I want to run a custom-built deep research equivalent locally. I also want to be able to run coding agents.

I don't care much about speed (though it shouldn't take a crazy time like 12hrs+ to deep research), but I'm aiming for quality outputs mainly.

What sort of specs would I be looking at, for this sort of build?

My research tells me \~256gb vram would be a good minimum to run some of the higher end models.

I'm thinking of building a server with 10 x Tesla P40 24gb (1/2 the speed of 3090 for 1/5 the cost) and dual Intel Xeon scalables (i.e. TYAN Thunder HX FT83- B7119)

Does this seem like a viable option to aim for? Did I miss any other high value option?