Authour: James Prior
Editor: Charles Oliver
Date: March 16th, 2011
A year ago, NVIDIA's Optimus system attracted significant attention for its capability to dynamically switch between a low power GPU integrated into the chipset or processor and a higher performance GPU on demand through application detection - the Verde program. Last September we talked with Asif Rehman, an AMD Product Marketing Manager, about hybrid graphics and switchable graphics. We queried why AMD's Radeon Mobility HD 5000 series were listed as supporting AMD Switchable Graphics while the last generation Mobility Radeon HD 4000 offered support for Hybrid Graphics, plus a few questions about AMD's mobility graphics program:
Rage3D: What is the difference between hybrid and switchable graphics?
Asif Rehman: Hybrid - both integrated and discrete working together simultaneously - is an extension of Crossfire, using dissimilar graphics. Launched 2-3 years ago by both AMD and NV, it is two GPUs aiding each other. For example, the Radeon Mobility HD 4570 can work with Intel GMA, not just AMD IGP graphics. The most difference is seen when running a mainstream discrete GPU, as a single discrete GPU can now run all apps; there's no need for Hybrid technologies to combine both GPUs performance.
Switchable's main benefit is most of the time for power saving; no need for higher power GPU to be used. Earlier versions of switchable graphics required rebooting to move from integrated to discrete, as you can have only one GPU enabled at a time - not a good user experience with BIOS manipulation and system reboots required.
The second version switched on the fly without reboot, but with a switching time for the OS to swap video drivers, as Windows didn't allow two drivers to be loaded for one adapter, which is how the display appeared to it. The time to switch was longer from IGP to Discrete than vice versa, and users experienced screen flickering as the swap occurred via muxers.
The third version is cross display: no muxers, using displays connected to the integrated GPU and nothing connected to discrete GPU. The Discrete GPU (dGPU) goes to low power or off state, until an application requires the power of discrete to be used to render, but outputted via the IGP's display connection. This version is switching between GPU and IGP based on application triggers; no reboot, no flickering; instantaneous.
Rage3D: Could switchable graphics support Eyefinity (IGP with discrete, if the partner provides suitable outputs on the product)?
Asif Rehman: There are various forms of Switchable graphics available that we supported.
Muxed: Where you can shut off either of the GPUs (IGP or discrete) completely. In this case, you can support eyefinity for the displays connected to a capable dGPU [discrete GPU] only.
Muxless: The Muxless Switchable graphics technology uses CrossDisplay technology meaning that all the displays are connected to iGPU only and dGPU will render the 3D graphics when needed. In this case, since displays are connected to iGPU all the time you therefore cannot support Eyefinity technology.
Both the above-mentioned options are available to our customers (OEMs) and depending upon what they want to offer to their end-users, they can pick one vs. other.
Rage3D: Are FUSION APUs and switchable graphics compatible technologies?
Asif Rehman: Yes, since Fusion APUs are unreleased products therefore I cannot comment on their capabilities at this point. However you can anticipate to have Switchable Graphics support available on fusion APUs in some form. [Ed: this interview occurred before Fusion APUs were released]
Rage3D: What's the end user experience benefit to multiple various power/performance level graphics processors in a mobility platform?
Asif Rehman: The whole user experience improves with no switching time but there is a lower battery life vs. a solution that powered off the IGP when using the Discrete GPU.
Rage3D: When does switching occur, is there a list of applications? Will Developers add their applications to lists in the driver?
Asif Rehman: The profile list can be modified by OEMs and by AMD to give the best out of the box experience for users. It's mainly aimed at the mobility market because of need for battery life, but possibly could be seen in all-in-one PCs. It might also be possible to scale to desktop platforms with high power discrete cards, for when used as a web/media consumption system between high-power gaming sessions.
Thanks again to AMD's Asif Rehman, for his insight into PowerXpress versions 1 through 3. We spoke with him in Septermber 2010 as part of learning more about AMD VISION technology, as we were hoping at the time to get samples of products featuring AMD Mobility Radeon parts. Unfortunately this never came to be, and we mothballed the information for future reference. The leaked slides found on BAIDU are regarding PowerXpress 4.0 and Bus Alive Chip Off, information regarding which were not discussed with Asif but provide detail about previous and current technologies in use.
The new leaked info, possibly from a China or Taiwan-based partner of AMD, shows how the new BACON technology improves upon AMD's previous generation with the use of faster switching technology, user customizable application listing, and customizable options for OEMs to select how PowerXpress 4.0 based systems operate. Our sources indicate that while the slides do not appear in the traditional AMD slidedeck form, they are authentic - graphics removed to protect the source.
Catalyst 11.4 Early Preview demonstrated how AMD systems will gain more options for system control inside Vision Engine Control Center/Catalyst Control Center. This includes control over PowerXpress 4, with the ability to pick which GPU you want to use, and when to fall back to the lower power GPU option - for example, when on battery.
The purpose of PowerXpress 4.0 is to provide a smooth, seamless switching operation, giving the best possible user experience. User experience is prioritized over battery life by keeping the iGPU enabled and for processing Windows Aero or other desktop compositing operations. The OS only sees the iGPU in device manager, simplifying device installation and updating, as well as multi-panel output management.
Bus Alive, Chip Off Now! keeps the data connections and basic ASIC powered, but not processing. When the driver detects an application starting that is listed to use the dGPU (discrete GPU), BACON mode is exited and the dGPU powered on. This adds about 150ms to the application start process. At application termination, the dGPU re-enters BACON mode, at a power draw of around 150mw. For the whole system, having the dGPU in BACON mode only adds around 200mw of power, key to extending battery life while preserving the smooth fluid user experience modern mobility users desire.
Combining BACON enabled dGPUs with AMD FUSION Accelerated Processing Units promises to be a plateful of win. Products using AMD's Ontario and Zacate APUs are well balanced in terms of CPU vs. GPU power, making for great ultra-portable and nettop PCs. Using a higher power HD 6400M or 6500M could offer a boost in GPU intensive tasks, or AMD Accelerated Parallel Processing (APP) enabled programs.
The next evolution of the APU is promising to be quite a bit more powerful in CPU terms, as it is rumored to use AMD's Athlon II x86 processing cores - up to four of them, with a 480 stream core SIMD array offering around ~550GFLOPS of power, too. While almost certainly the new 32nm parts from the family codenamed Llano will have great performance/watt, they may benefit from more powerful GPUs such as the newly revealed London codenamed mobility GPU parts: Thames, Chelsea, Heathrow and Wimbledon due in late 2011 and early 2012 - again, especially with AMD APP enabled programs. The London codenames appear to have a higher TDP than current entry level mobility dGPUs, hinting that they possess more robust iGPU performance to accommodate the more powerful APU processors.
Editor's Note: AMD contacted us to request differentiation between statements made by AMD and our own speculation based on leaked information not provided or endorsed by AMD. We have updated the article accordingly.
content not found