I'm planning to build a PC mainly for machine-learning experiments, but current component prices are making it difficult to justify. 32GB of DDR5 memory costs around $400 unless I choose an unfamiliar brand, and a 16GB VRAM GPU such as the RTX 5060 Ti is roughly $750. Do prices seem likely to improve within the next two or three years, or should I buy what I can now?
3 Answers
For machine learning, don’t assume you need the most expensive parts. Traditional ML and smaller local experiments may work fine with 16–32GB of system RAM and a GPU with 8GB or more of VRAM, depending on the models. You can start with a midrange AM5 system and upgrade the CPU or memory later. For larger models, renting cloud GPU time may be cheaper than buying hardware that sits unused, especially if you only run experiments occasionally.
A used system or a discounted prebuilt could be a better compromise than buying every part at current retail prices. You might also build around DDR4 if your workload benefits more from having 64GB than from DDR5 speed. Waiting only makes sense if you already have a computer that meets your needs; otherwise, the value of using the system for the next year or two may outweigh the possibility of a future discount.
Nobody can reliably predict hardware prices several years out. Memory and GPU demand from data centers is keeping supply tight, and manufacturers reportedly have much of their upcoming production committed already. Prices could fall if demand drops or the AI investment boom slows, but that might take years—and prices could rise further before then. If you need the machine and can afford it, buying a sensible configuration now is safer than waiting indefinitely.
For virtual machines, containers, and self-hosting, capacity is usually more useful than the modest performance advantage of DDR5. If 64GB of DDR4 costs the same as 32GB of DDR5, the larger capacity is often the better choice—provided the platform still has a reasonable upgrade path.