AMD combines five compute architectures on one board • The Register
With the launch of the Embedded+ architecture yesterday, AMD effectively asked the question: why choose one compute architecture when you can have five?
House of Zen’s latest offering combines a Ryzen x64 processor with a Versal AI Edge system-on-chip via PCIe so it can be used on a single board in low-power, low-latency data processing applications on the network, for example. edge.
AMD’s Embedded+ technology sticks a Ryzen Embedded x64 processor with Versal Adaptive on-chip onto a single board via PCIe… Click to enlarge. Source: AMD
The flagship processor can be chosen from the Ryzen Embedded R2000 family, which was launched in 2022 and has up to four Zen+ CPU cores, 16 PCIe 3.0 lanes, and up to eight Radeon Vega graphics compute units.
This chip has a dedicated PCIe connector for the AMD Versal Adaptive SoC, the first of which debuted in 2021. These Versal parts have a complementary set of AI engines, an FPGA, and four Arm-designed CPU cores — two Cortex-A72 and two Cortex-R5. In terms of ML processing, AMD claims that its top Versal chips are capable of pushing around 228 TOPS at INT8.
As the name Embedded+ indicates, this kind of thing is meant to be used in devices that are built to last in relatively tough conditions — general displays, devices and machines in the field, network edge processing, transportation and automotive, etc. It doesn’t have to be very sophisticated and powerful; Reliability, cost and energy efficiency versus performance, footprint and verification of the specific workload are often more important. Therefore, it is expected that older architectures will be used for these chips.
In fact, AMD has its sights specifically on industrial robotics, retail security and surveillance, smart city equipment, networking, machine vision, and medical imaging; Its customers will decide if the device has the latency, traction and processing pipeline for their applications.
“In automated systems, the value of sensor data diminishes over time and must operate on the latest information possible to enable deterministic response with minimal response time. In industrial and medical applications, many decisions must be made in milliseconds,” Chetan Khona, senior director, AMD division The vision for the industrial, healthcare and science markets poured in in a statement.
In order to achieve these latency goals, AMD encourages developers to break their workloads into smaller chunks that can be individually accelerated by the platform’s various compute architectures. For example, the Adaptive SoC’s FPGA and AI engines can be used to pre-process and classify streaming data from multiple sensors or feeds, while the Ryzen processor’s CPU and GPU run the control systems and GUI.
Of course, this always happens in mixed-core systems, and AMD isn’t the first to put a mix of architectures on a single board or even in a single chip. This is clear. What’s interesting here is that AMD is doing this not only with the Ryzen and Versal families but also with a strong focus on AI in the embedded and network edge, which it wouldn’t do if people didn’t want it. Ideally.
Sapphire’s Edge+ VPR-4616-MB features a Ryzen Embedded R2314 processor with Versal AI Edge VE2302 Adaptive SoC in a mini-ITX form factor – click to enlarge
Among the first systems to build on AMD’s Embedded+ design is Sapphire’s creatively named Edge+ VPR-4616-MB. This connects a quad-core Ryzen Embedded R2314 processor to a Versal AI Edge VE2302 Adaptive SoC on a mini-ITX board that is said to consume up to 30W. Sapphire also plans to offer the motherboard as a fully assembled PC with memory, storage, PSU and chassis. ®