I don’t know what board you mean, or what generation TR that would be. Having spent last year & a bit into ‘26 building up my first Threadripper workstation I will strongly caution against building one up except for the following cases: you NEED a ton of PCIE lanes for 4ea or more x16 GPUs or similar, you NEED 128GB or more of RAM, you need all that plus a lot of on board NVME and other connectivity, and your use case can take advantage of the large number of cores (specialized use cases).
The 24core 9960x and other 9 series Threadrippers have excellent per core performance, significantly more than previous generations and you won’t be held back like a 3000 series TR build would in normal use (plus the bugs on older platforms). If it will be a somewhat basic LLM box with 4ea 16GB or 32GB GPUs that don’t need PCIE 5.0 and it doesn’t need to perform the best, maybe an older DDR4 Threadripper build makes sense, still not cheap.
But a high end modern 7000 or 9000 series DDR5 Threadripper build will be around $1k motherboard, $1.5-5k CPU, $4-10K RAM (DDR5 TR must use ECC memory, older DDR4 platform could use normal non-ECC ram), $500-900 PSU, $200-1,200 case, plus the cost of 4 half decent GPUs worth buying and powering up (depends, but expect $3-6k on the lower end and $60k on the higher end for 4ea double slot cards)
Again, if you don’t have a good use case you are probably better served by a good used AM4 DDR4 build (5000 Series, like 5800x3D or 5950x ect…) or a more modern AM5 DDR5 build with normal consumer parts. Depending on motherboard choice you can get a lot of features and be able to run 2ea GPUs at x8 lanes each which is not too bad if everything is PCIE 5.0 like an AM5 build and modern GPUs (especially with something like a pair of 5060Ti 16GB cards that are 5.0 x8 anyways).
I am assuming you want to do local AI since it’s kinda what everyone is doing, especially with Threadripper, from text LLM usage for coding assistance to a glorified chat bot, or image, audio and video generation. Modern Nvidia GPUs are generally the best supported, then modern AMD, then Intel. VRAM capacity on GPUs will be the primary thing to get as much as budget allows, VRAM bandwidth is kinda secondary but is very important for generation speed, then PCIE link speed and GPU core capabilities, system RAM capacity, ect… Some use cases can take advantage of multiple GPUs well (text LLMS) some strongly prefer a single powerful GPU (often an Nvidia card with as much fast VRAM at that for audio, image & video generation)
For image, audio & video generation a single 5060Ti 16GB GPU is an excellent starting point with Comfy UI and the free tools within, it can be on a somewhat simple motherboard and still do good. For example I have a little ITX Minisforum BD790i motherboard with a modern laptop 7945HX CPU and a 5060Ti as a lightweight local AI rig. Strongly caution against a sub 16GB GPU, unless you have a killer deal on a used 8 or 12GB GPU as a stop gap.
I have a 9970x 32 core Threadripper, an Asus Pro WS TRX50 Sage Wifi A board, 128GB RAM, 4EA 32GB AMD R9700 GPUs, a Silverstone 2500Rz PSU (1650w at 120v), in a new Antec 900 case with watercooling for the CPU, not a lot takes really advantage of the 4 GPUs or the massive CPU, the lightweight 5060 Ti rig is as capable in things like Comfy UI for single generation workflows. Check out Digital Spaceport on YouTube to see some of the kinda odd local AI setups or use of more normal hardware.