In advance of the official opening of Computex 2017, ARM has unveiled its latest chip designs which will help fuel a countless number of our devices in the near-future. Those include the Cortex-A75 and A55 CPUs, as well as the Mali-G72 GPU.
Performance improvements are nice, but the most notable aspect of the new chip designs is the introduction of DynamIQ, an architecture that will give manufacturers far more flexibility in their chip designs. Whereas past big.LITTLE designs required two different processors to be treated as separate, DynamIQ will allow unusual configurations like 1+7 or 2+6 (cores) to be possible.
What makes DynamIQ truly important is that it gives manufacturers the ability to create chips with a variety of different configurations, resulting in finer-tuned designs for specific purposes. The flexibility is granular enough to allow each core to run with different specs and speeds, and improving things on the whole, all eight cores will be put in the same cluster, allowing cache to be shared between different cores.
As was highlighted in a very big way a few weeks ago at NVIDIA’s GPU Technology Conference, artificial intelligence is becoming more important than ever, and ARM clearly agrees. Upcoming Cortex processors are going to include brand-new instructions to accelerate AI performance by up to a promised 50x over the next 3-5 years. Performance like this will mean that not all of our AI processing will have to take place in the cloud; instead, our devices will do the heavy-lifting, improving response times and efficiency overall.
If you don’t think AI applies to you, you’re probably already taking advantage of it and don’t even realize. If you use Google Translate to decipher foreign language text on a bottle of an unknown substance, for example, the performance might not seem so hot. Move all of that processing to your phone, however, and scenarios like these could deliver results that are an order of magnitude faster.
On the topic of the new CPUs, architecture improvements could see performance boosts of up to 50% in some cases, most notably with floating-point operations, and with ARM’s own NEON SIMD processing. Compared to the Cortex-A73, the -A75 piles on an additional 15% memory bandwidth.
As for the Cortex-A55, it will also deliver big performance improvements, including nearly a 100% gain in memory bandwidth, as well as an 18% boost for integer and 38% for floating-point performance. For graphics duties, the Mali-G72 promises a performance bump of up to 40%, improving not just gaming performance, but also performance for VR and machine learning.
The major improvements to hit ARM’s processor designs come at a great time, as the company expects that its partners will ship over 100 billion chips over the next 5 years. Sooner than later, we’re all going to have a device capable of impressive AI inference and generally much-improved performance – and that’s downright exciting.