Three computer makers have decided to combine Nvidia graphics processors with ARM-based CPU cores for high performance computing — a first for the ARM architecture that has so far dominated the cell phone market. Cirrascale (pictured above), E4 and Eurotech are all building machines that will use the Applied Micro X-Gene boards in conjunction with a nearby GPU to handle the types of performance-heavy workloads popular in the oil and gas, scientific and industrial design industries.
Nvidia has opened up its CUDA code that lets developers compile their code to run on GPUs as a means to get ARM cores into the HPC sector. The combination of a GPU and an Intel x86 processor has become more common in supercomputing, in part because GPUs can do more work per watt than an x86 processor. As supercomputers grow in performance, they are also sucking up far too much energy, leading experts in the field to become worried that the next generation of machines will require too much power to be feasible.
And since ARM cores are as low power as they come, the idea is that replacing the Intel or AMD cores using the x86 architecture with something using ARM, would help cut down on power consumption. While it may seem far-fetched, the rise of 64-bit capable ARM processors hitting the market this year means that the HPC market could become a possibility. People in the HPC world are assessing ARM already.
Pat McCormick, senior scientist at Los Alamos National Laboratory was quoted in the release announcing the creation of ARM-based machines as saying, “We are working with Nvidia to explore how we can unite GPU acceleration with novel technologies like ARM to drive new levels of scientific discovery and innovation.”
Bringing an ARM-Nvidia duo into the HPC market isn’t as crazy as one might think. Nvidia has been pushing hard in a variety of computing realms, from cell phones to enterprise as a way to broaden its market and take on computing jobs that the x86 architecture can’t do as efficiently. Like ARM, it is hoping to ride the trend of workload-specific hardware into more markets that were once owned by Intel’s x86 architecture. It only makes sense that the two architectures would team up to take on a lucrative sector that is currently limited by power consumption.