I'm looking at an RTX 3090 mainly because its 24GB of VRAM would be useful for local AI workloads. Gaming performance would be a bonus, and it would still be a major upgrade from my current RTX 3060 12GB. The options I've found are from Palit, INNO3D, and EVGA, but the market is rough: even older cards are often priced above $1,000, while some RTX 5060 Ti models are approaching $900. Is a 3090 still a sensible purchase at these prices, and does the brand make much difference?
3 Answers
The RX 7900 XTX is probably the strongest alternative if you can find one in the $700–850 range. NVIDIA may still be the safer choice for compatibility with AI tools, but 24GB on a 3090 doesn’t justify paying a huge premium over newer or cheaper alternatives. If the XTX only reaches that price after trading in your current card, calculate the actual out-of-pocket cost before deciding.
Palit, INNO3D, and EVGA are all legitimate manufacturers. EVGA’s main drawback is that it no longer makes graphics cards, so warranty replacement could be difficult or impossible. For a used card, especially in a market where warranty support isn’t really available, condition matters more than the badge. Check temperatures, memory errors, fan noise, and whether the card has been used for mining.
Warranty isn’t really a factor for me since I’m buying second-hand in Russia. I’m mainly comparing the cards by condition, cooling, and price.
At around $1,250 used, I don’t think a 3090 is a good deal anymore. For local AI, a refurbished or used RX 7900 XTX around $700–850 could be worth considering, although NVIDIA generally has better software support. Another option is a modified 20GB RTX 3080, or Intel’s Arc B-series card with 32GB, provided your applications support the software stack. If you’re spending more than $1,000, compare the 3090 against newer cards rather than assuming the extra VRAM automatically makes it the best choice.

Getting one for $850 would only be possible for me with a trade-in of my current graphics card, so the real comparison is more complicated.