Intel is working in cooperation with universities to create a public suite of applications for multicore processors.

The semiconductor giant’s CTO Justin Rattner made the disclosure during his keynote address on Tuesday at the HotChips conference at Stanford University.  Such applications could be used as a benchmark for multicore chips.

To help with its own research on multicore, Intel has pieced together a suite of applications that demonstrate the breadth of computing functionality and performance that can be enabled by multicore processors. The so-called RMS suite – recognition, mining and synthesis – encompasses many applications within those functional areas.

The suite was created through some internal development at Intel, but also by using the code of some of Intel’s partners.

“There was interest in creating a public RMS suite,” said Rattner. “But there were problems with taking the suite public.”

The big problem came from the fact that Intel didn’t own the whole suite.  Intel developed a body tracking application and a ray tracking application and was happy to release that portion publicly. But many of the other apps in the RMS suite came from third parties, so Intel couldn’t offer them as part of the public suite.

To rectify the problem “we’ve gone around informally asking folks about contributing to an RMS suite,” said Rattner. Intel is contributing its body tracking and ray tracking applications.  And some universities have already pledged support. The University of Pittsburgh, for example, has contributed a cancer cell detection application.

Intel is working with Princeton University which will serve as a repository for the code contributions from the academic community. Professor Kai Li is spearheading the efforts there.

Rattner appealed to the HotChips audience at Stanford for contributions, too.

“We are here to ask you to contribute some code to this RMS suite,” he said.

The so-called RMS applications are expected to make up much of the computing workloads of the future.  Rattner broke them down, providing demonstrations for the crowd packed into Stanford’s Memorial Auditorium.

Recognition, he said, can include modeling and identifying by using multimodal data.  Rattner showed a demonstration of a speech to text application that relied on both voice recognition and lip recognition to capture and translate the words. 

“The performance of the lip tracking is almost as good as the audio piece,” he said. “But the signal to noise ration dramatically improves when you use both.”

A mining demonstration showed an image-based search that relied on textures and content. But even today, indexing the volume of data – in this case 10,000 photographs, took about 20 hours. The search, however, took only seconds. 

Synthesis, said Rattner, puts these models in motion to answer “what if” questions.  For example, Intel’s ray tracking application renders lights and shadows in graphical settings with photo realism.  While the application has not yet achieved real time rendering, that is on the way, said Rattner. Such software will be sought after by video game makers.

Together, this public RMS suite will pave the way for creating a multicore benchmark by establishing a set of emerging workloads.

“The benchmark needs to look at performance and energy,” said Rattner. “with applications that are highly-threaded and scalable, and they should cover the entire realm from visualization to financial and everything in between. We want applications that push the technology to its limits.”

AMD, Intel and others have pointed out the upcoming challenges of creating software that can make full use of multicore hardware to offer greater performance or reduce energy use.

Intel provided a peek into some of its research efforts around development for multicore during a briefing before its developers forum in March.