Imagination Technology has unveiled a set of new GPU IP cores to enable smartphone performance of up to twenty times faster, using the company's low-powered graphics solution. Technology giants like Apple, Samsung and Texas Instruments are expected to be the chief beneficiaries of the new technology.
"You are now talking about, from a gigaflop point of view, supercomputing in hundreds of milliwatts," Imaginations' marketing VP, Tony King-Smith said to The Register, when speaking on the company's new designs.
Based on the new architecture, Rogue - a set of PowerVR Series6 GPU IP cores - was introduced at the 2012 Consumer Electronics Show (CES). The PowerVR G6200 and G6400 cores have retained the basic architecture of their predecessors and have evolved into heterogeneous CPU/GPU systems.
The new GPUs have been optimized for better results when mapping polygons, while also complementing CPU workloads with enhanced computing capabilities.
"We're talking about compute mattering more," said PR Director David Harold addressing The Register, "So the architecture we're defining is about these compute clusters - these arrays of compute elements that go together."
The G6200 is said to comprise two of these compute clusters, while the G6400 includes four. Two or four groups of clusters are said to be made of execution elements, while each execution cluster is an array of execution units, according to King-Smith on The Register.
The new GPU technology is focused at taking over CPU workloads during intensive computing tasks, as the GPU's benefit with their superior horsepower and "the raw density of processing per square millimeter and per milliwatt", according to King-Smith.
Besides, the entire processor coding has to be changed, as the instruction set for a GPU totally differs from that of a CPU and to take advantage of the parallel processing capabilities of the Rogue architecture.
Furthermore, Khronos Group's open standard, Open CL, could pave the way for better optimized algorithms that enhance parallel processing capabilities.
"Most of your code will actually stay in C on the CPU, [except] the heavy lifting stuff. If you're trying to optimize on the CPU, that's increasingly the wrong way to do it," said King-Smith, emphasizing the need for OpenCL APIs to transfer parallel tasks to the GPU.
Quite befittingly, Imagination has developed full-fledged OpenCL drivers for both the Series6 as well as the Series5 GPU IP cores. Series6 GPU IP cores are apparently designed to meet intensive computing demands besides regular graphics processing tasks.
Taking advantage of multiple cores, the GPU core speeds are expected to touch 100 Gigaflops and even breach the teraflops territory, according to Harold. He further suggested a performance gain in the range of "a couple of hundred gigaflops" may be achieved at mobile power-consumption levels.
"20x or more of the performance of current generation GPU cores targeting comparable markets ... enabled by an architecture that is around 5x more efficient than previous generations," read a release by the company.
The Rogue architecture is primarily aimed at power-draining smartphones, which could benefit with their smaller profile. However, the company has refrained from divulging any information regarding the prices, release date or its hardware design partners.