Richland is here. The successor to AMD’s Trinity APU has been shipping to PC makers since January, and today, we can tell you more about what it has in store—and what it doesn’t. Although Richland will replace Trinity in notebooks and on the desktop, it isn’t really new silicon. Trinity’s Piledriver-based CPU cores remain, and the integrated GPU is still rooted in the graphics architecture of the Radeon HD 6900 series. Richland is even fabbed on the same 32-nm process as Trinity, albeit with some more efficient transistor tuning. AMD hasn’t taped out a new chip design. Instead, it’s taking fuller advantage of the hardware already built into the chip.
The biggest difference between Richland and Trinity can be found in the power management department. Both chips manage power states with an integrated 32-bit microcontroller. Trinity’s Turbo mechanism adjusts clock speeds based on how much power the chip is consuming. Real-time power measurements are used to estimate the temperature of the chip, and the algorithms are relatively conservative to account for different cooling solutions and ambient environments. The silicon actually includes a network of integrated temperature sensors that aren’t utilized fully by Trinity. In Richland, these sensors feed into a new Hybrid Boost power management scheme.
Microprocessors take a little time to heat up when they’re put under load, so changing clocks based on actual temperatures should allow higher frequencies to be maintained for longer periods—especially if the previous estimates were conservative ones. Intel realized extra clock headroom when it switched to temperature-sensitive Turbo tech in Sandy Bridge, and Nvidia benefited from a similar approach when it started considering temperature in its second-generation GPU Boost implementation. AMD’s clock-boosting mechanism has now joined the party.
In addition to gaining temperature sensitivity, Richland’s power management has adopted smarter algorithms to balance the processor’s CPU and GPU components. Trinity increases the clock speeds of whichever component demands more power, but Richland’s new Intelligent Boost algorithm is more discerning. When one of the chip’s components—the CPU or GPU—requests extra juice, Intelligent Boost attempts to determine whether the component is truly bottlenecking system performance. If it isn’t, Intelligent Boost will save power rather than raising the voltage and clock speed.
The power management algorithms are more refined, too. AMD has spent more time profiling individual workloads since Trinity’s release, and Richland’s algorithms have been updated accordingly. Some application-specific optimizations have been implemented, as well, although AMD says it uses those only sparingly.
Richland complements its more advanced power-management algorithms with finer-grained clock speed control. AMD has added more points along the chip’s frequency and voltage curve, giving the power management engine more options when determining optimal combination for a given workload. The processor’s TDP is also configurable, allowing customization for notebook makers with unique needs.
Thanks to these improvements, AMD claims Richland consumes 17% less power than Trinity at idle and 38% less when playing 720p video. The two APUs purportedly offer comparable power consumption when browsing the web, though. Those numbers are based on AMD’s fastest mobile Trinity part, the A10-4600M, and its Richland-based counterpart, the A10-5750M. Here’s how the latter APU looks next to the rest of the Richland chips being unveiled today:
|Processor||CPU cores||CPU clocks||L2 cache||GPU||Graphics ALUs||GPU clocks||Max DRAM|
|A10-5750M||4||3.5/2.5GHz||4MB||Radeon HD 8650G||384||720/533MHz||1866MHz|
|A8-5550N||4||3.1/2.1GHz||4MB||Radeon HD 8550G||256||720/515MHz||1600MHz|
|A6-5350M||2||3.5/2.9GHz||1MB||Radeon HD 8450G||192||720/533MHz||1600MHz|
|A4-5150M||2||3.3/2.7GHz||1MB||Radeon HD 8350G||128||720/533MHz||1600MHz|
AMD is only introducing standard-voltage 35W mobile chips today, and the A10-5750M is the flagship. It’s joined by another quad-core part and a pair of duallies, all of which share the same thermal envelope. Compared to the Trinity APUs they’ll replace, the Richland chips have 200-300MHz higher CPU clock speeds and an extra 53-65MHz on the GPU side. The L2 cache sizes and Radeon ALU counts remain unchanged, but the A10-5750M gets a bit of a boost thanks to support for 1866MHz memory. Other Richland APUs, much like mobile Trinity processors, are limited to 1600MHz RAM. (Desktop-bound Trinity chips can run their memory at up to 1866MHz, however.)
Although the integrated Radeons have received only clock-speed adjustments, AMD has given them new, 8000-series model numbers. The integrated GPUs can participate in Dual Graphics teams when combined with AMD’s 8000-series mobile graphics parts, one of which we looked at late last year.
Somewhat surprisingly, AMD isn’t making any lofty claims about the performance of the 35W Richland parts versus their predecessors. At CES in January, AMD touted a 40% jump in graphics performance and a 10-20% boost in CPU performance over Trinity, but that was for ultra-low-voltage models aimed at ultrathin laptops. The gains likely won’t be as dramatic for APUs with higher thermal envelopes, which suggests Richland’s eventual desktop incarnations may benefit the least.
There’s more to Richland than fancier power management and slightly faster frequencies. AMD has introduced several platform-level enhancements, including support for the quick resume and wireless connect tech built into Windows 8. Nevertheless, Richland APUs will drop into existing sockets and work with current chipsets.
Richland-based systems will come with a bundle of AMD-branded software, as well.
One of the bundled apps, Gesture Control, will use GPU acceleration to translate hand waving into commands for media playback, web browsing, and other applications. Gesture Control appears to rely on a webcam rather than a true 3D camera, which is great for compatibility but probably means lousy precision. Another app, Face Login, will let you use your webcam to log into Windows or access websites. Folks who want to stream content to remote televisions will be able to use Screen Mirror, which promises a low-latency connection and requires DLNA-compatible hardware on the receiving side. Screen Mirror appears to be broadcast-only, so it’s more of an answer to Intel’s Wireless Display tech than a competitor for Nvidia’s Project Shield. AMD will also combine Richland with a collection of video software—Quick Stream, Steady Video, and Perfect Picture HD—that will handle bandwidth prioritization, stabilization processing, and dynamic image adjustment, respectively.
The 35W mobile Richland parts listed above are shipping to notebook makers now, and they’ll be joined by 25W and 17W parts in the first half of the year. We don’t have a timeline for desktop versions of the chip, but we know they’ll work in the same FM2 socket as Trinity processors.
Richland may not be with us very long, because AMD is on track to deliver another APU later this year. Dubbed Kaveri, this chip will feature updated Steamroller CPU cores alongside a new integrated GPU based on the Graphics Core Next architecture. Kaveri will be built on a smaller 28-nm process, and rumor has it the chip could incorporate a GDDR5 memory interface.