28/07/2026 5:25 pm
Topic starter
Notifications
Clear all
GPU Forum
2
Posts
4
Users
0
Reactions
79
Views
0
Is the RTX 4070 Ti actually enough for Stable Diffusion or should I just bite the bullet and get a 3090? Ive seen a bunch of forums saying VRAM is everything but some people claim the newer architecture makes up for it. Im in Canada with a 900 dollar budget and need this sorted by next week for a project. Really dont want to waste cash on the wrong card.
12
28/07/2026 5:35 pm
Honestly, if you are stuck on that 900 CAD budget, stay away from the 4070 Ti for AI. You really need that 24GB of VRAM for training stuff. I grabbed a used NVIDIA GeForce RTX 3090 24GB GDDR6X locally for way less than retail. It crushes everything I throw at it. Just be careful with cooling on those cards, they get pretty toasty after a few hours of rendering...
2
28/07/2026 5:25 pm
Honestly, I used to obsess over this too, but after running a bunch of models, the VRAM situation is real. I started on a NVIDIA RTX 4070 Ti 12GB GDDR6X and while it is super fast for rendering images, I kept hitting walls when I tried training LoRAs or using high-res fixes. It just runs out of steam once you push the resolution.
- The NVIDIA RTX 3090 24GB GDDR6X is a total beast for this. Having double the VRAM lets you upscale like crazy without crashing.
- The 4070 Ti is great if you stick to basic generations, but it feels limiting fast. Basically, grab a used 3090 if you can find one under your budget. Being able to actually use ControlNet and hires fix simultaneously without getting OOM errors is such a game changer. I'm much happier since I made the switch.
