Best GPUs for Stable Cascade and Diffusion - 2024

[@Pixovert]

Stable Cascade course (discount until 25 February) cutt.ly/xwBnvtjV Amazon list - from entry level to the Big Boss geni.us/lDB6v3D - GIGABYTE GeForce RTX 3060 - 12 GB geni.us/fUJUqVf - maxsun GeForce RTX 3060 iCraft OC - 12 GB geni.us/xV4b - ZOTAC Gaming GeForce RTX 4060 Ti 16GB geni.us/hBNKvh - MSI Gaming GeForce RTX 4060 Ti 16GB geni.us/GQAqh - MSI GeForce RTX 4060 Ti Gaming X Slim geni.us/O20aZ - MSI Gaming RTX 4070 Ti Super 16GB Gaming geni.us/LAIGt - MSI GeForce RTX 4080 SUPER 16GB Gaming geni.us/Z10ofh3 - MSI GeForce RTX 4080 SUPER 16GB Gaming X TRIO geni.us/YTuN0 - RTX 4080 Super 16 GB geni.us/8cL42J - RTX 4090 24 GB bit.ly/3wsE5KR - High End GPUs

[@LouisGedo]

👋

[@blender_wiki]

😂😂😂

[@Kratos30000]

Which one would you recommend for comfyui/Animatediff? Thanks for the great video

[@MURSIXX]

is it possible to run SD on dual RTX4080 (for example) configuration?

[@Pixovert]

Probably not, You can do this with some 30 series cards and older professional cards.

[@foreropa]

Hello, is your Udemy Course still working as well, or do you have a new course I could check? Thanks!

[@shabadooshabadoo4918]

i literally just upgraded from my 1080ti to the rtx 4060 ti 16gb like 20 mins before seeing this video.

[@EggKnight]

How are your results? Is it really a good GPU for SD and SDXL?

[@shabadooshabadoo4918]

@EggKnight yeah its about 3 or 4 times faster at rendering ai stuff than my previous card, and i get the benefit of frame generation in games now so games that i might only get 35fps in without it i can get around 60-80fps in now. That said some games their frame generation sucks like Gray Zone Warfare. Other games like New World the frame gen is so good you dont even notice its turned on. Games that dont support it card performance is similar to my old card. One interesting thing thing though is i left my old card in and put the new one in as an experiment and then opened two instances of stable diffusion and configured each one to use a different gpu. So i can render ai images on both cards at once now. I can render the images in the max size which is like 2k by 2k usually without running out of memory but a lot of models just arent trained for that size and you get better results around 1024x1024. But i do it at 1400xwhatever. Like 1400x800 and it usually wont fall apart like maybe 8/10 times the image holds up and makes sense. Higher than that it just duplicates your prompt. Keep in mind this varies completely depending on model.

[@joaobarreira5221]

in the Stable Cascade course do you explain how to finetune the model?

[@Lowai14]

Hi keiven can I ask you about RTX 4070 laptop visrion ( 8GB) I am intending to get Lenovo legion I9 14th gen , is this suitable setup for stable diffusion R3 & cascade? Thanks for your advise

[@JahonCross]

Will my rx 6700xt be enough for this?

[@4XTP5]

Yes

[@baheth3elmy16]

I am using Cascade as A1111 Forge extension with my Nvidia RTX3050 8GB VRAM. It works, slow but works.

[@audyo123]

How long to complete a typical image?

[@4rrxw794]

@@audyo123 You can complete 512x512px image within 5-8 seconds on 1060 3GB VRAM (no tensor cores) with LCM mode or a 'turbo' model. 2060 SUPER can generate 1280x720px image in one second and this is with a ControlNet applied. But at the end of the day you realize it's actually 4k that's needed and this currently needs 20-24GB VRAM at least. Or a lot of time and workarounds. For example 4k image takes roughly a DAY (22-24 hours) on 1060 3GB.

[@prashanth384]

Can we use 2 toys with 12 GB ,Say 2RTX 3060 would it be able to give better results ?

[@Pixovert]

Great question!!! With recent versions of RTX cards you generally don't get an advantage. There was a way of doing this with older GTX cards, but that was mainly geared towards gaming - you can productively pair some RTX Professional cards but not the latest 6000 Ada Lovelace series.

[@andreiirimia414]

What about 3090 24 gb? It's quite good price

[@ShadowangelZero]

Is buying a used 3090 a good idea?

[@Pixovert]

There are pros and cons covered here: kzbin.info/www/bejne/iV64o5ibqayGo80 Get an RTX 4090 at HALF PRICE Using this One Weird Trick (?)

[@dogwood178]

What about 2060 12gb? For 150 usd

[@DarianWade]

Snagged a new Gigabyte 3060 for $226 after shipping and tax on Amazon yesterday. That's cheaper than a lot of used ones. Absolute steal considering the next best option would be double the price 🤭

[@milhouse777]

Hey, what is more important in rendering AI images, cuda cores or ram memory? for example what's better a rtx 3060 12gb or a 3070 (more cuda cores) with 8gb, can someone explain me how it works?

[@Bicyclesidewalk]

the 3060 12GB is a great card for SDXL and Cascade - running ComfyUI in Linux box - I have not had any issues as long as you don't ask it to do something impossible~

[@caioluccas3375]

I have a radeon 5700 and I was thinking of buying a new video card. The video card I have in mind is an RX 7900 XTX. Do you think it's worth it to play with Ai?

[@GOODWINweddingmovie]

Can you make your voice some little bit softly?

[@Johan-rm6ec]

Memory is the most important thing with Ai. 2x4060 ti 16gb gives 4070 ti super performance but with 32gb of vram. Nice if you want to do other Ai stuff also.

[@themanager4983]

I was told the old (sub £500) 4060's are trash (memory bandwidth?) and that you're better off (when starting out like me) with a 3060 12gb and waiting until you can afford/justify a 24gb? Even, perhaps, waiting for the next gen because the latest models are decidedly meh when measured on a price to performance ratio? Thanks for the content, greatly appreciated. Don't care personally, but some may find the vocal fry a tad grating.

[@Pixovert]

The reduced memory bandwidth and increased cache with the RTX 4060 Ti and other lower end 40 models actually pays rich dividends in power consumption, reduced temperatures, improved acoustics, and most likely improved durability and a bunch of other factors that don't necessarily appear in the typical benchmarks. Be careful of faulty "gamerbro" advice on the RTX cards - there is quite a lot of very poor decisions that are being made based on that advice.

[@4rrxw794]

The only factor that is actually a trash is the ridiculous pricing.

[@monolithsoft_guy]

@@Pixovert Oh wow, you''re the first person I see recommending the 4060 Ti. So it's actually good for AI art? I thought about getting a 4070 Ti Super with 16GB VRAM, would that be a good idea? thx

[@Andytlp]

@@monolithsoft_guy Theres only two things you need to consider for SD generations, vram and ram. So for example 4060ti 16gb version would technically beat 4070 12gb one if your goal is some ridiculous resolutions or larger image batch sizes. Yes it would churn the images slower but more at once, higher quality if you do get past 12gb buffer.

[@monolithsoft_guy]

@@Andytlp thx. So a 4070 Ti Super with 16GB VRAM would also be a good choice for AI art, right? How much regular RAM would you recommend? I thought to go with 32, but I was contemplating about 64GB. thx