The U.S. Environment Protection Agency is moving to tougher Energy Star specifications for computers. For the first time, the spec will define efficiency in terms of power consumed while a computer is on, rather than just on standby. Preliminary conditions for desktop computers include the following: The PSU must be rated for >80% efficiency, and not exceed 50~60W AC power during idle. Over 60 stakeholders met to discuss the proposed changes for the first time on March 15. We managed to speak with a handful of attendees after the meeting. The catch is the new energy Star spec is not effective until Jan 1, 2007. A lot can happen between now and then.
March 21, 2005 by Mike Chin
Washington, D.C. does not come to mind when most of us think about innovations in computer technology. Still, U.S. government programs and regulations do have a real impact on the PC industry, and smart industry participants keep an eye on relevant intiatives from D.C. It is precisely for this reason that over 60 PC industry representatives gathered in the U.S. capitol on March 15 to dicusss the Preliminary Draft for the revision of the Energy Star computer specification with the U.S. Environment Protection Agency. Many “stakeholders” in the industry were invited.
We associate the EPA with policing toxic waste dumping by big industrials, creating rules to protect endangered species and wetlands, or enacting regulations to deal water purity issues. The fact is that the EPA mandate to “protect human health and the environment” in the US theoretically covers virtually every arena of human activity. The EPA has long been involved in setting regulations to reduce pollution and improve energy conservation. Energy Star has been a primary program for encouraging energy efficient products since its introduction in 1992.
The Energy Star marque is familiar to almost anyone who has purchased electronic consumer products in the US in the past decade. The logo shown here, or variants thereof, can be found on numerous products in dozens of categories listed at the Energy Star web site. Please keep in mind that this is a voluntary specification program, not a standard that must met by government regulations.
Energy Star is a US program, but the label has become more widely recognized in recent years. In the US, Energy Star is now promoted by retailers, utility companies and even state governments to help encourage consumers identify and choose energy efficient products. It is also internationally recognized by countries such as Canada, Japan, Australia and the EU. The US program directors work more closely with their international counterparts to expand the awareness about energy efficiency. The program’s biggest boost may have come recently when George Bush publicly urged Americans to buy Energy Star approved products.
In the Energy Star program, Computers fall under Office Equipment. Currently, the Energy Star specification for computers, integrated computers (such as the iMac) and monitors is not demanding. It only specifies what AC consumption should be while in sleep mode during which as much as 10% of the included PSU’s power rating is allowed to be consumed.
It is no wonder that 98% of available computers in the marketplace carry the Energy Star label! Furthermore, the EPA has data that shows sleep mode is actually enabled on only about 5% commercially used PCs. It’s clear the Energy Star label no longer differentiates between ordinary and high efficiency computer products. (Please click here for full details of the current Energy Star Key Product Criteria for computers.)
The EPA has been aware of the need to raise the bar. Five years ago when the spec was last changed, only a quarter of all relevant products in the marketplace carried the label. Hence the draft of the revised Energy Star Computer Specification.
There were more than 60 attendees to the preliminary draft meeting on March 15. They included representatives from the CPU makers, the power supply companies, the tier one computer system builders, utility companies, and many more. The big graphics cards makers were missing: ATI and nVidia.
TOUGHER COMPUTER SPEC… BUT NOT TILL 2007
For the first time, the draft spec introduces power consumption targets while the computer is actually powered on and running. The Active Mode portions of the spec are completely new and considerably more challenging than any of the current specifications.The effective date proposed is January 1, 2007. (Please click here for documents pertaining to Energy Star’s revision to their existing computer specification.)
Here’s a summary of the Active Mode portion of the draft Energy Star’s spec for computers:
1) IDLE MODE
2) INTERNAL & EXTERNAL POWER SUPPLIES
Silent PC Review, more than most other PC hardware website, has long espoused high efficiency and low power dissipation as the intelligent ways to achieve low noise computing. Even for us, the efficiency and the low power demanded by the draft Energy Star spec are surprising. They can be met by only a few systems and PSUs today.
The draft Energy Star proposal cites research that typical office computers sit idle 90~98% of the time that they are turned on. This means that reducing idle heat / power is highly signficant for improved thermals and acoustics as well as for reduced energy consumption.
Let’s take a close look at just one of the computer categories, the desktop. It is easier to make some kind of assessment about the viability of this new proposed specification with a product category we know well.
IDLE POWER MAX 50~60W?
By all accounts, maximum allowable idle power was the single most controversial issue at the meeting. The reactions were said to be varied from hackles rising at intrusion into the heart of the PC makers’ own matters to curious questions about how idle was being defined. The latter question seems almost petty. Idle can be easily defined as when the OS is not working and no one doing any other work with the computer. It is not likely to vary by more than a couple of watts regardless of how exactly it is defined. Still, the reaction of the attendees is significant.
The proposed idle parameter for desktops is 50~60W AC! It’s a spec that can be achieved by a very small percentage of PCs currently sold or even assembled by enthusiasts in the US or Canada. It requires not only an efficient power supply but also high efficiency components, particularly the processor, which is still the biggest gas guzzler in a PC.
Among SPCR’s own wide range of PCs, including test bench systems, there are only three current units that that meet this idle spec:
#3: Shuttle SN95G5 with AMD Athlon 64-3500+ (Winchester), 1GB ram, Matrox G550 VGA, Samsung notebook HDD and >78% efficiency PSU (51W AC idle with Cool ‘n’ Quiet engaged.)
The processor names are bolded (along with Cool ‘n’ Quiet) because they are primarily responsible for the low power dissipation of the above systems. There is at least one other processor that belongs in the above list, the Transmeta, a super low power / high computing efficiency device used mostly in notebooks, but also at least in one corporate desktop by NEC and in some high efficiency server by NEC, HP, FIC and others.
The Athlon A64 3500+ Winchester core stands alone in this group as the only “standard” desktop processor, one that’s not too far off the mark from the pace set by the burning edge silicon from Intel and AMD. The way this processor achieves low idle power is no secret: Cool ‘n’ Quiet, a feature migrated from mobile computers where it was first developed to extend battery-life. The adoption of this software won AMD an EPA Energy Star Certificate of Recognition last week for advancing computer energy efficiency.
The good thing is that Intel has finally implemented dynamic idle processor throttling (dropping clock and voltage when not needed) with its latest 600 series (and 570) CPUs in the last few months. Any new CPUs from either company are not likely to come without some kind of built-in cool-me-down-when-resting feature.
Looking over recent roadmaps from both Intel and AMD for the rollout of multi-core processors over the coming year or two, decreases in average CPU and system power may be possible. This is not to say all the new silicon emerging from semiconductor labs will be lower in power dissipation, but that the computing-to-power efficiency will be improved across the range, and this may have a real impact on power consumption. Numerous multi-core technology sessions at IDF mentioned power at parity while performance ramps up ~80% compared to single core. AMD’s assessment for the highest TDF in forthcoming dual-core server parts is a mere 5W gain, and we can certainly anticipate performance gains simlar to Intel’s.
From IDF presentation, Low Power Technology by Chekhar Borkar and Shu-Ling Garver, Intel
If we make the leap of faith that at least one or the other of either Intel or AMD will release dual-core processors at lower clock speed and power consumption level because they can, and because they will still be more powerful and perhaps cheaper than current single-processor offerings, then reductions in average system and idle power could happen. The viability of the new dual-core systems will depend much on software optimized for multiple core multi-core compatibility has to be expanded to a much wider range of software.
THE VIDEO CARD CHALLENGE
Another big hurdle for low idle system power consumption is video cards. Recent ATI and nVidia cards are shown in the detailed power consumption graph below. While the highest cards on the chart reach mid-70W, there’s already more than talk of VGA cards in the labs that pull over 150W, and if an overheard conversation at IDF can be believed, even 220W. Regardless of the exact numbers, there is little question that the top ranked gaming cards represent a serious source of power consumption in high end PCs.
This chart was created by Xbitlabs and used with their kind permission.
With graphics processing unit (GPU) thermals and power reaching the level of CPUs, you’d think more effort would be spent in curbing their excesses. Although there may have been some flirtation with dynamic power management management features in the past, these features are noticeably absent from current lineups at ATI and nVidia. This key feature may become mandatory for any gaming system seeking Energy Star approval in 2007.
Note that excellent 2-D performance graphics cards that work well in offices and for most working people actually draw very little power, well under 20W. What this suggests is that if gaming cards and systems represent the cream of the crop, the small percentage of users at the very top willing to pay big dollars for sheer entertainment, then perhaps the focus should shift more towards the middle range where a mix of good 2D graphics, movie performance and some games is enough to keep users happy. Optimizing the performance of such video cards for both performance and low power consumption should be encouraged. To this end, there were calls at the draft meeting for a separate discussion and meeting involving the graphic card manufacturers.
Recent graphics card industry numbers for 2004 released by Jon Peddie Research show clear domination by integrated graphics cards. IGCs claimed 56.1% of the total graphics market, a year-over-year gain of more than three points in share. Interestingly, Intel is ranked as the biggest graphic card maker by virtue of the biggest IGC maker. The total number includes mobile graphics, which held 20.2% share of the graphics market. Only the remaining 23.7% of graphic cards sold in 2004 were discrete cards. ATI and Nvidia together accounted for 97.8% of the discrete graphics market in 2004. The breakdown of higher power discrete cards is not known and no one from either company would return my emails or calls. In any case, if 10% of discrete cards (or 2.37% of the total sold) is >30W, then perhaps the issue is moot. The 2007 Energy Star label simply will not go on gaming machines if gaming cards remain as they are today.
HIGH PSU EFFICIENCY
The requirements in the Energy Star draft for PSU efficiency are high, but certainly within reach now, as demonstrated by the first qualifying >80% efficient Seasonic PSU in the 80 Plus program we reported about recently. By the time the new spec becomes effective, >80% efficient PSU should be much more common than they are now. High efficiency in an EPS1U PSU may be a bit harder to achieve simply because of the small size, but allowing just a few dollars increase in almost any PSU category is probably enough for efficiency to be raised to >80%, even with current technology. With enough demand, high efficiency PSU components will also become less expensive as economies of scale slide up. (An Aside: Seasonic’s Vincent Chang announced just before the meeting in Washington, D.C., his company’s objective to achieve 80 Plus certification for all the PSU products they sell in the US before the end of the 2005.)
Aside from Seasonic’s 80 Plus approved PSU, there has been a general effort in recent years by many PSU makers to improve efficiency. Some of this effort comes from the natural competition among about the PSU makers. Efficiency is a natural peformance criteria that can be be used to differentiate from competitiors. The increased efficiency incentive has also come from Intel’s PSU guides, which are not mandatory standards, but specifications that almost every PSU and PC maker in the industry follows. This is part of the role that Intel has played since the early formative days of the PC industry: Set rules and guidelines for companies to follow in order to build products that work with each other.
Those who have followed the evolution of the ATX12V Power Supply Design Guide (now v2.01) authored by Intel’s power division can identify that it was the release of v1.3 in April 2003 that added efficiency guidelines at typical and light loads, and saw the minimum efficiency increased from 68% to 70%. Details for Energy Star and standby efficiency were also added in that version.
From AXT12V v2.0 Guide: “Minimum PSU Efficiency Vs Load”
Version 2.0 of the ATX12V guide released February 2003 underwent a few more significant changes. Aside from those related to higher power requirements for more power hungry systems, section 3.2.5. for Efficiency listed not only the required minimum efficiency, but for the first time, the recommended minimum efficiency. Those numbers are substantially higher. At a presentation in Europe late last fall, Intel unveiled the possible recommended numbers for this year: >80% efficiency at 50% load and >75% at 20% and 100% loads.
A interesting point is that the >80% PSU AC-to-DC conversion efficiency at 20, 50 and 100% load specified in the draft Energy Star spec is the same specification established for the 80 Plus program begin in the first quarter of 2004. Ecos Consulting, the highly specialized company that created the 80 Plus program, “only undertakes projects that make a positive environmental impact” and has been successful in helping utility companies encourage conservation among its customers rather than build more power generation capacity (which is more costly).
It turns out Intel’s power department personnel, and the Ecos Consulting folks have been attending some of the same power conversion seminars and trade events, and began influencing each other some time ago. It’s difficult to attribute particular ideas to any one specific group or individual; a confluence of factors lead to the same efficiency target being adopted for the draft Energy Star spec. It’s even possble that Intel’s ATX12V guide will recommend the same 80% across-all-loads efficiency by the time the new Energy Star spec becomes effective.
One thing is very clear: Computer PSUs have become noticeably more efficient over the three years that SPCR has been reviewing them.
CORRECT PSU SIZING
This was a topic of discussion in one of the sessions at the recent IDF in San Francisco, the concept of choosing a PSU whose efficiency curve is well-matched to the system power load. Such matching can yield incremental improvements in average power consumption and ensure that the right amount of power supply capacity is paid for. Correct PSU sizing is very carefully practiced by tier one computer makers concerned with maximum cost effectiveness.
SPCR’s own PSU testing has shown that power efficiency in PSUs varies not only with load but also from model to model. In a given line of PSU models, the maximum and average efficiency tends to be very similar; where the peak occurs depends mostly on power rating. Most PSUs reach peak efficiency between 50~75% loading, tail off a bit at maximum power and drop at least 10% at minimum load.
PSU efficiency data from SPCR review database.
A 250W PSU for a system that draws 200W maximum and idles at 50W means the PSU and system are well suited for each other, and the PSU usually runs at its most efficient. A 500W PSU in this system would never operate at maximum efficiency, and in idle, work at <10% below best efficiency. This is an example of incorrect, costly PSU sizing. It is practised most frequently by gaming enthsiasts who are encouraged to believe that greater power capacity is always better.
WHAT ABOUT POWER FACTOR?
The current non-mention of Power Factor Correction in the Energy Star draft is something of an oddity. PFC is mandatory for any >75W device in most other industrialized nations, including Japan, China and the EU. Why not make it part of Energy Star at least? Active Power Factor Correction, in particular, has the potential to greatly reduce harmonics and other garbage in the AC lines. For industrial or commercial operators who pay for VA rather than Watts, there is real money to be saved. The utility bill is reduced, and the size of Uninterruptable Power Supply units can also be significantly reduced.
UPS units are rated and priced by VA. If a low 0.6 PF power supply draws 200W from the wall, it draws 333VA while doing so. This means that to ensure an adequate overhead UPS, you might choose a 350VA or 400VA unit. But with an Active PFC PSU, the same 200W draw will allow result in ~205VA or lower. This means a single 450VA UPS would handle two 200W computers. Multiply the savings in the utility bill, UPS backup costs, reduced AC rewiring needs (due to lower current demand) for a corporate building housing several thousand computers, and possibly reduced air conditioning bills due to lower IT heat, and suddenly the numbers get pretty enticing for bean counters, CTOs and even CEOs.
MOTHERBOARD VOLTAGE REGULATION EFFICIENCY
So far we have identifed several efficiency bottlenecks: The high power draw of the CPU and the VGA, and the (in)efficiency of the AC/DC power supply. There is yet another rarely discussed botleneck. This is the Voltage Regulator Module on the motherboard itself.
Looking specifically at a single CPU desktop system, AC power is delivered to the PSU, which then coverts it into three main DC lines: +12V, +5V and +3.3V for delivery to the components. The DC current gets to the motherboard in two ways, the 20 or 24 pin main ATX connector and the dual 12V connectors (or quad 12V in the case of dual CPU boards). In the most current P4 and A64 motherboards, all the power to the CPU is obtained via the dual 12V lines. The VRM, whose components are usually arrayed close to the CPU, performans the function of converting the 12VDC input from the PSU into the ~1.4VDC (and other) voltages needed for the CPU. By now you should not be surprised that there are losses through the VRM.
The efficiency of VRM circuits on motherboards ranges considerably. The very best server motherboards desgined for critical enterprise or military applications can reach past 90% efficiency. The cost of the components for such circuitry is extremely high, as is the overall cost of the board.
Efficiency curve of high end server board.
(Graph courtesy of Andy Watts, Test Automation at Intel)
Typical retail motherboards range closer to 75~80%. Intel’s own retail offerings tend towards the high end of this range, possibly up to 85% for entry level server boards. In comparison, budget products may offer even less than 75% efficiency.
Efficiency curve of good quality Intel retail board.
The bottom line is price. Higher efficiency can be had for a price, and that price gets steeper the higher you go. Also, it is difficult to design out the bottom end drop-off in efficiency when the VRM is optimized for high power delivery (>100W). So it is not realistic to have high efficiency at Pentium M level power draw (barely 20W) and still have high efficiency ramping up past a hundred watts. In other words, even if you could get a P4 desktop-type CPU to idle really low, the power range might be too high for VRM linearity so that the overall power draw at idle would be little improved. For mainstream desktop and server systems, efficiency losses in the VRM will hit hard when trying to reach the idle levels proposed in the new Energy Star spec. It will be interesting to see whether any motherboard makers make any moves to improve VRM efficiency at this time.
LOW IDLE JUST FEASIBLE TODAY w/ SOME DESKTOP SYSTEMS
When you add up the efficiency losses in the PSU and on the motherboard, how much AC power does it take to power up, say, the Intel P4-2.8C (Northwood) that is powering the system I am working on now?
Let’s assume that the processor demands the theoretical maximum of 80W when I am working on large graphics while multitasking with 6~8 other programs. Assuming 80% efficiency in the VRM, this means 100W DC enters the board. For an 80 Plus certified PSU with 80% efficiency, 125W AC input is needed to ensure 80W of power at the CPU. This matches the highest peaks seen on my AC power meter when the system is worked very hard. The system efficiency is about 64%; conversely the total energy loss to heat is 36% or a third. Obviously, the motherboard is a significant bottleneck to improved efficiency in this desktop computer.
It’s difficult to calculate what happens with my system at idle for many reasons, but I can measure it with the AC power meter: 74W. The PSU I am using might not pass 80 Plus, but it would come within a hair. This is a 1GB ram system with a low power dissipation Matrox P650 video card and no extras except an additional hard drive, which I know draws about 6W AC at idle. So the lowest idle I can achieve here with just one desktop HDD is about 68W. Scaling back to 512mb ram might bring it down to the low 60s, but it would still be higher than the idle max of the proposed draft Energy Star spec.
An AMD Athlon 64-3500+ (90nm core) CPU system running in the lab with a single hard drive and otherwise similar equipment draws only 51W in idle with Cool ‘n’ Quiet enabled and 64W without. This is a substantially faster processor than the P4-2.8 for most applications. But lest you take away the lesson that this processor is somehow intrinsically superior, just wait. Competition between the big silicon part makers is such that we’ve seen one surge ahead technologically, only to be caught up and passed in the next surge by the other. As mentioned earlier, Intel has brought dynamic clock / voltage throttling at idle to their desktop processors, and there is no going back. Both major processor companies will be producing more efficient processors with smarter thermal management in the coming years.
Even today, 50~60W AC idle is clearly within reach for a certain segment of systems. By the start of 2007, we can certainly expect to see greater progress on the low power, high efficiency front, with technology and thinking from the energy-miserly mobile sector permeating more thoroughly through all the processor divisions. Higher efficiency PSUs will also be more widely available with increasing efficiency competition among PSU makers spurred on partly by rising Intel power supply design guide specs and recommendations.
IS 2007 TOO LATE?
Without looking at all the specs for the other categories of computers, we’d suggest that unless there was a serious hue and cry among the attendees, the specification has no teeth. From discussions with many experienced people who were there at March 15, the consensus seems cautiously optimistic. Yes, there were grumbles and complaints, and perhaps some things will have to be tweaked. But there are nearly two years before the new spec becomes effective, time enough to prepare for anyone who is interested enough. A great deal of computer technology sees the light of day in such a span of time.
It is a brave new world the Energy Star team is building, and we wish them all the encouragement. We hope for everyone concerned that the proposed specifications are not watered down. The target of 20~25% initial product compliance with Energy Star 2007 is a good one. If everyone is a star, then, most people would say, no one is a star. Allow the exceptionally efficient products to be recognized, and let them be examples for others to match. This is the whole point of a voluntary program to encourage high performance. Make it too easy and it becomes meaningless.
What is SPCR’s interest in the new Energy Star program, in Intel’s evolving PSU design spec, in Ecos Consulting’s 80 Plus program or in the prospect of more power efficient CPUs? The answer is simple: Low heat and high efficiency are the cornerstones for quiet computing. Less heat, less need for fan airflow, less noise: On a certain level, it really is as simple as that. The efforts to produce exceptionally efficient computers for Energy Star approval will create the foundations for quiet, ergonomic computers. Clever marketing people will not miss such an opportunity. If some exceptionally quiet, attractive and powerful computers are not among the Energy Star approved products of 2007, I promise to eat my crystal ball. In the meanwhile, if you don’t want to wait that long, SPCR remains dedicated to silent computing with today’s systems and components.
* * *
My thanks to the many individuals who helped me gather the information for this article:
Craig Hershberg, Energy Star, EPA; Dan Snyder, PR Manager, Intel; Brian Griffith, Power Delivery Architect, Intel; Andrew Watts, Test Automation, Intel; Chris Calwell, VP Policy & Research, Ecos Consulting; Peter Ostendorp, Research, Ecos Consulting; Vincent Chang, US Manager, Seasonic; Teresa de Onis, Desktop Branding Manager, AMD; Vic Bhagat, Product Marketing Manager, AMD