Phaze-A, 2080Ti

[rolnor]

If I want best resolution/result with the Phaze-A model, is the 2080Ti a good choice? Is it much better/faster than 1080Ti?
2080Ti is often seen with 11GB Vram, is that sufficient?
I dont want to compromise on the result and I can put in some money, I can also sell my 1080Ti and upgrade.

[rolnor]

I tried Phaze-A with the 1024 presets and 2048 output res. its running fine with my 1080Ti at batchsize 70. It seems fantastic! I can already see start of a good model at 1750iter.

[torzdf]

Glad to hear that works. Will be interested in the results. I've never had the patience to train up a 1024px model.

On the 2080Ti, I wouldn't buy this card now. 40xx series is just around the corner, and I expect that the 2080Ti's 11GB will look tiny in comparison to what will be available.

[MaxHunter]

I absolutely agree with Torz, wait until the 4000 series - and I'd even wait until the 4080ti, or if money is a consideration, you can get a really really good deal on a 3080ti or 3090. I have a 3080ti, and my finger is really itching at buying an additional 3090 because of the price!

[rolnor]

Thanx, I see no reason to upgrade from 1080Ti, the training is fast and I can run 1-2weeks, no problem. I can see on ebay that 1080Ti-cards are cheap now, is that because they are out-dated or maybe bitcoin-mining is less popular now?

[torzdf]

They are 3 generations old now, and the GPU shortage is over, so they are no longer really worth a lot.

[rolnor]

OK, You can get a pre-owned 2080Ti card for 250-300USD so making deepfakes, even at 1024 pxs is not expensive... My model is improving fast, I will post a link when its ready.

[martinf]

I tried Phaze-A with the 1024 presets and 2048 output res. its running fine with my 1080Ti at batchsize 70. It seems fantastic! I can already see start of a good model at 1750iter.

Impossible.

[ianstephens]

I tried Phaze-A with the 1024 presets and 2048 output res. its running fine with my 1080Ti at batchsize 70. It seems fantastic! I can already see start of a good model at 1750iter.

Impossible.

My thoughts too. Even with 24GB VRAM we only manage BS 2-3 with the DNY 1024 model. And seeing results from only 1750 iterations? Perhaps they are training on a single image for A/B Something's amiss here.

[Scrapemist]

I just started running dny 1024 on 48gb with Full fc, Split fc, Split Gblock and Split Decoders at bs 12.
Little stunned myself tbh.