Hi,
I’ve been using a 3070 Ti for a while, but it’s definitely not enough anymore. I’ve discussed this with AI, checked Reddit, and done some research. I’m still undecided.
I use Topaz Video AI very often, and I’d be really happy if I could run Starlight Mini locally.
Right now, it feels like there’s only one reasonable option: the 5070 Ti with 16 GB of VRAM. The 5080 is a bit more expensive, but it has the same amount of VRAM. The 5090 with 32 GB is just way overpriced at the moment.
I can afford the 5090, but it’s definitely not a small amount of money for me.
So I’d love to hear opinions from real humans 
who use Topaz Video AI.
Should I wait until the 5090 drops to a more reasonable price?
Or should I go with the 5070 Ti and upgrade in a year to something with more VRAM?
Personally, I have a 5080 and a 3080 ti. I plugged in the 5080 for a few hours to test it. I swapped it back for the 3080 ti because they were about the same for the models I use. This was before the optimizations were in place. Looking at the benchmarks now, I should get about double the speed or one fourth better on some models. The drivers were a big deal-breaker at the time. I still have the 5080 sitting in a box. I told a friend I’d save it for them so they can get one when they have the money.
If you’re willing to wait for Starlight Mini, I think the AMD 9070 XT is the best deal for TVAI processing right now. I use Nyx 3 a lot and that’s one that the 5080 only gets me a fourth better speed. The 9070 XT would get me double speed for less money. Because of all the driver issues Nvidia seems to have at the moment, I would go for the 4090 or 4080 before a 5080 or 5070 ti.
1 Like
Comparing the 4090/5090 and 4080/5080, the 90s have more VRAM than the 80s, but tend to benchmark only about 15-20% faster. At twice the price or more. For me, that’s past the line into diminishing returns even if I can afford it.
3 Likes
Thank you! I’ll wait a little bit longer and see how things go. The 5070 Ti is starting to drop in price here, and if we see support for these cards from Intel in the future @dakota.wixom, it would make more sense for me to build something more powerful https://youtu.be/vZupIBqKHqM?si=LeJtZVI2q-YhpvSx
