I'd advise against using a GeForce in a server environment for a variety of reasons:
GeForce cards don't have ECC RAM
Tesla GPUs are tuned for reliability, not simply in being error-free as possible, but for minimal downtime
Depending on the task, it may not run as well, if at all, on a GeFore card due to market segmentation.
If you decide to get GeForce cards, I'd advise at the minimum tuning them to run at base clock speeds when doing something. Running cards hard is fine for gaming, but I'd argue it's not fine for doing data crunching where accuracy is likely important.
Honestly, I wouldn't not when the Tesla is designed for workloads meant for data centers. The GeForce is designed for gaming...that's due to market segmentation.
I'd advise against using a GeForce in a server environment for a variety of reasons:
GeForce cards don't have ECC RAM
Tesla GPUs are tuned for reliability, not simply in being error-free as possible, but for minimal downtime
Depending on the task, it may not run as well, if at all, on a GeFore card due to market segmentation.
If you decide to get GeForce cards, I'd advise at the minimum tuning them to run at base clock speeds when doing something. Running cards hard is fine for gaming, but I'd argue it's not fine for doing data crunching where accuracy is likely important.