At both ISC High Performance and TERATEC Forum, I was struck by the convergence between HPC and deep learning that we saw and heard about. It also seems like there’s been a variety of new entrants into the HPC market with bigger and newer hardware.
To explain the growth, I will start with the market leader: Intel. Intel recently released their new KNL platform. This platform is the next-generation of many-core Xeon Phi architecture, combining the best features of Xeon and Xeon Phi architecture in one platform, making porting much easier. Placing many-core chips on the bootable node also opens new possibilities of scaling to much greater parallelism.
The next big announcement was around China’s newly deployed “Sunway” machine. It’s a 10 million core, 93 petaflop machine, and is currently the largest supercomputer in the world. When we add in that Fujitsu recently announced that their new post-K Japanese machine will be ARM-based, a recent entrant into the HPC marketplace, it’s easy to see that the whole industry has recently become a lot more competitive and that major players see large growth.
When you add in recent performance increases from NVIDIA and IBM (OpenPOWER), the competition is very strong with all the major players showing off their advantages.
Most of these silicon vendors aren’t just after the typical HPC targets of better and faster computational analysis, but also the ability to do deep learning on big data. This was a big topic of discussion given the rate of growth of new data and how unstructured it is. Overall this is a big growth area for HPC and is causing much of the expansion in the HPC industry.
This is why supporting dynamic analysis and debugging of HPC machines is a very important part of our portfolio and we keep adding new capabilities. To learn about the dynamic analysis products from Rogue Wave Software, visit – TotalView for HPC and CodeDynamics.