What GPU to buy? Games + local AI
Posted by Deep-Tomorrow4667@reddit | buildapc | View on Reddit | 3 comments
Hi,
A while ago, after years of using laptops, I decided to build my first PC. I’ve put everything together and it’s working, but I’m still missing a GPU.
My specs:
Ryzen 7 7700
ASRock B850 Pro-A WiFi
32 GB DDR5 6000 CL36
Lexar NM1090 Pro
Corsair RM850x
Endorfy Navis F280 AIO
DeepCool CG530 case
7× Arctic P12 fans (6 intake, 1 exhaust, plus 2×140 mm from the AIO as exhaust)
I’m mainly going to play AAA games. Right now I’m on 1080p, but I plan to upgrade one monitor to a 1440p ultrawide (21:9). I’m not super demanding, but since I already spent quite a bit on this build, I’d like a GPU that makes sense long-term.
I also want to start experimenting with AI locally. Nothing serious/production, mostly learning and curiosity. I don’t know, there’s just something exciting about the idea of running it on my own machine instead of relying on cloud services and servers somewhere out there.
Because of that, I’m basically looking at NVIDIA.
Cards that got my attention:
RTX 5070: seems decent, but only 12 GB VRAM, and I’ve seen people say it may age poorly / even be slower than 4070 Super
RTX 5060 Ti 16 GB: same price as 5070 where I live (\~$40 difference), better VRAM but lower performance
Used 4070 / 4060 Ti 16 GB: prices are close to new cards (100-150$ differences), and there’s no warranty, so it feels risky
RTX 5070 Ti: ideal, but out of budget
I’m a bit stuck and not sure what the best balance of performance, VRAM, and future-proofing is here.
Please help my make up my mind!
Ortizzer@reddit
If you want to get into local Ai while gaming, grab a used 3090. You will want the extra vram to run models.
David375@reddit
Look at benchmarks for the 5060TI and 4070. If they play the games you want to play at the quality levels you want to play them at, then you can't go wrong with more VRAM for the AI side. 12 and 16GB is the difference between having to offload layers to CPU and hamstringing performance of mid-size models, and keeping models + context entirely within VRAM for best speed. It would also help to go spend a few dollars on AI services like OpenRouter, figure out which models you actually care to run, and then backing out the requirements you need to run it with benchmark sites like https://www.canirun.ai/. Within a 4080 Mobile 12GB laptop, I've had no problem running smaller coding-centric models like Qwen 3.5 9B or Gemma 4 E4B within those constraints.
Blankthehustlerstone@reddit
This is a tough one but since your budget is restricted and your AI work is just casual right now I’d recommend 5070 since it’s a lot faster than 5060ti