There is no doubt that NVIDIA plans to debut the Tesla 3000 series part in the next couple of days. However, according to the sources close to the company, it will not be based upon GK110 chip, which is not ready, but rather launching a Tesla version of the GTX 690.
This is somewhat unexpected as in the past, NVIDIA kept the dual-GPU cards limited to consumer GeForces – mostly due to thermals and too high power consumption for the part. Things have changed, and the demand for power of different type – GPU compute one – outweigh any other negativity.
NVIDIA GeForce GTX 690 4GB is preparing for a second birth: This time as an 8GB Tesla card
Given that the GeForce GTX 690 was built with exceptionally high level of quality, do not expect a Tesla version to be all that much different with the exception of removing the display connectors but the essential one, enabling the whole second slot to be used for exhaust.
Every GPU is paired with 4GB GDDR5 memory for a grand total of 8GB, a world record for a GPU. Our sources warned us that this product is also suffering from the GPU Compute choices the company made with GK104 chip, and the new Tesla C3000 Series part will follow suit. We’ve been told to expect good SP performance (IEEE 754 "binary32" Single Precision), and GeForce-specific weak DP FP (IEEE 754 "binary64" Double Precision). According to the sources, "the big daddy silicon" is still not ready, while there is a lot of market demand for Kepler-based SP FP applications.
Given that Kepler GPU architecture increased the amount of SP FLOP (Floating Point Operation) from 64 to 96 in a single cycle, we understand the market demand for SP. While HPC customers will probably opt to wait for GK110-based Kepler boards, there is a large amount of customers that said that they don’t give a "rat’s behind" for DP and that they demand a Kepler part for their computational needs.
Thus, you can expect a seven-billion transistor Tesla card with 3072 CUDA cores, 512-bit memory interface, 8GB GDDR5 with over 300GB/s of video memory bandwidth (shared between two GPUs), as great looking and expensive high-end heatsink and increased cooling capabilities.
The price is still unknown, but in any case – it is a very brave move from NVIDIA to launch their first commercial high-performing dual-GPU parts (Quadro NVS from yesteryear excluded.)