Knights Landing can deliver about 3 teraflops of peak performance, which is about the same as some high-performance GPUs in the world's fastest supercomputers today. The chip mixes conventional x86 CPUs with specialized vector processing units, differentiating it from regular x86 chips in PCs and servers.
But the Intel chip could soon lag behind the competition. Nvidia late last month introduced the Tesla P100 GPU based on the Pascal architecture, which offers 5 teraflops of double-precision performance and could ultimately outperform Knights Landing.
Knights Landing will be used in supercomputers like Cori, which will go live later this year at the National Energy Research Scientific Computing Center in Berkeley, California. An earlier Xeon Phi chip is already in China's Tianhe-2, which right now is the world's fastest computer, with peak performance of 54.9 petaflops.
Innovations in Knights Landing could ultimately come to mainstream laptops and desktops. It has 16GB of high-bandwidth MCDRAM stacked memory, which Intel claims has five times more bandwidth than DDR4 DRAM and is five times more power efficient and three times denser than GDDR5 on GPUs. The chip also supports the new OmniPath fabric, which Intel says will speed up communication inside a computer.
The desktops will include programming tools like Intel Parallel Studio XE Professional Edition.
Sign up for CIO Asia eNewsletters.