I have built this config for my friend who
trains LLM models and requires a setup in mid range. Let me know if it require any change. Budget
$1700 USD (1.5 Lacs INR).
Processor (CPU) | AMD Ryzen 9 7900X |
Cooling System (CPU Cooler) | Cooler Master Liquid 240L Core 240mm CPU Liquid Cooler |
Motherboard | Gigabyte B650 Aorus Elite AX V2 (Wi-Fi) |
Memory (RAM) | Corsair Vengeance RGB 64GB (32GB x 2) 6000MHz DDR5 RAM |
Solid State Drive (M.2 SSD) |
Crucial P3 Plus 1TB M.2 NVMe Gen4 Internal SSD |
|
Hard Disk Drive (Internal HDD) | Western Digital Blue 1TB 7200 RPM |
Graphics Card (GPU/VGA) | Asus Dual RTX 4060 Ti SSD OC Edition 8GB |
Power Supply Unit (SMPS/PSU) | Thermaltake Smart BX1 750W 80 Plus Bronze |
Cabinet (Case) | Deepcool CG580 ATX Mid Tower Case |
The VRAM is definitely a concern for AI because the hard stuff is processed by the GPU and it wants to be able to store a ton of data there. If you selected the 4060 ti version that has 16 GB of VRAM, it would be much better. I don't think it's very much more in cost. The 4070 ti would also be a good choice.
The GPU is going to be doing the heavy lifting. You still need a good CPU, but if your choice is to go down a bit on the CPU so you can go up a bit on the GPU, you should do it.
Also, a 4070 TI with 12 GB of VRAM is better for AI than a 4060 TI with 16 GB of VRAM. VRAM isn't the only thing that's important in a GPU for AI.
Lastly, ignore whatever reddit has to say on the matter. They give terrible advice regarding training AI. They just pick whatever would be better for gaming.