Adjustments that turned things around.

Want to understand the training process better? Got tips for which model to use and when? This is the place for you


Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for discussing tips and understanding the process involved with Training a Faceswap model.

If you have found a bug are having issues with the Training process not working, then you should post in the Training Support forum.

Please mark any answers that fixed your problems so others can find the solutions.

Locked
User avatar
sergor
Posts: 18
Joined: Wed Jan 15, 2020 7:59 pm
Has thanked: 9 times

Adjustments that turned things around.

Post by sergor »

Hey all,

My specs before finding "lightspeed". :

Gigabyte D450 D3H Motherboard, AMD Ryzen 5 1600, 32 GB ram, MSI GtX 1050ti 4 GB GPU.

Very slow on training as in VERY. 7 iterations pr. minute

Changes : I was on Game Ready drivers (NVIDIA), I switched to Studio Driver (NVIDIA) and did a pip update of tensorflow. (Refer to NVIDIA site as to why Studio Driver is beneficial). These were the only changes made.

Result when training: 154 iterations pr. minute (Original - NOT Lowmem). I can't tell you why as I have software-ocd and ALWAYS have everything updated. HWINFO does not show overheating and fans are on nominal speed.

For whatever reason it has cleared a bottleneck of sorts, Studio Driver has an AI element that connects directly to the core and I believe that is the reason for the "inferno-fast"-training.

Hope it can help someone.

/Sergio

Locked