Monday morning in Taipei and the public holiday for the Dragon Boat Festival doesn’t slow the momentum of the technology industry. Computex opens tomorrow but first we’ve gathered the best and brightest of Asia’s tech industry to announce the release of Arm’s latest suite of premium mobile products, including a brand new High Performance GPU, Mali-G72. The Taiwan International Conference Centre plays host this morning as we gather in the cool of the air conditioning with the sun beating down outside. First up on stage is Nandan Nayampally, GM & VP of Arm’s Compute Processing Group to introduce today’s announcements.
He started by recapping last year’s launch and the emergence of those products in this year’s latest premium mobile and large screen compute devices, ‘accelerating the paradigm shift we need for an increasingly complex world’. Nandan discussed the requirement for ever more sensing and compute capability, and emphasized that this was not just general compute, but total compute, with the Arm platform well-equipped to deliver. The strengths of both Arm’s products and partnerships is evidenced by the fact that it took 22 years to ship the first 50 billion Arm-based chips but just four years to ship the second 50 billion. With a prediction of another 100 billion Arm based chips in the next 5 years, it’s vital we can support this industry acceleration.
Nandan emphasized the industry’s shift to distributed intelligence, moving greater capability from the cloud to the edge device, or ‘putting the power in the palm of your hand’. This, however, is no simple task. You need security and increased privacy for control over your content, requiring more single threaded performance. As we know, the mobile power envelope isn’t growing so we have to constantly adapt to do more with less in order to provide sustained and efficient performance and deliver this intelligence to scale.
Nandan then reminded us of the release of DynamIQ in March, designed from the ground up to support the needs of AI. DynamIQ provides low latency, acceleration connections and improvements in the CPU for the handling of AI workloads. With this fresh in our minds, it was time to announce the first processors launched on DynamIQ technology: Cortex-A75 and Cortex-A55. Cortex-A75 delivers a substantial uplift for performance, with over 50% improvement in raw integer performance alone, and even greater gains for specialized workloads. Cortex-A75 provides this performance across a broad range of market uses from mobile and large screen compute, to infrastructure and safe automotive, for use in autonomous vehicles, for example.
Cortex-A55 provides the LITTLE to Cortex-A75’s big, with 2.5x the power efficiency of its predecessor, bringing performance to the edge across a variety of applications and performance points from IOT edge gateways to mainstream, everyday devices.
Combining the two new CPUs in a DynamIQ big.LITTLE cluster gives you multiple configurations, and specifically doubles performance in a 1+7 configuration.
Of course, at Arm, we never work in silo, we always aim to adopt a system level approach for distributed intelligence, and this leads seamlessly to my personal Arm technology baby, the GPU.
So with that, Nandan handed over to the Media Processing Group’s very own VP, GM and Fellow, Jem Davies.
Jem began by focussing in on the mobile market in particular, where premium products and complex graphics use cases are driving the development of our High Performance GPU roadmap.
So who is better equipped to cope with these challenges than the provider of the number 1 shipping GPU? 1 billion Mali-based chips were shipped last year and ‘while that’s not quite up to CPU numbers, it’s pretty exciting’. It doesn’t end there though, and that’s why Jem was here to announce the launch of Mali-G72, taking the Bifrost architecture, launched in 2016, to new heights with 1.4x the performance of previous generation devices.
With the growing visibility of Machine Learning (ML), increasing demand continues to be placed on the GPU for compute capability to perform ML locally, on device, wherever possible. Jem explained that the clever bods at Arm ‘saw Machine Learning coming’ but even so, uptake has been significant so we’ve taken the existing ML performance of last year’s Mali-G71 even further with the Mali-G72. When questioned after the presentation, Jem explained that this ML uplift has been facilitated by ALU and data path improvements, amongst others, to reduce energy consumed by data movements.
It’s not just ML adding complexity though, VR continues to be a huge industry focus and we’re seeing a massive uplift in High Fidelity Mobile Gaming, bringing console quality visuals (and therefore graphical complexity), to mobile gamers. In China, this level of gaming is growing twice as fast as standard mobile gaming and the reason for this, as Jem put it, is: ‘Because it can. The graphics technology we provide is performant enough to support it’. This means we’re seeing our partners implementing bigger, more capable configurations to give the end user the best possible experience. As we know, bigger and better SOCs use more silicon, which is why the 20% area reduction of Mali-G72 will be welcome news to our partners, bringing significant cost savings to next year’s flagship devices.
Of course, that’s not all the improvements of the Mali-G72, indeed, we’ll see further innovations as our partners begin their implementations, but for now you can head over to my launch blog for all the techy detail.
With the Mali-G72 successfully announced, Jem returned to the system story to remind us that it’s not all about the big-name CPUs and GPUs. Bringing together the next generation of Premium Mobile IP takes other clever tech too. Whether it’s CryptoCell-712 IP Interconnect, Physical IP libraries enabling partners to get to market faster, or the ‘nattily titled’ SGM-775 (System guidance for Mobile) available free to partners for ‘smarter, more efficient devices,’ there’s no end to the cleverness happening in next-gen SOCs.
It’s great to see how we’re continuing to evolve our approach and are focused on complete solutions for distributed intelligence, visual computing and raising the bar on performance, all within the mobile power envelope.
For now, stay tuned and tomorrow I’ll be hitting the show floor to bring you all the excitement from Computex 2017!