Re: > I need it for local coding tasks....

• I totally agree with the others about prioritizing VRAM. Stumbled upon this discussion and wanted to add that you should look into the used market for GPUs because it is definitely the most budget-friendly way to hit your performance targets without blowing the 2k budget. Here is how I would split the costs:
• Try to snag a used NVIDIA GeForce RTX 3090 24GB GDDR6X if you can find one locally. It is a workhorse and that 24GB of VRAM is perfect for coding models.
• Pair that with a reliable motherboard and a solid CPU like the Intel Core i7-13700K 3.4GHz 16-Core.
• Don't skimp on the power supply; get a EVGA SuperNOVA 850W Gold Rated to keep everything running stable. I am super happy with my own setup running similar specs. It works well and I have had no complaints while coding daily.

Honestly, dont sweat it too much. That 2k budget is actually a pretty solid starting point if you play your cards right. I know it sounds scary, but you really dont need some crazy enterprise server. Focus on VRAM because that is basically what makes or breaks local AI. I would suggest aiming for at least 24GB of VRAM if you want things to run smoothly without too much stuttering. Honestly, a NVIDIA GeForce RTX 3090 24GB GDDR6X is a lifesaver here. You can often find them used on marketplaces for a killer deal. Make sure to check the seller history though, just to be safe. Pairing that with a decent CPU like the AMD Ryzen 7 5800X 8-Core 3.8GHz will keep you in budget while leaving enough cash for a solid power supply. Be careful not to skimp on the PSU because you really dont want a cheap unit frying your components. Also, grab at least 64GB of RAM like the Corsair Vengeance LPX 64GB DDR4 3200MHz just in case you overflow into system memory. It might be a bit slower than the GPU, but it saves the whole task from crashing. Just take your time with the build and make sure your case has decent airflow... honestly, these cards get pretty toasty when they are working hard. You got this.

> I need it for local coding tasks. Do I need a massive GPU or just lots of ram? Honestly you need both, but prioritize the GPU VRAM above everything else. Coding models love context, so if you run out of memory, the whole thing just grinds to a halt. Since you have a 2k budget, you should grab an NVIDIA GeForce RTX 4090 24GB GDDR6X. It is the gold standard for a reason. Pair that with at least 64GB of memory, like the G.Skill Trident Z5 RGB 64GB DDR5 6400MHz, to keep your system snappy while the model is loaded. I am super satisfied with my current setup and it handles local inference like a champ. Dont worry about the noise or heat too much, just get a decent case with good airflow and you will be golden.