ARM adds Ethos cores to machine learning catalogue
The N77 delivers up to 4TOPS of performance from 2048 8bit MAC units, scaling to 100s of TOPs in multicore deployments. It has up to 225 percent convolution performance uplift using the Winograd Fourier transform algorithm on 3×3 kernels, delivering up to 90 percent MAC utilization. The efficiency is up to 5TOPs per watt when using distributed internal SRAM, storing data close to the compute elements to save power and reduce DRAM access.
Target applications are computational photograpy, premium smartphones, AR/VR.
The Ethos-N57 is half the size and delivers up to 2TOPS using 1024 8bit MACs again with up to 90 percent utilization under the WFT algorithm. Typical applications would be mainstream smartphones, smart home hubs.
The Ethos-N37 is half the size again and delivers up to 1TOPS from 512 8bit MAC units within a square millimeter of silicon. This might be suitable for voice-command systems, real-time translation in digital TV and other systems.
Related links and articles:
News articles:
ARM launches two machine learning processors
Thoughts on Jem Davies leading ARM’s machine learning group
How To Do Machine Learning on Arm Cortex-M Microcontrollers
ResNet-50 – a misleading machine learning inference benchmark for megapixel images
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
