Our motto is “Gaming Energy Efficiency without Performance Compromise”. This work is
sponsored by the California Energy Commission, and covers the full spectrum of
non-battery-charged gaming platforms, as well as gaming applications.
Meet our team and find out more about our project here.
Claudio Capobianco, AMD Radeon Technology Group (RTG), Strategy, Director Product Marketing, sheds light on their outlook for energy and PC gaming
The energy used by a gaming PC can represent 500 to 5000 pounds of carbon-dioxide emitted to the atmosphere each year
About our Green Gaming lab
New insights into the installed base of non-mobile gaming equipment
Notable Industry Activities & Emerging Technologies
Powerful power management for GPUs
Research finds that World of Warcraft gamers’ average playtime is a whopping 3.7 hours per day
Green-up Your Game
Tuning up your rig
When buying new gear
Comings & Goings
Console makers convene at LBNL
Green Gaming News was delighted to speak with Claudio Capobianco, AMD Radeon Technology Group (RTG), Strategy, Director Product Marketing
GGN: Do you hear any discussion of “green gaming” as a competitive factor as a response to any environmental impacts in the industry/marketplace today? If so, what are the drivers (e.g., energy cost, footprint, heat, noise, ability to run on lower-power GPUs)?
GREEN GAMING NEWS
Issue Number 2 - February 1, 2017
AMD: All of the above. There’s no escaping it. Power consumption has a bearing on a host of key factors in the everyday operation of a gaming PC, including noise levels, heat, size, cost, and performance. That’s true for big, high-end desktops all the way down to thin-and-light gaming laptops. As a result, power efficiency is one of the key metrics we target when designing Radeon™ GPUs. In fact, we were so pleased with what we were able to achieve with our new Polaris architecture that we completed a Carbon Footprint Study so we could better understand the impact of its efficiency improvements.
An example of a recent AMD software offering that’s targeted at “green gaming”, is Radeon Chill, an intelligent power-saving feature for Radeon graphics that dynamically regulates framerate based on movements in-game. It’s designed to reduce system power consumption while maintaining the end-user gaming experience.
GGN: What technological or user/behavioral trends do you see that may influence future energy demands of gaming PC systems (upwards or downwards)?
AMD: I’d say four distinct industry trends come to mind, and they pull in different directions:
Advances in semiconductor process technology will likely continue to improve the power efficiency of new chips over time. These advances should enable future chips to run at lower voltages with less power wasted in the form of device leakage. Meanwhile, growing transistor densities should give us room to create innovative new architectures that extract more performance from each joule of energy consumed. If you look at what’s possible now with a low-power, thin-and-light laptop or even a mobile phone, it’s miles ahead of what a high-powered desktop computer could do a decade ago. We’re working to extend this very positive trend into the future.
Users expect ever-more realistic and immersive experiences. Whether it’s VR or traditional 3D gaming using high-quality render settings and high-resolution displays, gamers are always looking to push the envelope. As a result, the temptation is always present to push to higher power levels in order to achieve higher performance and better visuals.
Gaming PCs are moving into new thermally constrained form factors, like quiet mini-PCs for the living room and thin-and-light notebooks. In order to remove the wire tethers from VR setups, some companies are even putting powerful gaming systems into battery-powered “backpacks” based on notebook-class gaming hardware. We have to design more capable and efficient GPUs in order to fit into the constraints of these new platforms.
In recent years, AMD and RTG have played a key role in supporting the development of efficient new APIs (application programming interfaces) such as DirectX 12 and Vulkan. These APIs allow developers to access our graphics hardware more directly, with substantially less overhead. We’re hopeful these improved tools will deliver improved energy efficiency as developers make better use of our chips.
GGN: What effect do you think VR will have on the energy demands of desktop gaming systems, and GPUs in particular?
AMD: Desktop PCs and GPUs have had a pretty well-defined range of power consumption levels for years now. Thanks to efficiency improvements in hardware and software, we expect nearly all desktop gaming systems to be VR-capable in the not-too-distant future. Quality VR experiences will simply be expected, even by casual gamers. We do expect immersive VR to drive demand for high-performance GPUs in the coming years, but that doesn’t necessarily mean energy demand from gaming PCs will increase. Efficiency improvements should allow us to deliver compelling new experiences within current power envelopes.
GGN: Benchmarking approaches and methods that combine the measures of user experience (UXP) and gaming system energy use are clearly a key need in this space. What's your thinking on this? Can you also speak about related industry/marketplace trends?
AMD: Most reviewers today measure GPU performance in terms of average frames per second (FPS), and some measure power consumption in a basic way. There are finer-grained ways to measure performance by looking at the consistent delivery of frames to the display at the proper intervals to achieve fluid animation. Similarly, there are finer-grained ways of measuring GPU energy use over time.
In fact, we have several methods of modulating GPU performance intelligently so that the graphics chip doesn’t generate frames at a faster rate than the display can handle. We’ve also introduced FreeSync displays with variable refresh rates that can maintain quick responses and the illusion of fluid animation at lower performance and power levels.
AMD and RTG would welcome smart new standards for testing image quality, performance, and responsiveness in the context of energy-efficient operation.
Fossil fuels are the preferred feedstocks for electric power plants throughout most of the world. Generating the more than 20,000 billion kilowatt-hours consumed each year for all purposes results in emissions of about 13.7 billion metric tonnes of carbon dioxide, for an average emissions factor of about 1.47 pounds of CO2 per kilowatt-hour of electricity.
Our earlier scoping estimate found that about 165 billion kilowatt-hours are used by PC and console gamers around the world each year, corresponding to about 240 billion pounds of CO2, equivalent to that of about 16 million cars. For comparison, the average US passenger car emits about 15,000 pounds per year of CO2.
Outcomes will of course vary widely by locality. Power plants based exclusively on coal emit about one third more than the world average per unit of electricity, while those based all on hydroelectric emit virtually none. Similarly, the choice of gaming equipment and time spent gaming is also decisive. For US conditions, a middle-of-the-road gaming PC used to game only 1 hour per day in the cleanest US region would emit about 350 pounds of CO2 each year, while an that used by typical World of Warcraft players is about four times as much.
A key component of the CA Gaming project is our new Gaming Systems Test Lab. This lab will allow us to measure both the Unit Energy Consumption (UEC) and the User Experience Performance (UXP) of each gaming system in both active and non-active modes. Some of the measurements and the equipment used include:
System Power will be measured using a Chroma 66202 Digital Power Meter which can measure power to 0.1mW resolution with an accuracy of 0.1% of reading + 0.1% of range. In addition to power the Chroma can also measure power factor, total harmonic distortion (THD), and inrush current.
Component Power (such as CPUs and GPUs) will be measured using a Measurement Computing USB-1608FS-Plus DAQ that will sample voltages and currents at 50 kHz and record them on a 1-second average basis.
Video Image Output will be captured using a Datapath VisionSC-DP2 capture card which is capable of capturing
4k Ultra HD video at up to 60 FPS. Video will be stored in a RAID 0 (data striping) array of three 500GB SSD drives with a maximum write speed of 450 GB/sec. This is readily scalable if needed. The maximum amount of data coming through with 4K 60 FPS video, uncompressed and uncropped, is about 1.39 GB/sec.
Frame Rate statistics will be measured using multiple methods including FRAPS and OCAT software running on the test system and FCAT and custom video software analysis of the video capture.
CPU and GPU Temperatures will be recorded using software running on the test system.
Sound Levels will be recorded using an Extech 407750 sound level meter using A weighting.
During the coming six months about 25 systems (PCs, laptops, consoles, and other gaming devices) will be run through the Lab, with about 1000 tests spanning a variety of variables and sensitivity studies.
Electronic gaming conducted on computers, consoles, and a host of portable devices is a major social phenomenon, engaged in by one in three people on the planet according to some estimates. As the gaming population has grown and game-playing technology has become more powerful, the associated energy use has risen.
Gaming energy demand is the product of a combination of exceptionally diverse physical and behavioral factors. The drivers include a wide array of types and numbers of gaming devices and software, including a host of user-determined parameters such as mix of devices used by a given consumer, types of games played, and time spent in various operating modes, including active gaming, web browsing, standby, sleep, and off. The market is further characterized by user choices among various system and in-game settings, and choice of display technology (monitor versus virtual reality). The mix of online versus online game hosting also has energy ramifications. A broad variety of market data is thus essential to understanding the drivers and patterns of energy use in gaming. There is a large but highly fragmented literature, with no existing synthesis providing a profile of the market usable for estimating aggregate energy demand.
As part of the LBNL research program, we are putting together a unique picture of the California gaming equipment marketplace, in collaboration with Jon Peddie Research. This includes defining a number of representative machine types and performance levels mapped to estimates of the associated installed base centered around these benchmark systems. Methodologies for estimating the installed base must stipulate a cutoff point below-which the machines are not assumed to be used for a material amount of gaming. For this project, we have set this threshold at 1 hour per week of active gameplay. Such an assumption is arbitrary, with a lower bar corresponding to a larger installed base with lower average implied energy use per device, but of course relatively more devices. Within our project, installed-base information will be paired with benchtop power measurements of the representative individual systems so as to estimate aggregate gaming energy in California.
The current provisional estimate is that as of 2016 about 3.1 million desktop computers are used for gaming in California, plus 12.4 million consoles and other TV gaming devices. The significant reduction in the installed base of PCs used for gaming between 2011 and 2016 is attributed primarily to the rising popularity of consoles and mobile gaming together with a migration from casual to higher-end PCs for those who stayed with that type of platform. The numbers for PCs are projected to rebound in the coming five years while consoles decline, with a notable structural shift away from the less energy-intensive "entry-level" gaming PCs and towards more "high-end" (and typically higher power) PCs, as well as away from the lower-energy-use console product lines. That said, the industry is today making efficiency gains in virtually every product line.
Notable Industry Activities & Emerging Technologies
While there has been lots of focus on energy-efficient gaming componentry, software solutions offer a similarly significant potential to reduce energy consumption while maintaining or even improving the gaming experience. Some win-win opportunities exist in that there is no cash investment required in order to get these savings. Dynamic Voltage Frequency Scaling (DVFS) is an important part of this, which involves dynamically changing power states to better match the resources actually required by the computing process (e.g., graphics rendering).
Major GPU manufacturers have launched power-management methods to take advantage of the fact that framerates can be varied depending on the level of action in the scene. One set of trials using one of AMD’s tools (Radeon Chill, a subtab of WattMan) found 31% power savings (108 to 75 W) when applied to WoWc, as well as significant temperature reductions (88 to 77C; 190 to 171 F), plus quieter operation. More interestingly, responsiveness was actually improved, probably because there is less congestion (aka “backpressure”) in the pipeline of cached frames (average framerates fell from 125 fps to 62 fps). They note that the level of savings varies by game title.
A third-party account published in The Tech Report measured a power reduction from 250-260 Watts to 160 Watts with Chill activated.
The chart below shows an example for a session of Witcher 3, in which power reductions with Chill were as high as 22% during periods when little or no screen activity was occurring. Conversely, for only a handful of moments was the full defaulted framerate actually required to achieve the desired user experience.
Source: Tom’s Hardware.
Benefits appear to vary widely depending on the application (and type of activity happening within a gaming session). Games defaulted to use exceptionally high framerates, such as WoWc are the best applications.
Estimating the energy use of gaming requires good market data. A lot of the data we see is cursory and its origins ill-defined. Lack of useful documentation is no doubt in part due to the proprietary value of such information, but also to the fact that energy-relevant information hasn’t been the focus of most past market research in this industry. Of particular importance is information on hours of active gameplay.
Fortunately, some in academia do look at gaming markets and published their results in the peer-reviewed literature. We’ve recently come upon one such example. Sheng-Wei "Kuan-Ta" Chen, a Research Fellow with the Institute of Information Science at Academia Sinica (Taiwan) and Director or the Data Insights Research Laboratory has been looking in great detail at the behavior of MMORPG (Massively Multiplayer Online Role-Playing Game) players. One study by Kuan-Ta focuses on the behavior of 7,043 long-term players World of Warcraft. Over the period December 2005 to October 2007, Kuan-Ta identified an average of 3.7 hours of gameplay per day for these users, with the lower 5% playing 0.5 hours per day and the upper 5% playing 8.8 hours per day. Their entire dataset is available to researchers at no cost.
Cumulative Distribution Functions per Tarng et al. 2008
The results are useful not only for estimating energy use of these intensive gamer subsegment, but also in thinking about how to characterize the additional server-side energy requirements for networked games.
Green-up Your Game
Tuning up your existing rig: Some GPUs offer "Zero-core" technology which means they can be powered down when the display is off.
When buying gear: Avoid bottlenecks: DIY machines often end up with over-spec’ed GPUs that their CPU or display can't even take full advantage of. Right-size components to work together.
More gamer tips here.
Comings & Goings
We hosted a significant gaming console industry outreach event here at LBNL on December 18, 2016. Representatives of all the major console manufacturers participated (Sony, Microsoft, and Nintendo), as well as the Entertainment Software Association and energy expert Jon Koomey, who presented some preliminary research results focused on benchmarking the energy performance of consoles. The productive meeting included an exchange of information about the LBNL research project goals and latest developments in console technology. The meeting was held in LBNL’s new supercomputing facility, which guests had a chance to tour after the meeting. Interaction will continue as the bench-testing phase of our project ramps up.