California-based graphics IC design company Nvidia has just unveiled details about the most anticipated GPU of this year, on its official website. We are, of course, talking about the huge GK110 graphics processing unit (GPU).
The initial presentation took place yesterday at the GPU Technology Conference (GTC).
Nvidia has presented the new TESLA K10 and TESLA K20 GPUs .
TESLA K10 is practically a GTX 690 dual GPU card with the usual small tweaks for professional use and the software toolset that comes with it.
The company claims a huge 4.58 teraflops of single-precision floating point performance for the K10, along with the impressive 320 GB/s bandwidth.
This is only 20% faster than AMD’s Tahiti-based Radeon HD 7970 card that comes with a 260 GB/s bandwidth. It’s quite an achievement for AMD to deliver such a great compute performance well ahead Nvidia and with much less fuss.
Nvidia’s “fuss” is generally well-targeted and quite profitable, as the company is working hard to integrate its technologies in the high margins / high profits server / professional markets.
Practically, Nvidia is doing a much better job than AMD at integrating its hardware with the much needed software that would reveal the computing power inside GPUs.
As Intel proved for the past decade, software is a very important part of hardware development.
Nvidia’s K10 will come with a full 8 GB of GDDR5 memory complement.
The other TESLA card introduced by Nvidia is the yet mysterious K20. Nvidia didn’t share too much info about it, as this is based on the unreleased GK110 GPU.
The graphics card company already has contracts based on the K20 design and this is what they had to say about its new K20 cards:Tesla K20 supports the Hyper-Q and dynamic parallelism capabilities. The GK110 GPU is expected to be incorporated into the new Titan supercomputer at the Oak Ridge National Laboratory in Tennessee and the Blue Waters system at the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign.
Nvidia says that K20 will be able to deliver three times the double-precision performance of the Fermi-based TESLA accelerators.
On the first look, this makes the GK110 GPU 50% more powerful that the 1,000 USD GeForce GTX690.
K20 will also come with more GP Compute technologies as listed in the table above.
While we commend Nvidia for its commitment to the professional GPU Computer market, we’re starting to feel pity for the GTX 690 buyers.
Luckily, there is a severe shortage of Kepler GPUs so the wallets of Nvidia enthusiasts are safe.