![]() ![]() So the GPU will work, but data transfer from the CPU to the GPU will be slower. So normally what happens in most people's builds is the GPU takes up 16 of those lanes in the top slot, which means that if you put a second GPU in the bottom slot there would only be 4 spare lanes left. The motherboard you chose has two x16 length slots, but they won't both run at x16 speeds at the same time. If you want to leave the option open for a dual GPU build with 128GB ram I think the most cost effective route is to go with AM4 (either 2x 3090 or add second GPU later):, since there are affordable motherboards that support this and the CPUs have more PCIe lanes than the Intel equivalents too. 'memory support') to make sure the RAM you're buying is guaranteed to work. In that case I'd shop around for a Z690 motherboard that has the aesthetics/feature set you need, ideally sub $200, and then check the motherboard's QVL list (a.k.a. If you want to keep the RGB and CPU with a single GPU setup, the main ways to save $ are to swap the PSU, motherboard, RAM and GPU to cheaper models. ![]() You really don't need a beefy CPU for deep learning though - in your budget you may as well since it does help with preprocessing, but if you wanted to save even more here you could do. This build saves you $1k and keeps the same peripherals, but the CPU/aesthetics are worse. 1200W) - there is also a 1600W SuperFlower PSU for $100 less than the Corsair model. If you aren't planning to add a second GPU, you could opt for a cheaper PSU (e.g. Unfortunately you might have to settle for one of the slots running at x4 from the CPU though unless you go with an expensive DDR5 motherboard that supports x8/x8 (e.g. Also I'd stick with DDR4 if you want 128GB RAM but you could probably find some cheaper 3200 CL16 or 3600 CL18 kits to save a few more $ if you don't care about RGB (same goes for the cooler).Īre you planning to add a second GPU at any point? If so, you'll want a different motherboard, ideally one that has two PCIe x16 slots that both run from the CPU and are spaced as far apart as possible. I think if I were you I'd rather have the 24GB GPU though. The way you could save money on the GPU and keep fast training times is if 12 or 16GB VRAM is enough for your models that way you could use the 4070Ti or 4080 instead. The main disadvantage of having the 3090 is everything trains quite a bit slower, so it's up to you if that's a worthwhile tradeoff or not. I'd rather do that than downgrade the GPU on your budget, although if you want a huge saving you could probably pick up a used 3090 for $600-700. Those specs will be great but if you want to save a bit of money you could downgrade the CPU a bit. | *Prices include shipping, taxes, rebates, and discounts* | I imagine the biggest cost variable here is the graphics card, and I wanted a really high end one, but I'm not sure how much I can get away with in terms of downgrading a bit from like a RTX 4090 to a 3090 or something.Īgain, I'm new to this whole process of PC building, so any insight is appreciated, thank you! If anyone wants to give me any suggestions on how I can improve the build I have below, I really appreciate it!īelow is a list I have on PC part picker so far. I also am not quite sure if there may be any dimension/wiring compatibility issues that I may not be familiar with, as I've never built my own PC. I've been using resources like PC partpicker and other blogs and youtube videos I've found online to try and make sure everything I have is compatible, but I'm worried that I may be putting extra money in certain places I may not need it. I'm ultimately trying to use this PC for a lot of deep learning/reinforcement learning work, and my ideal budget is in the $3k-4k range. ![]() I'm currently in the process of building a new PC for grad school, but I'm trying to make sure I can be as cost efficient as possible, while maximizing performance for the work I'm trying to do. Pick, Assemble and Install: Video Guide.No intentionally harmful, misleading or joke advice.No excessive posting (more than one submission in 24 hours).No selling, trading or requests for valuation.No self-promotion, advertising, begging, or surveys.No submissions about memes, jokes, meta, or hypothetical / dream builds.No titles that are all-caps, clickbait, PSAs, pro-tips or contain emoji.No submissions about retailer or customer service experiences. ![]() No submissions about sales, deals or unauthorized giveaways.No submissions about hardware news, rumors, or reviews.Please keep in mind that we are here to help you build a computer, not to build it for you. Submit Build Help/Ready post Submit Troubleshooting post Submit other post New Here? BuildAPC Beginner's Guide Live Chat on Discord Daily Simple Questions threads ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |