Published in PC Hardware

AMD rolling out new adaptive SoCs

by on23 April 2024

Versal AI Edge Series Gen 2

AMD’s rolling out a couple of new adaptive SoCs – the Versal AI Edge Series Gen 2 for AI-powered embedded systems and the Versal Prime Series Gen 2 for more traditional embedded systems.

Usually, multi-chip setups can be a bit of a headache with all their extra baggage, and a single hardware architecture doesn’t quite cut the mustard across all three AI stages – preprocessing, AI inference, and postprocessing.

To get around these headaches, AMD’s developed a single-chip heterogeneous processing solution that smooths out these steps and cranks up the performance.

The Versal AI Edge Series Gen 2 adaptive SoCs are all about giving AI-driven embedded systems a boost from start to finish, and AMD’s chuffed to bits about the improved safety and security it’s built on. They’ve thrown in a high-performance processing system with Arm CPUs, the latest AI Engines, and some programmable logic, making a device that handles the computational shenanigans needed in embedded AI applications.

AMD is chattering about how the Versal AI Edge Series Gen 2 SoCs are just the ticket for a whole range of embedded markets, especially those about security, reliability, long lifecycles, and safety-critical stuff. We’re talking about autonomous driving, industrial PCs, self-driving robots, edge AI boxes, and healthcare tech like ultrasound, endoscopy, and 3D imaging.

The integrated CPUs’ processing system is packing up to 8x Arm Cortex-A78AE application processors, up to 10x Arm Cortex-R52 real-time processors, and support for USB 3.2, DisplayPort 1.4, 10G Ethernet, PCIe Gen5, and a bit more.

These gadgets are up to snuff with ASIL D / SIL 3 operating requirements and tick the boxes for other safety and security standards. They offer up to three times the TOPS/watt for AI inference and up to ten times the scalar compute with beefy CPUs for postprocessing.

AMD’s Adaptive and Embedded Computing Group bigwig Salil Raje said the demand for AI-enabled embedded applications is going through the roof, and it’s pushing the need for solutions that lump together multiple compute engines on a single chip for the slickest end-to-end acceleration, all within the power and space limits of embedded systems.

The first silicon samples of the Versal Series Gen 2 are expected to pop up at the start of next year, with production set to kick off towards the end of 2025.

Last modified on 23 April 2024
Rate this item
(1 Vote)