During the GTC 2012 opening keynote, NVIDIA announced two Tesla cards, K10 and the K20. While the K10 is effectively a 8GB Tesla-version of the GeForce GTX 690, the K20 brings the monster silicon – 7.1 billion transistors.
The long awaited Kepler GK110 made a public appearance during yesterday's GTC 2012 opening keynote, arriving to the world stage in the form of NVIDIA Tesla K20 GPU Computing card, a DP-FP GPGPU beast. Curiously the actual availability is pushed all the way to Q4 2012, in time for the actual debut at SC’12 in Salt Lake City . The unpublished goal is to have a Tesla card sit on top of Top 500 list of world’s most powerful supercomputers with upgraded Jaguar, i.e. Titan at Oak Ridge National Laboratories (ORNL). Given that there will be around 15,000 Tesla K20s in the system, that is one heck of a competitor for the crown.
If you are wondering what NVIDIA did pack in the 7.1 billion transistors, the answers are as follows:
- 2880 CUDA Cores
- 15 SMX Clusters
- 384-bit Memory Controller
- Up to 24GB of GDDR5 memory
- 2nd Gen ECC
- Hardware GPU Silicon Virtualization
- Hyper-Q (Slashes CPU idle time by allowing multiple CPU cores to simultaneously utilize a single Kepler GPU, dramatically advancing programmability and efficiency)
- Dynamic Parallelism (Simplifies GPU programming by allowing programmers to easily accelerate all parallel nested loops – resulting in a GPU dynamically spawning new threads on its own without going back to the CPU)
- 50-85% Double Precision Rate to Single Precision
- At least 1.5 TFLOPS DP FP64
- Target: 250 GB/s bandwidth
The reason for the wait until the fourth quarter does not lie with the company, though. The one spec remaining was the amount of memory that Tesla K20 will have, and the answer is not the easy one; amount of memory. Currently, NVIDIA has access to 6GB and 12GB memory (up to 24 chips per board), but there are developments that might double the amount of memory in a single chip. Given that Tesla is a part with 70% gross margin, there is plenty of space for NVIDIA to even order a small quantity and pay the associated premium of high-density 8Gbit (1GByte) GDDR5 memory chips.
The K20 is not the only GK110 based Tesla, though. We’ve learned that the company will offer more affordable products as well, most likely with 1-4 SMX cores disabled, i.e. 192-768 CUDA cores might end up disabled from lesser ASICs.
GK110 will not appear in Quadro or GeForce products this year. It is expected that NVIDIA will fight AMD’s new Graphics Core Next based FirePro at SIGGRAPH 2012 (early August 2012), with GK107 and GK104 based parts, while professionals will probably have to wait for GK110-based Quadro until April 2013 (NAB 2013). GeForce gamers should get their hands on GeForce GTX 700 Series sooner, though. Did we just mention GTX 780? Dang…