ARM unveils Cortex-A76 and Mali-G76 - higher performance, better power efficiency
Each year new flagships come out with new chipsets, which in turn require new cores. To that end, ARM unveiled a new big core – the Cortex-A76. And there’s a new GPU too, the Mali-G76, along with a new video processor for the 8K future.
Compared to a current Cortex-A75 (10 nm) running at 2.8 GHz, an A76 (7 nm) running at 3.0 GHz promises 35% higher performance and 40% better power efficiency. And since AI is trendy, the new CPU has nearly 4x improvement in speed when it comes to machine learning. Memory bandwidth is nearly double too.+
ARM is calling these “laptop-class” cores and the earliest designs will pair one or two Cortex-A76 with a number of A55 cores. This doesn’t mean that the A75 is going away, though, it sounds like the A76 is a large core and smaller, cheaper chipsets will still use the A75.
The Mali-G76 is the latest GPU from the Bifrost family and when built on a 7 nm it promises a 50% boost in performance compared to a G72. And 30% better energy efficiency with an improvement in machine learning too (2.7x).
Finally, there’s the Mali-V76 video processor. It can encode 8K video at 30 fps, when mobile cameras and storage catch up. And ARM is promising a 25% enhancement in video quality too thanks to algorithm improvements. The processor can decode 8K video at 60 fps or 4K at 120 fps.
Related
Reader comments
- TwelveMoon
- 24 Jun 2018
- thw
A57? Did you mean the successor of the A55 efficiency core? A57 is used long ago as a high-performance core put in big cluster, and if things gone wrong, it will overheat badly. Put 8 of those A57's, overclock at a constant speed of 2.5 GHz each, and...
- Vegetaholic
- 04 Jun 2018
- sXj
Looks good, we need better graphics on today's phones, looks like graphic progress stagnate on mobile phones, welcome improvement, maybe it's mean we will get GTA 4 port soon :)
- Kangal
- 03 Jun 2018
- uCX
You both are wrong. Its not about bandwidth, nor about per-core performance. It's about balance. You need to have large cache to store memory code, then you need good per-core performance to have high-chance of correct rasterisation each cycle...