Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
EE Times-Asia > Processors/DSPs

Interconnected MIPS, GPUs on Imagination's horizon

Posted: 04 Nov 2014 ?? ?Print Version ?Bookmark and Share

Keywords:graphics processors? technology? Imagination Technologies? Nvidia?

Imagination Technologies wants to take on Nvidia with a new technology that will connect its multi-core MIPS processors with large arrays of its PowerVR graphics processors (GPUs).

This would allow its IP customers to create chips that would compete directly with Nvidia's Kepler and Maxwell chips for higher-performance computing. (See: High-end GPUs drive up Nvidia's revenue)

Alongside its i6400 Warrior core, Imagination is developing the coherent interconnect technology to link hundreds of MIPS and PowerVR cores and is happy to support customers with such developments, said Jim Whittaker, executive vice president of the MIPS division, speaking at the recent Multicore Challenge conference in the United Kingdom.

The key difference is that Imagination is an IP supplier rather than a chip company, so it will support any chip customers that want to create a cluster chip. "Nvidia is a very different company, and we are not going to spend all our time on the high end, but if a customer wants to go there we are certainly able to help them," he said.

"As we go to multi-core we have the option to increase the number of threads, the number of cores in a cluster up to six, and we can have multiple clusters so it gives you lots of dimensions to scale up the compute resources, and we are working on a fabric that can bring all this together. The interesting thing is you can connect together all these resources for extremely powerful compute clusters, and what we will be supporting is fully coherent operation between CPUs and GPUs. So multi-core is not just about putting down multiple cores but multiple threads and not just in CPUs."

As Nvidia found, the development environment around such a chip is vitally important, and Imagination is looking to the OpenCL programming environment for this. "OpenCL on GPU cores does provide a very well defined and standardised programming model for massively parallel systems," said Whittaker.

Nvidia's Kepler K1 chip contains 192 graphics cores alongside four ARM Cortex-A15 processor cores while the successor, codenamed Erista, will use the newer Maxwell GPU architecture to provide twice the performance and is scheduled for release in 2015.

- Nick Flaherty
??EE Times

Article Comments - Interconnected MIPS, GPUs on Imagina...
*? You can enter [0] more charecters.
*Verify code:


Visit Asia Webinars to learn about the latest in technology and get practical design tips.

Back to Top