Intel chipsets consistently get better host->device and device->host memory transfer speeds (~6 GiB/s vs ~4 GiB/s), but that is with PCIe gen2 devices. Since the Tesla C870 is only PCIe gen1, there wont be any performance differences. But you will see a difference in transferring to the GTX 260.
I’m not aware of any other performance differences related to CUDA. If cost is an issue, the 780 chipset might be cheaper: it uses very inexpensive DDR2 memory. I haven’t kept up on all the intel chipsets, but most are now using the very expensive DDR3 memory.
If you have a small budget, why’d you get a tesla? Funny thing is that the 260 will be faster than it. For a display card, you can get something cheap or use onboard graphics. (If you put a c870 and a 260 in the same case, be sure to get a BIG power supply.)
Because if you need more than a gig of memory, you buy a Tesla (or a Quadro, but that’s even more money)? Also, you could probably get away with a 750w power supply if you’re going to be using a 45nm Core 2.
I’d use the 780i, just because 16GB of DDR2 is $500, and 8GB (4x2GB) is less than $200. Right now, the 780i doesn’t actually support 16GB, but apparently there’s a new BIOS coming out very soon that adds support for it. Something to keep in mind!