NVIDIA is the leader in high-performance computing accelerators, but Intel is quickly becoming a credible force in this market.
Sourced through Scoop.it from: www.fool.com
Although NVIDIA’s high-performance graphics processors are generally designed for gaming, these chips can also be used for computationally intensive applications outside of games. For years, NVIDIA has sold graphics processors under the “Tesla” brand into supercomputers, servers, and professional workstations to accelerate certain compute-intensive workloads. During NVIDIA’s fiscal year 2015, the company’s “HPC and Cloud” business — which probably consists primarily of sales of Tesla accelerators — was $279 million, growing about 52% year over year. This is an impressive growth rate, albeit off of a fairly small base. However, despite the fact that Tesla has been a success story for NVIDIA, there is one potential threat to this business that NVIDIA investors should keep a close eye on — Intel‘s Xeon Phi. Which is designed to handle very computationally intensive workloads in a faster, more efficient manner than traditional CPUs. Although NVIDIA’s Tesla is essentially a graphics processor being used for non-graphics computational tasks, the Xeon Phi is made up of a bunch of relatively low-performance general purpose processors coupled to high-performance vector processors suited for computationally intensive tasks. Intel’s first commercially available generation of Xeon Phi accelerators, code-named Knights Corner, generally didn’t deliver the kinds of performance that competing Tesla accelerators did. Further, although Intel claims that Xeon Phi’s compatibility with the traditional x86 instruction set is a valuable asset in terms of attracting developers, NVIDIA’s incumbent position in the accelerator market probably dulls that potential advantage for Intel. Intel, however, might be set to gain an edge on NVIDIA with its upcoming Knights Landing processors for high-performance computing. Intel’s next generation Knights Landing promises major performance improvements, comes in socketed CPU form, is expected to launch later this year. The chip promises to bring substantial performance improvements over the prior generation Knights Corner due to a combination of architectural enhancements and a move to next generation 14-nanometer manufacturing technology. Intel says that Knights Landing should be able to perform six trillion single precision floating point calculations per second or three trillion double-precision floating point calculations per second. In contrast, NVIDIA’s top single-chip Tesla GPU, known as the Tesla K40, can do about 4.29 trillion single precision floating point calculations per second or 1.43 trillion double precision floating point calculations per second. Another interesting thing about the upcoming Knights Landing chip is that Intel will be offering it as a stand-alone CPU as well as a co-processor connected over the PCI Express bus. The stand-alone CPU version, according to Intel, should deliver more than 25% better power efficiency relative to the add-in-board variant.