CUDA available:", torch.cuda.is_available())
print("Device count:", torch.cuda.device_count())
print("Current device:", torch.cuda.get_device_name(0) if torch.cuda.is_available() else "CPU")
print("Supported architectures:", torch.cuda.get_arch_list())

输出如下无报警即成功：

PyTorch version: 2.7.1+cu126
CUDA available: True
Device count: 1
Current device: NVIDIA GeForce GTX 1060
Supported architectures: ['sm_50', 'sm_60', 'sm_70', 'sm_75', 'sm_80', 'sm_86', 'sm_90']

结合GPT，Grok后测出的符合的版本是：

Python3.12在CUDA版本12.6的情况下需要安装

  • torch 2.7.1
  • torchvision 0.22.1
  • torchaudio 2.7.1
pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu126
comment: