Poor achieved settings with a rtx 3090

[petabyteHD]

Sorry for my English... Hi, i have tried the two last versions available (10_20_2021 and 11_20_2021) (dx12 version doesn't work for me) and i can't get more than 128 of resolution and 6 for batch size, (all rest of settings by default). I think it's a very poor setup for a rtx 3090 with 24gb vram. I remember to train a few months with 256 res with a older version (that i can't find anymore). I have 32gb system menory ryzen 5900x and w11 with latest drivers. I only installed th DLF package, nothing more,. Seems like the lsatr version only uses about 8/10 gb vram.....

[azhangliang]

信息已收到。

[gxf888]

这是来自QQ邮箱的假期自动回复邮件。   您好，我最近正在休假中，无法亲自回复您的邮件。我将在假期结束后，尽快给您回复。

[MalachiMassey]

I'm having similar questionable results. I have a 12GB NVIDIA GeForce 3060 RTX and a 32GB CPU, but I can't run any model past 128 resolution, regardless of batch size and other settings. This is super annoying since I can't run pretrained models as they are generally around 256 resolution. The "Deepfake native resolution progress" image on the main GitHub page boasts a 448 resolution model trained on an 11GB card. How can I improve model resolution and not run into OOM errors (I've tried putting the model on CPU and reducing the batch size down as low as 6 [12 works for 128 resolution model] to no avail.)?

[joolstorrentecalo]

Did you ever find the answer? If so, would you mind sharing it and closing this issue?