16/06/2025 12:52 am
Topic starter
Notifications
Clear all
Desktop Computer Forum
2
Posts
3
Users
0
Reactions
1,018
Views
0
I'm a graduate student transitioning from using university clusters to needing my own workstation for deep learning research. I'm primarily working with PyTorch and TensorFlow, training neural networks for computer vision tasks. My models are getting increasingly complex, and I'm tired of waiting in queue for GPU time on shared systems. I need something that can handle large datasets, train models overnight without breaking a sweat, and ideally support multiple GPUs down the line. I'm thinking I need serious VRAM for the larger models I want to experiment with. Budget-wise, I can probably swing $4000-6000, especially if it means I can be more productive with my research. What's the sweet spot for GPU selection these days? Should I go for a single high-end card or multiple mid-range ones? Any advice on CPU and RAM requirements would be super helpful too!
2 Answers
1
14/02/2026 1:15 am
Tbh I just saw this and I went through the exact same thing a few months ago! I'm still a bit of a beginner with the hardware side, but setting up my own rig for research taught me a few things about where the money actually goes: * VRAM is basically everything. I tried to save a bit of cash at first, but I kept hitting 'out of memory' errors on my vision models. Its way better to spend the extra money there first.
* I decided to go with a single really powerful card instead of two smaller ones. I was worried about their being driver issues or it being too hard to code for multiple GPUs as a beginner.
* Don't forget the power supply! My current setup pulls way more electricity then I expected, and I almost cheaped out on that part.
* Cooling is a major hidden cost. My room gets so hot training overnight that I had to buy extra fans for the case!!! Are you planning on building it yourself or getting it pre-built??? Honestly, doing it myself saved me a ton of money that I basically put back into more system memory. Good luck with the research!
0
17/06/2025 2:00 pm
