Best Value Graphics Card for Stable Diffusion XL | MSI

When it comes to AI models like Stable Diffusion XL, having more than enough VRAM is important. The newly released Stable Diffusion XL (SDXL) model from Stability AI can now generate improved faces, legible text, and more aesthetically pleasing art using shorter prompts. However, these enhanced capabilities come at a hardware cost, specifically, VRAM requirements and GPU performance. Here MSI, we use current and previous-gen RTX graphics cards to better understand the minimum hardware requirements and what performance you can expect from them.

#RTX40 #SDXL #msi #StableDiffusion #AI
--------------------------------------------------------------
Follow us to get the latest news!

[DurzoBlunts]

Props for going out of the way to do this, great job on the workflow usage.

[jeff]

Can't believe MSI did video on Stable Diffusion and telling people what best value GPU is. I'm looking for upgrade and this was eye opening. Thanks MSI!

[thewebheadgt]

So more VRAM makes the image generation faster?

[CabrioDriving]

Can't believe what I see - MSI doing such a great video. This is an awesome show of performance. I typed MSI 4060 Ti 16GB Stable diffusion and got this video. wow.

[EinSwitzer]

I can dig it, y’all at amd s acquisition problems of links to expansions of code you never think of / gddr6 speeds the chip has to do a lot of memory placement which is memory in itself / training software then things to use with the extra variables

[ActionBrother]

How does this compare to a 4070 or a 4080?

[EOSGUY]

What about normal SD 1.5? I'm interested to see the difference between 4060 ti 16g and 4070 ti

[jacobs.]

4060ti or 4070super? Worth the extra price?

[NaReN_]

Just saving money to buy this, can’t wait 👁️👄👁️

[K_R_N.]

I should have watched this before I bought a 3080

[vityapelevin]

1:13 at 3080 from below it looks much better and more photorealistic

[HPCAT]

You do know there's other colors than green, right?

[РоманСырватка]

That's bullshit. My result, 3080 10 GB, 1024x1024 +Lora ~10 sec., + 1 Control net ~13 sec. +2 control net ~17 sec.

[EinSwitzer]

Bios chip memory / double it also 400mhz make it 1ghz bios chip

[jesselozier]

I'm having trouble understanding these results. I matched these settings with my RTX 3080 8GB, and I get 27s on the first run (because it has to load the model), and 12s on each generation after that. How does my 3070 8GB (or a 3060) outperform a 10GB 3080?

[wedsg]

i just checked my 7900xt with the same settings in automatic1111 = steps 20 SDXL 1024x1024 does 8.5 seconds on rocm in Linux (ubuntu 22.04) and 11.5 seconds on ZLUDA in windows. Direct ML is out of the question as it's tremendously slow and inefficient.

[FederickTan]

Is there a big difference in performance and speed in AI tasks like stable diffusion and video rendering etc between RTX 4080 super and RTX 4090?Which one should i buy as I seldom play games or should i wait for 5090 at the end of the year?I am not a video editor or hold any jobs related to designing or editing, just a casual home user.

[QQ]

i'm going to build a PC with MSI 4070 ti Trio...
can make a video compare all 40 series GPU?

[xtlm]

why is a 3080 so much slower then a 3060

[maximood-tired]

Thank you, really interesting results