San Diego, Ca. — Murthy Renduchintala jokes abouthow hard it is to pronounce his surname, butthere's no mistaking the message of theexecutive vice president of QualcommTechnologies Inc.
Like six-foot-something Renduchintala,Qualcomm sees itself as a tall, capablepresence that goes its own way in mobiletechnology. It expects to continuedeveloping custom CPU, graphics, DSP, andother cores and combining them in uniqueways in its SoCs.
“I don't care who copies Qualcomm, butQualcomm will never copy anyone,”Renduchintala said in a press Q&A at theUplinqevent here. “The ARM A15 core is good, butit's not good enough for Qualcomm, so wehave to enhance it.”
Both Samsung and Mediatek claim theirlatest eight-core SoCs beat Qualcomm'squad-core Snapdragon in benchmarks. The tworivals both use ARM's off-the-shelf A15 andA7 cores combined using ARM's Big.littlemulticore architecture.
“What's important is the aggregate userexperience rather than superficially thenumber of CPUs,” said Renduchintala. “Youcan use benchmarks to show whatever youwant, but if you take a broad panorama ofbenchmarks, Snapdragon is still the besttechnology out there.”
But Snapdragon is getting a bit long in thetooth in the fast-paced silicon world. Itslatest members were announced late lastyear, demonstrated at CES in January, andare now emerging in smartphones and tablets.Qualcomm will probably announce follow-onsbefore the end of the year.
Meanwhile Renduchintala downplays CPU corecount as a metric for the quality of amobile SoC.
“We could integrate 12 or 24 cores bytiling identical CPUs together. Thecomplexity is not in laying down the siliconbut in the software to make use of that manycores — real-time software processing with8 to 10 parallel tasks is pretty complicatedand more trouble than it's worth.”
Qualcomm claims it gains advantage bycombining “the right number” ofbest-of-breed CPU, GPU, DSP, imagingprocessing, cellular, and other coresoptimized for performance per watt.Renduchintala, raised in England, comparesit to an 11-man soccer team with the rightmix of players.
He dismisses ARM's Big.little approach ofganging performance and power optimizedcores into an SoC. “It's one way ofaddressing the challenge, but we have adifferent idea” he said, adding that the ARMapproach of multitasking between big andlittle cores implies that neither is welloptimized for the task at hand.
To read more, go to “Of LTE and NIH syndrome.”