As we previously disclosed, NVIDIA will introduce a Tesla K10 (Kepler GPU) products during the GPU Technology Conference 2012, which starts tonight at San Jose, CA. The new naming convention is expected to separate the product from the previous generations, which carried the C/M/S XXXX (i.e. C2070 for discrete card, S2070 for a 1U server) naming convention.
In a brochure provided by SGI, the company announced that you can purchase a "factory built, preconfigured Tesla K10 GPU solution of software and hardware starting at less than $200K." If you are wondering what level of performance $200K can get you, second sentence gives you the answer; "all you need for a powerful, 75 teraflop peak performance deployment in a single rack."
The company is already shipping units in volume to customers, with finance, oil & gas, radio astronomy. We have already heard of customers from the above mentioned industries acquiring a significantly large amount of Kepler-based Teslas, replacing the GT200b and Fermi based parts.
