The folks at AMD are really starting to rock-and-roll in embedded space. Historically, 90% to 95% of their revenue has come from the PC arena (they play in both consumer and enterprise-level markets), but about two years ago they started to diversify.
While continuing to evolve in their core markets -- maintaining their traditional PC strongholds and growing in the desktop and graphics channels -- they are also ferociously attacking growth markets, including professional graphics, ultra-low-power clients, dense servers, and many areas of embedded space. Some example application areas are as follows:
- Gaming machines: casino gaming; arcade gaming; over-the-top set-top box gaming
- Digital signage: retail signage; PSO/kiosk; quick-service restaurants
- Medical imaging: portable medical equipment; clinical workstations; MRI, CT, XRay
- Industrial control & automation: industrial HMI/PC; machine vision; surveillance; smart servers/gateways
- Thin clients: financial; education; converged devices
- Communications infrastructure: enterprise/switch/router; service provider networking; IoT infrastructure
As part of this, AMD just announced the second generation of their R-Series accelerated processing units (APUs). Code-named "Bald Eagle," these little beauties boast two or four "Steamroller" central processing unit (CPU) cores along with a new graphics processing unit (GPU) core based on the AMD Radeon HD 9000 platform.
These 28nm APUs offer next-generation performance-per-watt compute efficiency in the x86 product category by allowing system designers to take advantage of a Heterogeneous System Architecture (HSA).
A vast number of embedded applications involve a mixture of serial and parallel algorithms. CPUs are inherently serial in nature, while GPUs are massively parallel compute engines. The idea behind HSA is that work is dispatched to the appropriate cores based on the type of tasks involved: serial tasks are dispatched to the CPUs, while parallel tasks are dispatched to the GPU.
A key aspect of HSA is heterogeneous Uniform Memory Access (hUMA), which is a shared-memory subsystem that provides both the CPUs and the GPU with peer access to the main system memory space. As an example of the peer relationship between the CPUs and the GPU, the GPU can act as the master and initiate tasks on the CPUs (traditionally, GPUs have adopted a subservient role to CPUs).
If the computational and graphics performance of R-Series APUs live up to AMD's claims, then these devices could be game changers. For more information, visit the AMD Website.
— Max Maxfield, Editor of All Things Fun & Interesting