Hardware
1. Only NVIDIA discrete graphics cards are supported, AMD graphics cards and core graphics cards are not supported.
2. 6G video memory is the entry point, otherwise most models will not be able to carry it.
3. CPU\hard disk\memory is not required.
Software
system selection
supports WIN10, WIN7, LINUX, does not support mac (because mac does not support NVIDIA graphics card)
graphics card driver
graphics card driver Upgrade to the latest
method: Baidu
graphics card hardware acceleration . If you are using an RTX30 series graphics card, you need to upgrade WIN10 to 20H4 or a newer version, and then turn on hardware acceleration. Otherwise, the system
will be stuck in the first iteration during training.
Please Baidu for the upgrade method. 
Virtual memory settings
In the process of loading and reading face data, a large amount of memory (here refers to memory, not video memory) is required. It is likely that the memory is exhausted and the program stops, and the "page file is too small to complete the operation"" memory error is reported. "Wait for error.
It is necessary to set a large virtual memory, and use a part of the hard disk space to temporarily use it as memory.
Setting method Baidu "How to set virtual memory"
DFL software version selection
30 series graphics card can only use the version after December 12, 2020, the earlier version does not support. The latest is the 2021 version 0104
20 series and earlier graphics cards are optional. It is recommended to use the latest version
Note that after the 1212 version, the software download has two installation packages, the one marked "RTX2080ti_and_earlier" is the software installation package for the 20-series and 10-series graphics cards, and the other is the software installation package for the 30-series graphics card. Please do not install it wrong, otherwise the program cannot be used normally.
reference link
deepfacelab Chinese Forum