I’m an engineering grad student, and I’ve been tasked with finding parts for building a shared workstation for my lab. Our work includes deep learning, computer vision, network analysis, reinforcement learning, and more, so we need a workstation that can handle a wide range of computational needs for multiple users. We work primarily in Python (PyTorch, Tensorflow, OpenCV), work with large datasets, and often test things in Jupyter notebooks if that makes any difference. There are 6 students in my lab, but the vast majority of the time, only 1-2 people would be working simultaneously.
We're thinking of going with 2x Nvidia RTX 4090 GPUs (48GB VRAM total). However, I've never built a PC, and I’m feeling quite overwhelmed by all the options for the other components.
I'm hoping to get some advice or insight from anyone who has experience with deep learning builds, dual GPU builds, or just PC builds in general. I’m not sure how to balance spending and ensuring we get the right parts for our needs. The budget is \$10k, so I have some flexibility, but my PI would obviously prefer to spend less (I'm thinking \$7 - \$8k range).
These are the parts I need help with and what I've found based on the research I've done so far:
- GPUs: Most build info I've seen are centered around gaming and only use single GPU's which seems to be sufficient for most gaming needs. I haven't been able to find as much reliable information on doing dual GPU builds. Is the Dual 4090's even feasible? Like would packages like PyTorch even be capable of taking advantage of the two GPUs' combined function and memory?
- CPU: People seem to recommend AMD over Intel so I'm looking into Ryzen and Threadripper (although I'm still open to Intel). I know CPU is not as important as GPU, but I think some of my colleagues could still benefit from improved multithreading capability so I don't want to go completely minimal on CPU.
- Motherboard: I'm using pcpartpicker for all of this, and it tell me whether the motherboard is technically compatible with the CPU, GPU, etc. However, I've read that doing dual GPU can run into issues of physical spacing and airflow, which pcpartpicker doesn't account for as far as I know. Any recommendations on models or general advice on selecting a motherboard?
- Cooling: I've also read that the combination of CPU and dual GPUs can have issues with airflow and heat generation, which is why I want to make sure we have a good cooling system. Liquid cooling seems to be better than air cooling, but it's also more complicated and can leak. But like the CPU, there are just so many options so I want to make sure I'm getting a sufficient cooling system, but not overspending. Would liquid cooling be more worth it for a dual GPU build?
I'm less worried about selecting storage, memory, case, power supply, and peripherals, but if anyone has particular advice on these, it would be greatly appreciated as well. Thanks!
