I have had a 10GB RTX 3080 for a few years, and it has done me well for most of that time. Since getting heavily into Topaz Video, though, the 10GB VRAM is proving to be a limitation. I have an option on a 24GB 3090 for just over £1,000 (about $1300) from a friend, and am very tempted. (I can’t afford a 4000 or 5000 series with 24+GB of VRAM!)
My render times are really quite slow on the 3080 because it has to page between VRAM and system RAM a lot, and even with DDR5 6000 it’s quite a drag.
I can see that Starlight Precise is using 20-22GB of total RAM for most of my renders. Is the switch to the 24GB 3090 going to be a revelation, or just a slight improvement do you think?
I get that people’s experiences will differ, but just looking for a note or two from experienced upgraders.
You would see an improvement with the models and various workflows when upgrading to a 3090 card. How much would depend on the specific file specs and the model being used, though.
I’m predominantly using Starlight Precise 2.6 to recover heavily compressed 320x240 AVI (mainly DivX or XVid but with some WMV) VHS transfers, then running that result through a slight Nyx to take the edge off the sharpness in places. The target resolution is a straight 4x upscale to 1280x960.
The CUDA capability of the RTX 3xxx series is 8.6, which is below the 8.9 Topaz specs out for the latest builds of Video. So you’d be rolling the dice on future compatibility, especially for local Starlight processing.
A good point. It might be an issue long term, but 1k vs 5k is worth the punt in the short term I think. Especially if running “legacy” models remains an option as it does now.
Maybe you can find a used 4090-24GB card, price around 2000$, or new 5080-16GB for about 1500$, or a 5070Ti-16GB are below 1500$…anyway, I wouldn’t buy an RTX 3090.
Sorry, but a 3090 for $1,300 is an insanely bad deal. It’s even worse if you already have a 3080, and incredibly stupid if you want to make good use of Topaz Video AI in the future.
I can only agree with Mayday—you’re better off looking for a 4070 Ti, 5060 (16GB VRAM version), or better yet, a 4080, 5080, or even a 4090.
Welcome, if you are looking at the SL models and future compatibility, I would refer to what Gene stated and keep that in mind. The technology is continuing forward and systems/hardware are becoming outdated faster in certain situations.
As Mayday pointed out look around and see what other options there are that are the newer series even if they do not double the VRAM. These will still show improvement in processing due to the newer architecture. However, sometimes these are not available or fall outside of the budget, and in the current market both situations are very likely.
Thanks for the advice folks, I guess I’ll pass on the 3090 and save up a bit (or given the current state of the market, a lot!) more and see if I can get a 4- or 5000 series in a few months.
What does an RTX 3090 cost now? I got an RTX 4000 Pro Blackwell. Same memory as a 3090, about 5% slower and consumes 60% less power. It cost $1600, but with the AI bubble the price may have gone up. Again.
Yes also a good option! Also the none pro, is a candidate, based on prevoius “Ada Lovelace” archiecture, or let’s say simmillar a RTX 4070(Ti) but with 20GB Vram. Open the field, compare prices you get one, speed and Vram you need, and then buy
