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.
Game developer Bob King reflects on application-side energy efficiency opportunities
Overclocking: energy use can rise faster than performance
Our market segmentation provides a framework for energy analysis that considers a wide range of gaming platforms, modes of operation, and user types
Surveys show big variations in active gameplay time depending on platform, type of gamer, and other variables
Notable Industry Activities & Emerging Technologies
AMD’s Radeon RX 480 GPU achieves 40% system energy use reduction in active gaming mode compared to the same system built with its predecessor GPU
Tom’s Hardware tests show that TDP, true power under real gameplay, and power under simulated benchmarks vary substantially for a given GPU, while ratio of full and idle power varies by more than a factor of two
Comings & Goings
Technical Advisory Committee meeting
Youth education - Kennedy High students visit the Green Gaming Lab
Green-up Your Game
Tuning up your rig
When buying new gear
Green Gaming News was delighted to speak with Bob King, a 25-year video game industry vet, including roles as Art Director on The Sims and Animation Director on Spore.
GGN: What are the key things that drive energy requirements for games?
BK: Of course, anything that is graphically heavy (taxing or stressing the graphics card and processors) is going to require more power. A specific example would be the extent of particles used when creating particle-based effects.
GGN: Will virtual reality games use more or less energy than their desktop counterparts? Will people game more or less with VR?
BK: For a start they will use somewhat more power because of the added gear (goggles, sensors). More importantly, as processing power increases so will energy needs. This is in keeping with the technological growth patterns of the first 40 years or so of the video game industry. I do think that VR sessions may be shorter than current gaming sessions, purely due to comfort. Wearing glasses or standing/moving or motion sickness will severely cut down play times. There are people who play MMOs for days on end. I don't think VR will get to that level of usage for quite some time.
GGN: Do you think that game design can integrate energy efficiency as a metric that's fed back to the gamer in real time? If so, is Watts/FPS enough or are there more nuanced yet practical metrics that would incorporate more about user experience?
BK: I absolutely think it can be done. What has to happen first is to clearly establish a gaming reason as to why it should be done. I don't think that energy conservation is the first thing on a gamer's mind, so incorporating a fun and repeatable gameplay metric that requires energy conservation will be the real trick. Gamers are eager to make games out of just about anything, so adding something into the design that provides a "buff" or "power-up" should not be difficult. The benefits to the player have to be real, though. I am certain that gamers will not sacrifice performance or gameplay efficiency in order to save energy. We're not there, yet.
Watts/FPS is a bit 'on the nose’ and not very interesting. It could be done that way, but again, making the benefits to saving energy have some sort of value in-game is the sweet spot to shoot for. It has to have value to the gameplay experience as a whole and not feel like a tacked on metric.
GGN: Do you think that game genre is a major driver of energy requirements, or are there factors that span multiple genres?
BK: Hadn't really considered it, but I will bet that MMOs are the biggest energy hogs. I am basing that upon overall time spent playing games. MMOs tend to have some pretty die-hard addicts and are using more energy per session than say, Candy Crush.
GGN: Do you see ancillary benefits of lower energy use (cooler, quieter system, lower energy bill at end of month, etc.) as motivating to a gamer? Would games needing less power enjoy a larger potential user base by virtue of being able to run on more than just the most powerful platforms?
BK: In my experience with gamers, temperature is more important than noise. Noise can be canceled out with headphones, but high temperature is a real threat to any system. Gamers are going to be more worried about frying a motherboard than they will be about noise.
I don't think that energy use is seen as a large enough issue by gamers at this stage. That doesn't mean it isn't, but gamers just aren't thinking about energy usage yet. If you could somehow tie improved energy efficiency to improved user experience, you'd better believe that gamers would take notice. Not sure that is the most profitable/sensible way to go, but in order for the gamer to care there either needs to be something they are not getting by using too much energy or there needs to be an in-game incentive/reward for using an acceptable amount of energy. It is a pretty simple equation, actually.
Ability to play a given game on a broader range of platforms is always going to be better for the software industry when it comes to increasing user base.
GGN: We wrote a bit about dynamic voltage frequency scaling (DVFS) in our last issue, e.g., as utilized in AMD's Chill tool? We've wondered if game developerss could have this in mind as they code, i.e., design with making the most of power-management capabilities in mind.
BK: I don't think it's only a matter of whether or not Devs *could* have this in mind, but actually they *should*. Allowances toward green-friendly development could be easily instituted into standards and practices when creating software and games. It would even make a lot of sense to have someone on the part of the team whose job it was to make sure that green gaming guidelines were continually adhered to and followed.
Everyone knows that overclocking CPUs and GPUs adds computing power, but we’ve heard less about the effect on electrical power. Outervision’s PSU sizing calculator provides some insight. They use mathematical algorithms to estimate actual power requirements as distinct from TDP, which is only a loose approximation of actual power requirements. We ran the calculator to generate sensitivities for +/- 20% of base rated CPU clock speeds and for GPUs at up to 50%, based on maximum recommended overclocks per Tom's Hardware "Safe GPU Overclocking guide”.
Notable are the differences in slopes of the curves lines (which represent % increase in power per % increase in compute). Variations in power use varied from 9% to 12% across this over clocking range for the CPUs shown in overclock mode and -10% to -15% in underclock mode, and from 26% to 61% for the GPUs shown in overclock mode. The general pattern is that for every percent increase in clock, CPU energy use increases by about ½%, whereas for GPUs power use increases substantially more quickly than performance. These numbers haven’t yet been validated with measured data, but we plan to do some measurements along these lines later in our project.
GREEN GAMING NEWS
Issue Number 3 - May 8, 2017
As most of the gaming industry’s revenues are from game sales, much of the available market data focuses on consumer choices and expenditures for software. Because additional consumer expenditures come in the form of energy bills, our efforts at market segmentation focus on drivers of energy use. This certainly includes selection of game titles, but many other factors as well.
Aside from the energy use of each piece of equipment, two key determinants of aggregate energy use are installed base and duty cycle. In our last issue, we presented preliminary estimates of installed base for California developed for this project by Jon Peddie Research (JPR). Here, we summarize the continuation of that work to encompass other factors and bring it all together into a few summary visuals.
JPRs recommendations on average active mode are generally consistent with what we see in prior estimates, although time active gameplay varies widely (see next article). Estimates of the non-active-gaming parts of duty cycles for PCs and laptops from past studies apply to ordinary computers and their use cases. We’ve found no prior studies focusing on gaming computers.
As part of defining installed base, we have set the lower threshold for identifying a gaming platform as one used for an annualized average time spent gaming at 1 hour per week of gameplay or more. We have further segmented the resulting overall California installed base into a variety of segments including:
Four platform types: desktop PCs, laptops, and TV gaming devices such as consoles and Apple or Android TV (NVIDIA Shield)
Three levels of equipment: entry, mid, and high-end … defined largely in terms of price tiers
Range of desktop systems to be included in the test regimen.
We estimate that as of 2016 about three million desktop PCs and laptops, and twelve million consoles and other TV-gaming devices used for gaming in California. Utilization is divided into a series of modes spanning from “off” to “active gameplay”. We identify four types of users: light, moderate, intensive, and extreme (reflecting hours per day in gameplay mode), each with their own duty cycles.
These charts collectively describe our current working segmentation of the California marketplace which will be used for the purposes of energy use estimation.
Our project team is now making energy measurements of individual systems in the lab to inform the macro-level estimates.
Perhaps no single variable has a greater influence on gaming computer energy use than time spent in active gameplay. This metric can be expected to vary widely, based on type of platform, game, and gamer enthusiasm level. In addition, preferred modes of gaming have evolved over time, most notably towards mobile gaming in recent years, which in turn implies a shift in time budgets among competing platforms with widely differing power requirements on both local clients and, in some cases, remote servers. While a few sources report highly aggregated values representing some semblance of “all gamers”, for energy analysis purposes finer distinctions are important. There is no existing study or database that characterizes time in gameplay at the necessary level of detail for comprehensive macro-level energy analysis. We have scoured the literature and assembled an overview of what others have reported.
To be useful in estimating energy use across a population of gamers, per-person time spent in gameplay must be associated with a particular installed base of equipment. All other things equal, the more inclusive an installed base value the greater the number of casual and intermittent gamers that will be included, in turn reducing the weighted-average time in gameplay (but of course across a larger population). Conversely, small intensive gamer cohorts may have high per-platform energy use but relatively low energy consumption in aggregate. At one end of the spectrum, ESA estimates that 65% of US households have an average of 1.7 people in them who game at least three hours a week (implying a gamer population of about 212 million). At the other end of the spectrum, NPD estimates that 7 million people game 7 or more hours per day. Partway along this spectrum NPD defines a subset of about 14 million “core gamers” who game 1.3 hours per day. Each of these values are nominally “correct” for the populations they intend to describe. Thus, a nuanced map of the market comprising multiple tiers of gaming equipment together with tiers of user types and levels of activity must be assembled. More on that in the previous article.
Most existing estimates of time in gameplay are found in market research reports, with few from academia. Use varies by type of gamer, demographics, and for specific game titles. Definitions of gamer types are non-standardized while methodologies are not often explained in an unambiguous manner or the level of statistical significance noted. For example, the period over which users are actually asked to average their gaming activity (days, weeks, etc.) is often not always documented; nor is the inclusive period of time (weeks, months, etc.) over which the respondent asked to consider. The underlying survey questions are infrequently disclosed, further confounding efforts to qualify the resulting information. These studies are rarely published in the peer-reviewed academic literature. Some examples of best practices include Tarng et al. , Rideout, and Gentile et al.
The chart below presents our compilation of diverse estimates, normalized to hours per day to facilitate comparisons (survey periods tend to span much longer periods of time). Note that the vintages of the information as well as the gamer populations and equipment types reflected vary from study to study.
Note that multiple matching descriptors represent estimates from different sources for the same given user group. Full bibliographic information about the studies will be included in our forthcoming project reports.
Notable Industry Activities & Emerging Technologies
Remarkable innovations in graphics cards are being introduced with some regularity. AMD has taken the time to develop and share an analysis of the energy use of their previous-generation GPUs in comparison with more recent releases.
The study evaluated AMD’s RadeonTM RX 480 GPU (Polaris architecture) compared to the previous generation R9 390 GPU when used within a gaming computer. While both are 5-teraflop cards (similar computational power), their relative TDPs are considerably different: 150 and 275 watts, respectively. AMD’s measurements identified a 40-percent reduction in PC energy use and carbon emissions from the active gaming mode, and a 32 percent overall GHG reduction for the gaming PC daily usage model including non-gaming modes of operation (based on a duty cycle developed in prior work by the LBNL green gaming project).
While AMD notes that “sometimes these performance advances have required higher levels of power consumption,” they conclude that recent advances have made it possible to maintain or improve the gaming experience while simultaneously trimming energy use by hundreds of millions of dollars a year nationally. Notably, energy savings occurred in all active modes (gaming, video streaming, and web browsing).
Central to the energy discussion are the power requirements of gaming PCs under load. It is expected that simulated benchmarks may not always be representative of actual gaming, as they are designed to emphasize “worst-case” stress-test conditions more than actual in-game loads. Tom’s Hardware recently released a comparison of actual power for a variety GPUs under the Futuremark benchmark in comparison to an actual gameplay using Metro Last Light 4k. For the 10 GPUs shown it’s in fact not always the case that power under the simulated benchmark is higher than gameplay. Interestingly, sometimes TDP exceeds actual power measurements, but other times it is well below them.
Another interesting observation that can be made using this data is to look at the ratio of gameplay power to idle power. There’s a large variation, suggesting varying effectiveness of power management across the different setups. In the best case, power use in active gameplay is about 11-times that of idle, while in the worst case the ratio rises to about 25-times.
Comings & Goings
GDC 2017: We attended the Game Developers Conference again this year and had a blast. We found high-receptivity and interest in the research we’re doing and made great new industry contacts.
Technical Advisory Committee meeting: Our TAC convened at LBNL in February for an in-depth review of the project. In attendance were representatives of AMD, the Entertainment Software Association, PC Perspective, EPA/ICF, NRDC, and the California Energy Commission. We included a tour of our new gaming lab and walked through our test set-up.
Youth Education: Kennedy High School visited the Laboratory recently, receiving tours of our supercomputing facilities as well as the Green Gaming Lab. Others were tutored in green gaming during the Lab’s Daughters and Sons to Work day.
The two high-end desktop PCs to be tested. Front: Digital Storm Velox, Core i7 6700K with ASUS Z170-E motherboard, NVIDIA Titan XP GPU, and HydroLux PRO: Exotic Custom Cooling System Rear: DIY with Core i7 5820K with EGVA X99 Classified motherboard, dual AMD R9 Fury X GPUs, and Corsair Hydro H110i GT 280mm.
Data acquisition system for power and frame rate and quality. Chroma digital power meter 66202 at bottom right.
Green-up Your Game
Tuning up your existing rig: Some GPU settings (e.g., AMD Catalyst or NVIDIA Control Panel) allow you to set a "performance" or "adaptive" power setting. Most games can have FPS set to a level of your choosing (say 60 FPS max). This allows your GPU to work less hard, especially in older games - without any visible difference being in user experience! For example: Civilization 5 can make your GPU render the menu screen at several thousand FPS if you let it.
When buying gear: Get into G-Sync or Free-sync - a recent generation of display-side hardware/software that reduces tearing and other distortion normally encountered with less powerful (and less power hungry) GPUs. See article in our last newsletter.
More gamer tips here.
* * *
You’ll find lots of information about green gaming at our website.
Send feedback and suggestions of topics you'd like to see us cover to: Evan Mills