Intel on Monday announced a new brawny HPC processor and a family of network fabric components that incorporates silicon photonics technology. The new chip, the next-generation of the Xeon Phi family, comes with up to 16 gigabytes of high-performance memory and more than 60 computing cores. The fabric lineup, called Omni Scale will include PCIe adapters, switches and software, as well as director switches that replace electrical transceivers with silicon photonics for improved speed and fewer cables. Intel is promising better performance and greater efficiency with the new tech — something it might need considering the introduction of ARM-plus-GPU-based chips into the HPC market that Intel presently dominates.
Nvidia has teamed up with ARM to offer a combo of graphics processors and ARM-based CPUs for the high performance computing market. But is ARM ready to take on HPC?
Intel has launched its latest top-of-the-line chips for servers. These are the bruisers that make up the silicon in high performance computing and super fast financial transactions. The Xeon E7v2 class of chips features up to 15 cores, a massive amount of in-memory data capacity to make processing large amounts of data on chip possible, and performance that’s twice the average of the previous generation of chips. These processors hold a few surprises, as the Register details in its in-depth exploration of the silicon and the business case.
Supercomputers have graduated from linear algebra to differential equations, and so the guys who rank the world’s top machines have decided to change the benchmark that determines which computer is on top.
A highly dense memory technology introduced in 2011 takes another step closer to reality with the launch of new interconnection specifications. At this rate, we’ll see the new tech in devices in 2014.
The largest driver of this growth is the need for complex and compute-intensive data analysis and simulation jobs that require these types of specialized platforms. While the cost is typically high for traditional HPC approaches, many businesses are seeking and finding cloud alternatives.
Despite the idea that a server is a server, the needs of different computing customers differ widely. For those thinking about selling infrastructure, software or even services understanding the difference in computing and IT styles will help you hone your pitch and find your buyer.
In the near term companies are eking more performance out of chips without using too much power by using co-processors taking a cue from the supercomputer industry. In the long term, though making faster and more efficient chips will take place at the subatomic scale.
Morgridge Institute ran Cycle Computing software atop Amazon Web Services to build a cell knowledge base that, over time, could help doctors build the tissue types they need to resarch and cure disease in a petri dish.
Working with Schrödinger, which specializes in computational drug design, Cycle Computing built a 50k-core AWS cluster that screened 21 million compounds in less than three hours. The cluster enabled the company to use a much more accurate screening process than other technology.