I'm currently diving into my first deep learning projects and realized my laptop just isn't cutting it anymore. While Google Colab is okay, I really want a local setup to experiment more freely without worrying about session timeouts.
My budget is pretty tight, ideally under $450. I've been looking at the RTX 3060 12GB because of the extra memory, but I'm also curious if the newer RTX 4060 is worth it despite having less VRAM. I'm mainly focusing on computer vision and some light NLP work.
Does anyone have recommendations for the best value card for a student right now? Is it better to buy new or look for specific used models?
> I've been looking at the RTX 3060 12GB because of the extra memory I actually had a bad experience with a sketchy used card that shorted my whole system once, so I'm highkey cautious now lol. Before I give advice on the NVIDIA GeForce RTX 3060 12GB GDDR6 or the NVIDIA GeForce RTX 4060 8GB GDDR6, can you clarify what power supply youre running with? You gotta make sure your PSU is reliable enough for ML. I'm super satisfied with my stable rig now, but safety first!
sooo, quick question - what kind of model sizes are u actually planning to run?? like, are we talking basic resnet or heavier transformers? honestly, since ur budget is $450, u can find a used NVIDIA GeForce RTX 3060 12GB GDDR6 for like $250-280 easy. helps save cash for other parts!! TL;DR: i need to know your dataset scale before recommending the NVIDIA GeForce RTX 4060 8GB GDDR6.
Seconded!
In my experience, id grab the NVIDIA GeForce RTX 3060 12GB. I used one for my first CNN projects and that extra memory is actually a lifesaver for batch sizes.
Solid advice 👍
Same here!
This^ Also wanted to add that the caution about used gear is real. I once had a second-hand card fry my whole motherboard and it was a total nightmare... honestly not worth it on a student budget. Ngl, it's satisfying to build the rig yourself tho, and my current DIY setup works well with no complaints. What's your current power supply like?? And are you planning to do the install yourself??
Noted!