ARM takes VR/AR mobile with GPU core

New Products |
By Peter Clarke

The core is intended to enable demanding graphics use cases such as virtual reality and augmented reality and 4K displays on 2017 mobile equipment by providing a 50 percent performance improvement over the company’s current leading GPU, the T880. But it is also expected to find deployment in large-screen devices such as 4K TVs. The core has been launched alongside the Cortex-A73, formerly known as Artemis, as a premium suite of cores for next generation smartphones.

The graphics performance and efficiency uplift is enabled by adoption of the Bifrost architecture – following on from the Utgard and the Midgard architectures that have been behind all of ARM’s Mali GPUs to date (see ARM’s Bifrost steps up graphics, bridges to machine learning).

ARM’s Mali series of GPU cores has become the leading licensed GPU. Source: ARM.

The Mali-G71 introduces a number of innovations to improve raw graphics performance and to improve computational efficiency and the ability to operate within a heterogeneous computing environment. In such mobile devices all the computing resources must work within a common and increasingly demanding thermal envelope ARM executives stressed during briefings held in Austin, Texas.

For example the Mali G series and Bifrost scale up to 32 shader cores compared with 16 shader core maximum in the Midgard architecture as exemplified by the highest-performing Mali-T880

The Mali-G71 is also the first GPU from ARM to comply with the Vulkan 1.0 API specification released by the Khronos Group in February 2016. Vulkan provides a lighter overhead cross-platform access to graphics and compute on GPUs compared with Open GL ES.

“Mali-G71 delivers the premium VR experience,” said Espen Oybo, GPU product manager for ARM and which he benchmarked as 120Hz refresh rates on 4K resolution screens with 4ms pipeline latency. And as a comparison this offers 20 percent higher computational efficiency and 40 percent better performance density compared with the Mali-T880 if implemented on the same process node and under the same conditions.

The addition of system-wide hardware coherency with fine-grained bufferng reduces memory overhead. Source: ARM

Among the innovations included in Bifrost and G71 to help achieve these results are: index-driven position shading; something called claused shaders; quad-thread fragment processing and support of full hardware coherence in caching of memory. A quad is a 2 by 2 pixel group which provides 4 threads.

This support of full coherence allows the Mali-G71 to sit alongside a set of processor cores – ARM would hope Cortex-A series processor cores – with a common memory system coordinated by the Corelink CCI-550 interconnect.

Like the Midgard architecture and GPU circuits based on it Bifrost and Mali-G71 support the OpenCL 2.0 programming language and its definition of shared virtual memory. However, Mali-G71 also supports the extra option of fine-grained buffers. The use of full hardware cache coherency is a foundation for heterogeneous computing. It reduces the cost of sharing data, ARM said.

When asked if ARM had made particular provision for supporting general-purpose computing on the GPU – so-called GPU-compute – Oybo said: “The first and foremost priority is graphics performance. We do look at GPU-compute use cases as well. 

The core is generally being aimed at the 10nm manufacturing production node and HiSilicon, MediaTek and Samsung Electronics have taken licenses on Mali-G71. Commercial ICs including the GPU core are expected late in 2016 for inclusion in devices in 2017.

“Next-generation premium experiences will be defined by pushing the boundaries of what’s possible with mobile VR and AR,” said Jae Cheol Son, senior vice president, processor development team, Samsung Electronics, in a statement issued by ARM. “A scalable GPU like the Mali-G71 will help Samsung design teams address increasingly complex mobile VR and AR use cases.”

While the Mali-G71 is able to address demanding use cases in mobile it also allows trade offs of die area and power consumption so that the same basic architecture can be used for other large-screen compute devices such as industrial gateways, in vehicle infotainment and smart TVs.

Related links and articles:

News articles:

ARM buys Apical for $350 million

Samsung FinFET Exynos 8 has custom cores

Semi IP licensing growth slows, vendor ranking stabilizes


Linked Articles
eeNews Europe