PC Master Race

When training you'll want way more VRAM than you need to run inference - get a 90 series GPU for the memory.

Fortunately, my worn is not LLM related but just simple neural networks, but I don’t know how that might affect best practices for hardware.

Okay. If this is something you already do on existing machines, you'll be in good position to know how much memory you actually need, and then maybe give yourself a little room to grow. My guess would be that you're not working on massive models so you'll probably be fine with a mid-range card.

At the same time, a lot of AI/ML stuff is becoming mainstream and requires a ton of VRAM to get good performance. If you do any work with graphics, audio, or video, you might find yourself running large models without really thinking about it. There are lots of use cases for speech recognition models, for example, which are quite large. Photoshop already has some medium-sized models for some tasks. Noise reduction for audio can also be quite demanding (if you want to do a really good job).

As for system RAM...the world of DDR5 is indeed complicated. I don't think there's a huge need to go over 6000MHz RAM, and faster RAM brings some compatibility issues with some mobos/CPUs. It's also usually faster to use two sticks than four. So 2x32GB would be better than 4x16 in general.

For GPUs in particular, new gens with more VRAM are on the way, so buying the high-end now might leave you with something that feels obsolete by the time you grow into it. If you spend $750 now and $750 again in 2-3 years, you might end up better off than if you spent $1500 today and waited twice as long to upgrade. Particularly if you are able/willing to sell your old equipment to offset upgrade costs.

Consider going with a rail mounted 19'' case. I have 4HE hanging under my standing desk in a chenbro case. Nothing special, moves with the desk and can be easily pulled out and opened on rails. standing desk is motorized, so it's heavy enough.

It's quite the opposite, though. PC components have never been as compatible as they are today with the inclusion of different standards like ATX and stuff.

As for you figuring stuff out, here's how I pick parts:

Coolers: I go see the temperature tests to decide on which one fits depending on noise vs temps vs price.

Motherboards: Here are the main bits I look at

• Compatible socket
• Amount of USB ports
• Amount of Sata ports
• Amount of M.2 slots
• Other ports you'd want to use
• Supported type of RAM (DDR4 vs DDR5)

Then there's extra

• Chipset (top-tier chipsets are often a waste of money)
• Bluetooth and wifi (can always be added later, but nice to have built-in)
• A button for updating BIOS (bypasses need for CPU)
• Troubleshooting LEDs (very handy when tinkering)
• How chunky the heatsing is (bigger = better)
• Amount of power phases (black cubes around the socket, more = better, only need to pay attention to those when going for high power-draw CPU)

PSU: Very simple, go to power draw calculator or multiply power draw from pcpartpicker by 1.3 or 1.4, that's your Wattage requirement. Then find a list of reliable PSUs, look for cheapest reliable one that has enough Watts. It's a good idea to have some overhead as well. Alternatively to a tier-list is knowing which manufacturers are good.

Cooling for RAM: ignore cooling for RAM, not important at all. It's mostly for looks.

RAM clock speeds: MT/s, aka Mhz, is bus width. Higher amount = more data can pass at once. But we're currently at a point when 6000mhz doesn't make much difference against 3600mhz. So, latency is more important. Google, which combination of clock speed + timings (they look like 36-38-38, can also be written as CL36) has lowest latency, go with lowest.

Pcpartpicker makes sure things you put together are compatible with each other. So, start with CPU and GPU.

Nvidia is gold standard for AI stuff, that's why it was so heavily recommended. If training is not a major concern, then AMD is the obvious choice for Linux gaming (unless you need HDMI 2.1).