Power8 Before It’s Too Late
May 5, 2014 Dan Burger
Now that Power8 has broken out of the gate, the eyes of the Unix world will turned to see if IBM can rein in the challenge Intel has created in this enterprise computing horse race. The quarter-by-quarter decline in Power Systems revenue has not been a pleasant ride for IBM’s AIX side of Big Blue. We’ve often heard, “What’s good for Power Systems is good for IBM i.” That’s why we’re wishing Power8 the best.
IBM, as it introduced the first five Power8 boxes at the Impact2014 conference, also plugged into a social media audience with a promotional video that left no doubt that X86 was in the crosshairs of Power8. Comparisons clearly put Power8 in the role of the most powerful, the most trusted, and the most capable system when it comes to big data, analytics, and the cloud.
Doug Balog, general manager of the Power Systems division, was quick to set the tone by establishing infrastructure as the focal point and then loading the presentation with evidence that IBM will continue to be the king of infrastructure, while X86 is still a PC architecture.
Balog touched a many competitive advantages for Power, but singled out advancements in multithreading, memory close to the processor, and I/O bandwidth as particularly meaningful factors when considering what systems are best suited to handle the 800 percent of data growth that’s forecasted to spill into data centers in the next five years. And, by the way, 80 percent of the data avalanche is unstructured data, a mostly untapped gold mine, according to big data miners.
“Companies are out of floor space and out of energy, yet the data keeps growing,” Balog said, while claiming 65 percent sustained system utilization from Power Systems compared to less than 25 percent sustained system utilization being the norm in X86 server farms. He also dished up some availability comparisons, saying Power Systems experienced 73 percent fewer outages and 92 percent fewer performance problems, to make his point about simplified infrastructure management. That pitch works well in conjunction with Power8’s bigger, more scalable workloads and enterprise-strength memory.
During the video, Balog calls a few of his Power colleagues to the witness stand. The first was Rob High, VP and CTO of the Watson Group. Watson and the allure of cognitive computing is headline-grabbing material for Power and its message of next-generation computing. And to attach this promise to something real, IBM brought in MD Buyline, an ISV that is applying Watson evidence-based technology to the selection and buying of medical equipment so its customers can factor cost, quality and outcome data into their purchasing decisions.
To accomplish these kind Watson-esque data-intensive workloads, High emphasized “Power offers over five times the memory bandwidth of commodity systems.” I think we all know who’s wearing the ugly commodity systems shirt in this picture.
To elaborate on the unleashed performance in Power8, Balog brought in Bob Picciano, senior VP of the information and analytics group. Picciano’s the speed-of-thought analytics guy who sees a mountain of data and his first thought is: How fast can it be analyzed? When he talks about threads per core compared to the other guys, he notes Power8 has four times the number of threads per core than X86 and 170 percent more bandwidth.
Arvind Krishna, general manager of development and manufacturing for Systems and Technology Group, carried another batch of positive Power stats that reiterated the four times as many threads per core and the 170 percent more memory bandwidth that Picciano brought up. (Just in case you forgot.) Krishna also added that Power8 delivered 30 percent more memory capacity than X86-based servers.
His message included a tip of the cap to the OpenPower Foundation, which contributed to the Power8 war chest. Krishna then took the time to describe CAPI, the Coherent Accelerator Processor Interface that allows accelerators such as GPUs and flash memory to connect directly to the processor sharing the same address space. The result is improved performance and reduced latency.
CAPI was created by IBM and is being enhanced through the OpenPower Foundation in a cooperative effort between IBM and Nvidia. That partnership has also turned out Java accelerators on a GPU that optimize the placement of algorithms used for analytics. Ian Buck, vice president for accelerated computing at Nvidia, also participated in the video by praising the match up of Nvidia GPUs with Power8 CPUs.
If you’d like to watch the video, it can be seen at IBM’s Digital Event Center.