Since we launched the Tesla K20 GPU accelerators at the Supercomputing 2012 (SC12) show in November, researchers have been eagerly waiting to get their hands on them.
To benchmark the new Kepler GPUs on real-world scientific applications, we offered a number of researchers using our last-generation Fermi GPUs early access to the new Tesla K20 and K20X GPUs before product launch. We asked them – around 40 individuals from science and academia — to take a free test drive of the new family of Tesla GPUs, and share their experiences.
The results are in.
Two-thirds of the users (65 percent) who took the test drive reported speed-ups of 2x or more, right out of the box, without making any code changes. You can just imagine what they will achieve with a couple code optimizations.
And, this was done on a wide range of real scientific applications like GROMACS, GTC, and CST, not on synthetic benchmarks, showing how the new Kepler GPUs can make an immediate impact on research.
Accelerating Clean Energy, Cosmology, and More…
The researchers who participated in the test drive represent a variety of scientific domains, from astrophysics to biology to material science. And they all experienced how Kepler offers an instant performance boost over Fermi for their CUDA-enabled applications.
Below is a video from two of the researchers who wanted to share their experiences in their own words.
The first is from Dr. Inanc Senocak at Boise State University. His research focuses on simulating wind patterns to forecast wind-generated energy, and to determine how to efficiently supply wind energy to the power grid.
The second is from Prof. Esteban at Clua of Universidade Federal Fluminense in Brazil. His work centers on developing cosmological models of the movement of stars and galaxies as a means to better understand the origin of the universe.
Both researchers saw a tremendous benefit from the acceleration they achieved using Tesla K20 GPUs.
These are just two real-world examples of the power of GPU acceleration on Kepler. There are many other examples from the test drive, and you can access the full list here.
Take the Kepler Test Drive Yourself
We were so pleased with these results that we decided to continue the GPU Test Drive Program to give even more GPU users like you a chance to experience Kepler.
Would you like to see how much faster your CUDA code or GPU-accelerated application runs on Tesla K20 GPUs?
The GPU Test Drive Program is a free and easy way to find out how Kepler can work for you. You can sign up for the program at www.nvidia.com/GPUTestDrive.
We think you’ll be impressed with the power of Kepler.
If you do take the test drive, we’d love to hear about your experience. Please be sure to tell us about your results and experiences by emailing us GPUTestDrive@nvidia.com.
Tesla K20 Documentation:
Tesla K20 GPU Accelerator Board Specs (PDF 472 KB)
Tesla K20-K20X GPU Accelerators Benchmarks (PDF 504 KB)
Tesla K20 GPU Active Accelerator Board Specs (PDF 507 KB)