Tomshardware botches Intel Atom energy efficiency tests badly

Update 8/20/2008 – I need to fix my mistake because I quoted some initial informal numbers from Jack that didn’t include the 3.5″ HDD or dual-threaded tests.  Now that the I have his full data set, I’m need to make some corrections.  I apologize for my mistake and I will fix it below, but my insistance that Tomshardware is off by a LOT has not changed though I was wrong about saying the Intel 945GC/Atom board could run on 802.3af.  Tomshardware on the other hand only did NOT correct the mistake despite acknowledging my emails to them, they went ahead and and published more results using an 850W power supply which is horrendously stupid.

For those who are going to say that Thorn who critized me in the comment section was right all along despite the fact that he did not run ANY tests, he was still wrong by a long shot because he insisted that the Tomshardware numbers aren’t that far off.  It turns out that Toms hardware was off by 25.6W from the Sparkle 220W power supply and my hasty post based on informal data that mistakenly excluded the hard drives was off by 11 W.  Let this be a lesson to me for posting informal data and once again, I apologize to my readers for my mistake.

After praising Tomshardware for doing a good job fixing their flash storage efficiency article, I must point out that Tomshardware’s made another horrendous error by claiming that and Intel D945GCLF system with an Intel Atom 1.6 GHz processor consumes 59W idle power.  Mr. Dandumont who authored the article claims that he used an “80 Plus” power supply implying that he was getting at least 80% efficiency on the PSU (Power Supply Unit) but that is a huge mistake.

UPDATE 7/21/2008 – Tomshardware’s Senior Editor Matthieu Lamelot has responded that they used the Tagan EasyCon U15 530 W power supply.  My 600W guess was fairly close and it means that Tomshardware loaded their PSU to roughly 3.6% load and that translates to roughly 30% efficiency which makes their benchmarks very inaccurate and misleading.  Tomshardware should at least test with a good 80 Plus 220W PSU but ideally they should test with a sub-100W PicoPSU.

The 80 Plus rating only claims greater than 80% efficiency if you’re talking about workloads between 20% to 100% output power and my guess is that Mr. Dandumont used a 600+ watt PSU which means he was likely loading the system at less than 3% workload.  3% workload on a PSU translates to a horrendous efficiency of less than 30%.

Tomshardware claims that the same system with a 7200 RPM 3.5″ hard drive consumes 59W idle and 62W peak using an energy efficient power supply but this isn’t even in the ballpark in terms of accuracy.  Tomshardware needs to correct this error and start using some more appropriate and smaller power supplies for testing computers that draw less than 50 watts of output power.

Update 8/20/2008 – My PhD friend Jack who is a very knowledgeable and meticulous tester tested the Intel D945GCLF with a Sparkle SPI220LE 220W PSU and got a measurement of 34W idle and 36.4W peak input power consumption.  Since the “80 Plus” SPI220LE gets around 73% efficiency at 19.2W output load according to Silent PC Review, we can reasonably estimate at 75% efficiency that the entire system actually consumes ~27W output power from the PSU.

Without the hard drive, the system peaked at 27.2W which means the output power from the PSU would be ~20W which is too much for 802.2af PoE.  The Intel 945GC chipset is unfortunately a bit power hungry so when the next version of Atom with on-die controller and graphics built in to the CPU, we can hopefully see the output power requirement dive down below 10W but until then, 802.3af is out of the question.

The SPI220LE is above 80% efficiency when the output load level is above 20% and here we’re only at around 8.1% load level so the efficiency drops.  I’ve done some testing with the Sparkle 55W Open Frame PSU and found that it runs at around 80% efficiency for the Intel 945GC/Atom board despite the fact that it’s not 80 Plus certified.  This is because the 55W PSU is running at optimum load levels.

46 thoughts on “Tomshardware botches Intel Atom energy efficiency tests badly”

  1. I wonder what the folks at Intel would have to say about this slanderous article produced by Tom’s?

    Thank you for again attacking Tom’s Hardware and showing how they continue to lose relevance in the tech world. I used to be one of their faithful followers back in the Pentium II days, but am now finding out that they are simply not reliable as a information source anymore.

  2. Hi George,

    Congratulations about getting the SSD vs HDD comparison right from Tomshardware. You are and have always been a myth-buster of technology and are so good. After hearing that you are leaving ZDNet, I really had to follow you via Google and have been glad to see that you are still active in writing tech blogs as before.

    I realized from your article that putting a more powerful power supply than what is really needed may not always be a good idea because of efficiency loss. I am building a system with Intel Core2Quad Q6600 processor, two 500GB harddrives, EVGA 8500GT nVidia graphics card, SLI-capable mother board (Abit Fatal1ty IN-9), and other usual stuff like DVD-RW drives etc. Can you provide some inputs on how much power supply do I really need? I am using Antec EarthWatts 500W.

    Also, I have wondered for some time what is the cost/benefit relationship for efficiency in the long run. For example, a processor with higher power-to-performance ratio may be cheaper but would consune more power and may turn out to be costlier in the long run. Are you familiar with any study along these lines.

    Thanks – Alok

  3. The Q6600 and is slightly more power hungry and it’s an older 65nm technology. You might want to spend a little more money on the Q9300 45nm quad-core which has SSE4 which makes video encoding tasks much faster as well as other things. All Intel 45nm quad-core processors consume much lower power than Intel 65nm processors while being stressed by the average application.

    The 8500GT is cheap and low power but it’s not a good gamer and you might want to consider getting the 9600GT for a little more money. The 9600GT is good for gaming and uses fairly low power.

    For the PSU, whether you do your parts or my recommendations, the SeaSonic 330W 80 Plus PSU is what I recommend. It’s a very silent, reliable, and affordable PSU and it’s more than enough power for what you’re doing.

    George Ou

  4. …the chipset’s TDP is officially rated at 22W and the CPU’s at 4W. Adding your 9W for the hard drive, 3W for a single DDR2 DIMM and approximately 2W for the fan and VRM inefficiency, I’d end up with a full load power consumption of around 54W at 73% PSU efficiency. From that perspective your numbers don’t look any more realistic than Tom’s Hardware’s.

  5. Thorn, Jack and I both MEASURED the numbers. Jack measured a peak power of just 25.4 watts for the ENTIRE system including a 9-watt 3.5" 7200 RPM hard drive. It’s quite silly of you to go spouting off theoretical numbers of 54W by adding TDPs when you’ve never done any measurements. For you to criticize our measured numbers with your theoretical numbers is laughable.

    Tomshardware at least measured the numbers but they used the wrong power supply that resulted in less than 30% PSU efficiency. The idea that an Atom system would consume more than 50 watts is ludicrous. I’ve built dual-core Intel Core 2 systems with "65W TDP" CPUs where the entire system consumes less than 50 watts. http://blogs.zdnet.com/Ou/?p=934

    George Ou

  6. I don’t think it’s silly to question a result that concludes a peak power consumption of around 7W for the CPU and chipset only (taking your numbers and deducting 3W for RAM, VRM and fan, which is conservative) while the actual manufacturer advertises a 26W TDP for those components and the engineers decided to put a rather large heatsink with active cooling on the chipset. Why should they waste money and marketing potential like that?

    After all your criticism on Tom’s Hardware’s results is also just based on the guess that their results suffer from horrible PSU efficiency.

  7. Again, STOP counting TDPs. TDP is a number useful for people who design heatsinks and cooling systems; it should not be used for computing peak power, idle power, or average power consumption. I am telling you that Dr. Jack and I both MEASURED peak power consumption in the mid 20 watt range. I am telling you that I have Core 2 Duo systems that measure 40W idle and I even have Core 2 Duo systems that measure 50W peak. You haven’t measured any thing and you’re in no position to be criticizing my criticism of Tomshardware measurements. I’ve informed Tomshardware and they’re reevaluating it.

  8. Thorn,

    There are major problems with the way Tom’s measures power in general. Specifically, for the Atom reported numbers — this one was horrendous. First, they do not state the exact configuration, most importanly the PSU, but I can surmise … I get roughly their numbers (59 W idle) for the same Atom board when I use a 500 W PSU (CoolerMaster).

    The problem — high power PSUs extraordinarily bad efficiency at low loads, such that the measured input will appear almost constant at very low loads. What I mean by this is when I measure total system power for the 500 W PSU, I measure at the socket ~59 W idle and ~60W load for the atom board …. so how to fix this?? Well, you use a lower power, more efficent PSU with a known (measured) efficiency curve. The exact same setup, exact same board, etc. just plug in a Sparkel 220 W PSU rated at 80%+ efficiency (73% at 20W actual) and I get completely different numbers, more than 1/2 of what Tom’s measures (~ 24-25W Idle, and 25-26 full load).

    Tom’s and your mistake is assuming that an 80% efficient PSU is 80% efficient at all loads, from 0 to highest rated power, this is not true.
    http://www.codinghorror.com/blog/images/power-supply-efficiency-comparison.png

    If you plot the load on the PSU vs actual efficiency, most typical PSUs (say 400-800 W rated) begin to falter at about 20% of their rated load and the efficiency drops off. In the plot I linked above, at 50 W load the efficiency has already dropped well below 70% for both PSUs ….

    Since we are trying to obtain load information for a product coming well under 50W (actual consumption), the PSU efficiency will drop even more… you can see the rate of decline in the low power region in my example, it is not uncommon for these to end up 20 or 30% efficient at 20W load.

    To avoid this, I acquired a Sparkle 220W 80+ CPU, for two reasons … the lower wattage PSUs will be more efficient at the lower power loads and, two, because the actual efficiency near the estimated consumption has been measured and published:

    http://www.silentpcreview.com/article773-page4.html

    I can show clearly that using a high power rated PSU will give you Tom’s results, and using a lower, more efficient PSU will give you a completely different answer.

    Second, you clearly do not understand what a TDP is … this is not a consumption spec, it is a spec to which your thermals must be designed. Actual consumption is significantly less. It is because engineers know their product well that TDPs exceed the actual but a wide margin (this is called a guard band in engineering speak), basically it is extra engineering margin to avoid catastrophe.

    In the end, you are not correct, and Tom’s completely botched this one (as they do many others)

    Jack

  9. Jack,

    I perfectly understood the point about PSU power efficiency under minimal load in George’s article and I fully agree that it may have a significant impact on actual results. However, his assessment that Tom’s Hardware was using a 600W+ PSU was just a guess and I wanted to point out that your results appear to be on the lower end of the scale, based upon what the engineers decided would be the upper limit for the thermal envelope.

    I also think I know pretty well what the TDP represents and there’s of course a direct relation to electrical power consumption, since the device can hardly generate more heat than it draws in electrical power. It can actually draw a lot more power for short intervals since a TDP-sufficient cooling solution will be able to cope. I’m quite aware that there’s a safety margin, because there are manufacturing tolerances and the TDP is based on worst case conditions like maximum die temperature. That safety margin is nowhere in the range of 400% like George’s calculations based on your measurements would suggest, though. Intel would not put a relatively expensive large heatsink with active cooling on a part that will reliably only draw less than 5W in reality – not for such a low-cost solution (this is not an enthusiast overclocker’s board). They placed a much smaller cooling solution on the actual CPU which is rated at 4W TDP after all.

    I do not doubt your results, but I’m also quite sure that other people’s results are equally valid even if they show significantly higher power consumption. They just ended up with less excellent quality components and had worse testing conditions (George did not provide any temperature readings).

    In the end, this motherboard will be used in cheap nettops with cheap power supplies in cheap cases with minimal cooling. I found one excellent review of the board which takes at least care of the second aspect. Unfortunately it is in German:

    http://www.minitechnet.de/intel_d945gclf-littlefalls_1.html

    They tested the chipset without the fan and the heatsink temperature climbed to 67°C within 5 minutes. That’s definitely not a 5W part. The whole setup, using a cheap external 80W PSU, consumed up to 46W under full load. They provide CPU and chipset temperatures and let the load test run for 12 hours to let the components heat up. Would you say that their tests were also botched or for what reasons do your results bear more validity?

    Again: Tom’s Hardware’s results may be off for several reasons and maybe they even used an inappropriate PSU, but 25W is not the ultimate truth either.

  10. Thorne, you simply do not understand power measurements. It’s not that Jack and I are getting "lower end of the scale" results; it’s that we’re getting MORE ACCURATE results by using the closest to correct size power supply. The most accurate result is to use a sub-100 watt PicoPSU power supply which can do 80% or more efficiency at 20 watt output load. In that case, a full Atom/945 system with 3.5" hard drive should get around 23 watts of input power at the socket.

    Was it a guess that Tomshardware used 600W or higher power supply? Sure, but a very educated guess based on lots of good experience. Since Tomshardware never disclosed the detailed methodology or hardware used which is a very sloppy practice, we made a logical guess which is almost certain to be correct.

    I’ll say this one last time to you since you seem to refuse the well experienced and well educated explanation to you. Anyone who gets more than 25 watts is using an inefficient power supply. The best result you can get is from a PicoPSU which should get under 24 watt peak for the entire system.

    George Ou

  11. Thorn -

    Tom’s published an article, and reported a power consumption construed to reprent the power consumption of the Atom based board under question. He further states, incorrectly, that the efficency of the power supply is 80%. Tom’s did two thing wrong

    1. He failed to disclose the full details of the system such that anyone could reproduce the results.
    2. He mislead the reader by implying that the power consumption was X watts at 80% efficiency.

    I don’t doubt he measured what he measured, but he did not account for the error in his measurement, more than 1/2 of which is in the PSU.

    Your error is using TDP to validate his result….

    It makes no difference really, reality is that Tom’s measurements are off by a factor of 2 because they induced the error with a poor powersupply. I can reproduce his numbers with a poor power supply and I can get closer to actual by using a more efficient low-load supply with actual measured efficiency to conclude at most the atom board consumes < 20 W full load.

    Jack

  12. Jack,

    I’m not trying to defend Tom’s Hardware here. I’ve seen numerous flaws and lack of disclosure in their reviews, especially when it comes to measuring power consumption. I also agree that the results are misleading, since they imply 80% efficiency without providing any details about the PSU in use. I’ve hinted at all that in the title of my first comment.

    I posted a comment because of the equally misleading conclusions George draws in his rant. He takes your results (based on a single board sample and tested under your specific conditions) and concludes in the end, that a system with a 2.5" hard drive should run on Power over Ethernet. That’s boiling your results down to the level of an engineering decision, for which the TDP is a more accurate guideline than a single test. When building such a system you have to expect that it draws significantly more than 30W under load, because that’s what the manufacturer tells you. I could have looked up the electrical specifications for the board, but I already knew the TDP ratings and the laws of physics allow me to use them for a rough estimate as well.

    George,

    you’re insisting on your point that the system may not exceed the power cosumption results you got from Jack’s and your own tests and that’s your mistake. You tell me to stop talking about manufacturer guidelines and accept your measurements as the ultimate truth. If you do not trust Intel’s specifications, I can only suggest to test a couple of dozen samples from different charges in closed cases of various types and under a variety of external temperature conditions. Let them run for a couple of hours under full load and then tell me no system ever exceeded your 25W initial assessment. If you also provide some temperature and voltage readings, I will shut up and accept that it are only Intel’s specifications which are way off.

    Like stated before, I’m also convinced that system builders will mostly use $10 PSUs for Atom-based nettops and that actual end-user results will be in the 30-45W range the review I linked to suggests or worse. I’m not seeing too many solutions using a PSU that’s almost as expensive as the entire board, as you did.

  13. The Sparkle SPI220LE is $49 which is actually a very cheap power supply. Please do not tell me that what I’m testing with is unrealistically expensive. You went off on a rant claiming that it would be normal to get 40+ watt measurements when you’ve NEVER done a single measurement and you do not know what you are talking about.

    I’ve done tons of measurements INCLUDING the 945/Atom system with a 2.5" HDD and I’m measuring EVEN LOWER power than Jack but I haven’t completed testing YET to publish the results, but I’ve seen the results already confirm Jack’s results. I’ve measured many 3.5" HDDs and I’ve seen hundreds of measurements on hard drives and they all confirm around 9 watts idle unless it’s the Western Digital Green Power drives which are 4.5W idle.

    Jack’s results of 18.5W output power and 25.4W input power along with the data from Silent PC review is very conclusive evidence that the Atom system without the 3.5" hard drive is sub 10 watts. It’s common knowledge that 2.5" HDDs are typically under 3 watts and that translates to 13W peak power consumption. With a PoE DC converter that is 90+ percent efficient, the input power on the Ethernet jack is under 15 watts which is within the 802.3af PoE spec and most PoE switches go a few watts higher than spec.

    I don’t mind intelligent and well researched arguments that dispute my analysis but your arguments are neither.

    George Ou

  14. And that’s either because you don’t or you don’t want to understand them. Here they are in short:

    1. Power consumption of the board varies a lot depending on manufacturing tolerances for individual components, including CPU, chipset, VRM and RAM.
    2. Power consumption also depends heavily on component temperatures. Room temperature, case airflow and duration of operation have a significant impact on that.
    3. Manufacturers provide guidelines that guarantee stable operation under specified worst case conditions.

    When Intel recommends 7.75A current capacity on the 5V output rail of your power supply solution for this board, saying that the 15W PoE can deliver will suffice after DC-DC conversion loss is a bold and dangerous claim. You’re matching your measurements, calculations and estimates to a specification. Engineers match specs with specs.

    On a side note: _peak_ power consumption is something entirely different. Most 2.5" hard drives draw up to 5W at startup/wakeup – 3.5" drives easily exceed 25W there.

    I don’t care if you’re thinking my arguments are unintelligent. I’d call them trivial facts which you just refuse to accept. Please take your board, attach a 2.5" hard drive, connect the system to a DC-DC converter, put the whole setup into a small case and let it run under load for several hours. You can measure power consumption at the DC input of the converter to take PSU efficiency out of the equation. Then publish the results together with temperature readings from chipset, CPU and fan intake, disclose the kind of load you ran, what peripherals were connected and what screen resolution you used. That would be a valid test with sufficient information to verify it.

    I NEVER (to use all caps like you did) claimed to have done measurements, but I provided a link to a review of the same board with more details disclosed than you have and with a setup that is more realistic than yours (no, $49 is not a very cheap power supply). They used an 80W PSU (Intel recommends 75W minimum) and got significantly higher results than you did. So, again, what’s wrong with their figures?

  15. 1. There is variance, but it’s not more than a few percentage points. We’re talking +/- 2 watts at worst here.

    2. We’re talking room temperature here. We’re not talking about the arctic or the desert. There is no need for air flow or case fans because we’re only talking about 25W total system power consumption.

    3. We’re not talking guidelines here.

    Regarding PoE, it’s actually 13 watts MAXIMUM including worst case on the CPU and Storage. 2.5" drives idle at a little over 1 watt and peak at around 3 watts. This motherboard/CPU maxes out at under 10 watts. PoE can usually supply about 18 watts and if you knew anything about enterprise-class 802.11n routers, they’re all well over the 802.3af spec. This is the main reason Cisco is putting out PoE switches that support 20 watts.

    Now thorn, you came here spouting off how we’re wrong and that 45W measurements would be normal when you’ve admittedly never done any measurements yourself. I’ve repeatedly told you that I can get a Core 2 Duo system at 40W idle and 50W peak power consumption but that’s gone over your head. Now you’re changing the subject to the fact that maybe I’m pushing too close to the PoE standard but it sounds like you’re admitting it would at least be supported under normal operation. What you have done is gone off with a half-baked untested theory and you’re trying to come here and tell us we are wrong, and I’m saying that’s a very dumb thing for you to do.

    If you think $49 is an expensive power supply, not only do you know nothing about power measurements; you know nothing about the hardware market. I can assure you that the PSU Tomshardware tested was at least twice as expensive as the one I tested but they simply used the wrong size and most likely used something in the 600W or more range. I know what they like to do with their low-power tests and this is always a big mistake that they make. They consistently measure much higher power consumption which is simply wrong. When you do benchmarks, you’re always aiming for the best results and if you get results that are less than half as good, you’re doing something wrong.

    George Ou

  16. 1. Manufacturing variance is already much greater than that. Taking the Atom as an example, Intel sells chips from the same production line either as a Z530 or a 230 which differ in maximum specified current draw by as much as 60% at the same voltage, temperature and clock frequency. They naturally spit out chips with much lower real maximum current draw and bin them as Z530s. They will sell anything that’s acceptable as a 230 as a 230 and will even sell dies which are eligible for the Z530 product as 230s if market demand dictates it. The same applies to the chipset and RAM, so we have a large variance for the components alone.

    Leakage power increases anywhere between linearly and quadratically with temperature. VRM efficiency and loss for passive components also depends significantly on temperature. Don’t tell me that this is an insignificant point but do some research yourself. Start with Intel’s minutely specification for thermal solutions.

    2. Do you think you have room temperature in your computer case, the heatsinks and down to the die level? Intel thinks that its reference active cooling solution is sufficient to keep the chipset temperature below the critical 99°C as long as the temperature at the fan intake does not exceed 38°C. You don’t think there is a need for air flow? Then why did Intel’s engineers decide to put a fan on a motherboard for which every penny in cost counts? Remove that fan and put the motherboard in a cardboard box and see how long it will work. You can even watch the power consumption rise to check my other argument.

    3. You’re not talking guidelines – I am. They’re not there for fun, but for system integrators to build reliable while cost effective solutions.

    PoE: Just go ahead and build the thing. I don’t believe it’ll work and I have proven to you that Intel doesn’t expect it to work either. I can only repeat myself.

    My own measurements: I provided you with a link to measurements with different results, which you have decided to ignore. You also just call my theory half-baked without showing any evidence for its weaknesses. You just keep repeating that you’ve done your measurements, dismissing everybody else’s and ignoring that I already gave you the reasons why your results may be on the lower end of the scale.

    Changing the subject: I’m not. I never said that Tom’s Hardware is right or that 45W is the ultimate result. I posted my initial comment (which started with "While their numbers may be off…") because you used the occasion to base some very unprofessional calculations and estimations (or should I say a half-baked, untested theory?) on your own or Jack’s results. The statement regarding the possibility to power a system with PoE proved to me that you were overestimating the relevance of your results for applications beyond your desk under a large variety of conditions. That’s exactly what manufacturer specifications are there for and that’s why I used them as an argument.

    Cheap power supply: I said they’re not very cheap and now you’re claiming I’d consider them expensive. System integrators use $10 PSUs for low-cost solutions and I expect them to do the same with Atom-based Nettops. That is what I call very cheap and it incidentally was what the reviewers I refered to were using.

  17. Your arguments are either wrong or they’re irrelevant to the discussion. You also don’t speak for Intel so don’t use that argument here. Chip-to-chip variation isn’t 60% or anything close to that and you’ve provided zero evidence for that.

    It’s quite stupid of you to compared wholesale manufacturing costs for PSUs to the $49 retail pricing I quoted. I can assure you that computer makers can build a $10 wholesale 50W to 100W PSU that gets at least 73% efficiency.

    The idea that you’re implying Nettops will be inefficient is very laughable. MID Atom devices have been measured at 3W average or less for the whole system and Atom devices are sub 10 watts I can assure you because they wouldn’t last 6 hours on a 6-cell battery if they were inefficient.

    You’re wasting my time here.

    George Ou

  18. I’ve provided hard facts and verifyable evidence and the only answers I get from you are that my arguments were wrong, irrelevant, stupid or laughable without any attempt to provide evidence to the contrary or further explanation.

    Since this is leading nowhere, just answer me one simple question straight: Why does Intel specify a maximum power dissipation (and therefor power consumption) of 22W for the chipset and put a much more expensive cooling solution on the board than would be necessary, if under no circumstances this part will ever draw more than a few watts like you measured?

  19. "Since this is leading nowhere, just answer me one simple question straight: Why does Intel specify a maximum power dissipation (and therefor power consumption) of 22W for the chipset and put a much more expensive cooling solution on the board than would be necessary, if under no circumstances this part will ever draw more than a few watts like you measured?"

    Thorn, now you’re just being annoying when Jack and I have already answered that for you. TDPs, for the last time, is useful for heatsink and cooling system design. It is not useful for determining maximum power usage under real-world loads. You can take that answer if you like or not, but I’m not going to waste any more time repeating myself and Jack as already given up on explaining this to you long ago.

    George Ou

  20. It’s an evasive answer and I already responded. We have physical laws in our universe and microchips are not beyond them. There’s no secret fuel reservoir on a chip to power additional heating. All the heat a cooling solution may have to dissipate is generated by electrical power. If Intel thinks you may need cooling for 22W TDP, they are likewise telling you that the chip might draw 22W in electrical power. The chip may actually draw more power at peak times (a term that you consistently use mistakenly).

    There is of course a safety band as Jack mentioned, but it’s nowhere that broad that Intel would state a 22W TDP (and include expensive active cooling in the reference design) for a chip that may never draw more than 3-5W under load. They put a much smaller heatsink on the CPU itself accordingly, since that is only rated at 4W TDP. It is rated as drawing a maximum of 4A current at 1.15V, by the way, so the theoretical peak power draw is actually slightly higher than the TDP.

    So, again: In a closed box with bottom-of-the-bin parts and an ambient temperature of 38°C (rated as maximum for the reference board) the chips will operate close to their maximum case temperatures and the setup will draw much more power under load than what you measured. This is a theory based on Intels engineering specifications and the laws of physics and you can’t invalidate it by patronizing me or insisting on your own results which do not take these aspects into consideration. What kind of load were you running by the way?

  21. I’m trying to figure out if it is better to use a PicoPSU 90W with a 60W or 90W adapter!
    Can you help me? Is 60W enough/efficient or not ??

  22. Either the 60 or 90 will work well. So long as you’re between 20% and 100% load, you’re doing well. But 60W PicoPSU is more than sufficient for an Atom system and you’ll actually get better efficiency (over 80%) than the 220W Sparkle SPI220LE. However, this will only save you around 2W over the 220W and the PicoPSUs are more expensive.

  23. There’s no question the PicoPSU is more efficient. What I don’t like about it is the price and the fact that it needs an external brick. I’d love to see a 100W 80 Plus PSU with a built-in AC adapter.

  24. Since you apparently decided to moderate away difficult questions, I have one for you that you might be willing to answer:

    How do you come up with a figure of 32% efficiency for the PSU that was used by Tom’s Hardware? There’s a review of a similar PSU here:

    http://www.silentpcreview.com/article813-page4.html

    They’re measuring 70.9% efficency for the AC input power Tom’s Hardware measured for an idle system (59W input, 41.5W output) with a 650W mid-range unit from Corsair. Why would Tagan’s 530W PSU perform so much worse?

  25. Thorn, I get tired of responding to repeat comments. If you got something new to say, fine but I’m not going to post repeats.

    The SPC review you posted is a different power supply. Power supplies have significantly different efficiency curves and you’d be foolish to think they were similar.

    Jack has already shown that his exact same Atom + 945 mobo + 3.5" HDD can measure 24W idle with the SPI220LE or 59W idle with his 530W PSU. 25.4W peak input power x 73% efficiency gives you 18.5W output power. 18.5W output power divided by 62W input power on the larger and less efficient PSU gives you an efficiency of just 29.9% efficiency.

    As for my update, I was off by about 2 points in my quick update and I’ve corrected it.

    George Ou

  26. I agree that different power supplies will have different efficiency curves and that’s why I asked you where you got the values for the Tagan EasyCon U15 from. You started off with a guess of some random 600W+ PSU and I’ve shown you a test of one that works at >70% efficiency when measuring 59W at the AC input. Now I’m asking for an explanation why the Tagan used by Tom’s Hardware should perform worse (actually much worse than any PSU tested by silentpcreview in that range) efficiency-wise.

    What I think you’re doing is just reapplying your results to Tom’s measurements and calculating the efficiency of that PSU based on the assumption that there can be no significant differences in the actual power consumption of the setup, which is what I’ve been criticizing all along. That’s about the strategy of reasoning the catholic church used to defend the flat-earth model over centuries.

    The reason I keep repeating my arguments is that you choose to ignore hard facts. Heat dissipation < = power consumption is a physical law.

    You also refuse to disclose the kind of load you used for your power measurements. If you chose something like SuperPi, for example, you’re not putting any significant load on the Northbridge, which is the real power hog on that board. It’s CPU-centric with good cache-locality. Try again with 3DMark. Also put the board into a small, closed case and run the load for a couple of hours, if it doesn’t make any difference in your opinion anyway. I’m telling you it does.

    Stop trying to patronize me and have an open mind.

    P.S.: 36% load is still wrong in the update. I think you meant 3.6% load.

  27. My guess of 600W + is fairly close since the actual PSU is 530W.

    The Tagan PSU tested by Toms and the 500W PSU tested by Jack both resulted in close to 60W power consumption. I have power supplies that test the same system I have at around 50W but it drops to the mid 20s when I test with a good PSU. Those power supplies simply don’t handle single-digit workloads well while the one SPCR tested was quite impressive for such a large PSU. I don’t know why you’d think this was strange.

    You take theoretical measurements and you try and use that as evidence. I’ve done significant data gathering and testing and I’ve come up with a very reasonable hypothesis as to why Tomshardware got the power measurement they got. I would bet money that if Tomshardware used an SPI220LE for the exact same system, they’d get similar results within a few percentage points which still puts them well within the sub-30W input power range.

    I have always published my numbers and if I’ve cited Jack’s data, I’ve always published his methodology. Take this for example: http://www.formortals.com/Home/tabid/36/EntryID/53/Default.aspx

    Jack uses the OCCT 2.0 and I’ve used WPrime in the past though I’ll be switching to OCCT because that generates a higher power draw.

    George Ou

  28. Good artical with in depth knowledge. I have not read Tomshardware for a long while. They may not be biased, but they are wrong from time to time. I just stick to Anandtech.

  29. Despite the confusion and misunderstandings, this was a very good discussion. I learned quite a bit from it, more then other places intending to inform others about power consumption and power supplies. It also gives me a better idea on what to look out for when reading publications.

    Thank you Thorn, George, and Jack

    P.S. When you get a confusing (or confused) question, ask a question back. It saves a lot of frustration, plus it can save from alienating your contributors/readers.

  30. I don’t remember how I stumbled upon this site, but it was a fun read. Fun in that George thinks he has a clue about what he has measured vs. Thorn who is clearly right in everything he’s saying but keeps getting his statements ignored and hand-waved over post after post.

    George, Jack, why don’t you just measure the current supplied to each rail by the PSU and multiply by the rail voltage and voila, you will get instantaneous power. If you don’t have an ammeter, why don’t you just put a like a 500W load onto the 600W PSU in addition to the board + hard drive in your tests, so whatever you’re claiming about the inefficiency of the PSU below certain load will simply disappear. You will realize that Thorn is right.. If you don’t know howto add such loads, may I recommend either LOTS of ceramic high wattage resistors, or like a dozen 12V light bulbs, or maybe even like 40 hard drives.

    Good day everyone.

  31. It completely slipped my mind to question what power supply Tom’s was using for their test until I read this article. Thanks for bringing it to attention.

    In case anyone is still interested, Tom’s published a Atom vs. Athlon comparison a few days ago. This time around, they used a Coolermaster RS850-EMBA 850W power supply. I have no idea about the efficiency of this PSU, but it can’t be spectacular.

    http://www.tomshardware.com/reviews/Atom-Athlon-Efficient,1997-10.html

    Their conclusion is that the platform with Athlon running at 1 GHz is more efficient and beats the Atom at most benchmarks may likely be valid since they used the same PSU for both. It is disappointing that they did not choose a better power supply for this test despite earlier criticism.

  32. 850W PSU is pretty horrid for an Atom-based system. Yeah I did see it and it was pretty stupid despite the fact that I had extensive talks via email with them.

  33. Hello,
    I stumbled upon this site and I think it was interesting to read. I must say though that if you are the gentlemen in the picture above your image belies your attitude when posting. I believe the inherent communication block both sides were having with Thorn’s argument is that both parties were using 2 seperate arguments and not using any of the other’s to address the case at hand. I also think that some posts were a bit patronizing to Thorn (No, I don’t know him). You can address the posts as elitists or as advisors however if you post something asserting that you are OK with someone questioning results or methods; don’t call those questions or comments silly.

    There is a stigma that ego plays a large part of poor communication. I can see that George, Justin, and Thorn all seem pretty smart so language plays a big role in credibility. I found myself wondering who is right and just not caring once it was clear that either side would submitt.

    Why not de-escelate the situation and someone run a dern test already. I don’t care who does but I’m really curious as to the results. Those of us that aren’t mathmaticians or engineers are dying to see something crazy. Or maybe I’m on the wrong website and wrong post.

    Respectfully and with earnest curiosity,
    Derek

  34. When we get someone like Thorn criticizing us when Thorn has done ZERO testing when Jack and I have done the testing, it gets very frustrating. What Jack and I say are based on sound evidence and what Thorn is saying is based on unsubstantiated theory.

    If you’re wondering who’s right, why would you afford Thorn the same level of trust when he ADMITEDLY did no testing? It sounds like you missed that portion of the conversation.

    Note that Justin (the person in the photo top right) has nothing to do with the conversation.

  35. I have not tested anything, but neither have you (correctly)

    Looking at the input power used (with probably a kill-o-watt meter or something) is not an accurate test. And no site (excluding silentpcreview testing of PSUs) does this test accurately.

    You state:
    "The most accurate result is to use a sub-100 watt PicoPSU power supply which can do 80% or more efficiency at 20 watt output load. In that case, a full Atom/945 system with 3.5" hard drive should get around 23 watts of input power at the socket."

    No!! The most accurate test would be to measure the current on all the rails on the motherboard itself. Which is something you have not done.

    So I googled around a bit and came across this.

    http://download.intel.com/support/motherboards/desktop/d945gclf/sb/e35968001us.pdf

    Read page 51.

    "For example, for a system consisting of a D945GCLF board (with integrated Intel Atom
    processor at 4 W), 2 GB DDR2 RAM, one hard disk drive, one optical drive, and all
    board peripherals enabled, the minimum recommended power supply is 75 W. Table
    20 lists the recommended power supply current values."

    Now why in the world would they recommend a minimum 75W supply for all of that.

    You say yourself that a sub 100W power supply would be even better efficiency, so lets say a 75W power supply is only as efficient as your 200W supply @ 73% efficiency. that would give about 55W to the board. With your numbers, subtracting 9W for the hard drive, we’re down to 46W for the board.

    So intel is saying that 46W is minimum recommended to operate this board and you’re saying 9.5W. I am sorry but that’s just laughable. You know how happy they would be if they can say minimum required power supply is 30W. They would have a field day…

    It seems from many sources that board consumes about 5-8W more during full load than it does when idle. You cannot throw your "inefficient power supply" math at this since this is a delta of max-idle and such a small change in load would not alter the efficiency by enough to make a difference.

    That means that your 9.5W rating (which I’m assuming is under full load) would be 1.5-4.5W idle. which would be almost less than the fan :)

    Those sources are listed at http://www.linuxdevices.com/news/NS4407123606.html Under Power Consumption.

    So good luck and try again :)

  36. To E.

    My original numbers accidentally excluded 3.5" hard drives and multi-thread tests. The correct output load using the SPI220LE is ~27W with 3.5" HDD and ~20W without 3.5" HDD. Note that the SPI220LE got more efficient with the 3.5" HDD because it moves the output power closer to the ideal range. The load testing tool was OCCT 2.0 running dual threads which is not quite TDP levels but pretty darn close and a lot more intensive than any application other than maybe LINPACK.

    But even with LINPACK or the Intel TAT utility, it’s doubtful that the power would go up more than one watt because the total power consumption of the Atom is so low to begin with. That puts the overall output power well within 30W even with a 3.5" HDD running idle. The 3.5" HDDs are capable of running up to 30W during spin up mode which knocks the total system output power up to 50W but we can avoid this by using 2.5" HDDs and stay well within 30W output power even during spin up.

    Intel quoting 46W minimum output power is pretty dead on since it MUST factor in for 3.5" spin-up cycles is fairly accurate. It’s strange that you would compare this 46W number to the motherboard/CPU power consumption.

    As for your Linuxdevices.com link, it quotes HardOCP numbers of 38W for the Atom system in idle which 4W from Jack’s idle using the Sparkle SPI220LE and about 6W from my idle with the Sparkle 55W open frame PSU. That is consistent with my point that Jack and I are using the most efficient power supplies to get the closest possible numbers to the actual output power. All those much higher numbers that Linuxdevices.com quoted are insanely wrong because they’re using oversized and inefficient power supplies.

    As for the recommendation of a 75W PSU, that’s a standard conservative number to recommend to ensure a big enough safety margin. If I was the maker of this board, I’d say the same thing to avoid people coming back to me complaining. But the raw numbers I’m quoting here have zero safety margin so you can’t compare these two numbers.

    Lastly, don’t be a wise guy at the end of your post.

  37. As for the title on linuxdevices.com that "Nano smashes Atom", that conclusion is based on single-threaded performance testing that doesn’t even factor in the clock disparity with Nano at 1.8 GHz and Atom at 1.6 GHz. You do realize that in the NetBook market where the Atom uses a much more power efficient 945GSE chipset, Atom runs at 1.6 GHz at 2.5W or 4W TDP (1.9GHz with turbo button on some NetBooks) while Nano runs at 1.3 GHz at 8W TDP. At those stock speeds in the NetBook market, Atom would beat Nano on single-threaded tests.

    Furthermore on dual threaded tests, Intel Atom 1.6 smashes Via Nano 1.8 by a respectable margin and ExtremeTech shows the numbers here (http://www.extremetech.com/article2/0,2845,2326856,00.asp). Now this is a real shock to me because Via Nano (Isaiah architecture) has 94M transistors which is double the 47M transistors on the Intel Atom (Silverthorne architecture) which would imply that Via Nano should run faster clock-for-clock. But it’s clear that just the reverse is true with Intel running faster using half the transistor count and lower clock speeds when you fair test that takes advantage of the full potential of the Intel Atom processor.

    The reality is that Via did a great job with the limited resources they had in developing the Isaiah architecture and it’s a great drop-in replacement for older C7 processors, but let’s not kid ourselves or the public that it’s a better architecture than the Intel Atom.

    If this level of biased reporting wasn’t bad enough, Engadget linked (http://www.engadget.com/confirm/5ceaea3384ee610e499b7a32166d7467/) to a YouTube video from Via showing problems with Atom Netbooks having problems with Windows Media 1080p videos and it turned out that there was something wrong with the Atom NetBook that Via tested. Even after this was proven wrong http://www.youtube.com/watch?v=JVf3PtNF-j0 and Engadget was forced to acknowledge this, they had to throw a parting shot that Atom performance sucked. Now can you imagine how badly the media would skewer Intel if they had pulled a stunt where they put up a video of a defective Via Nano NetBook choking on video playback?

    I realize that people have a natural tendency to want the underdog to win, but these wide spread biased reports all over the tech media are flat out sloppy and even reprehensible.

    Media bias in favor of the underdog is nothing new and I even recall some readers telling me that they think they should be able to factor in their own biases. I agreed with him 100% but that I still believe he should be entitled to unbiased information so that he can add whatever bias he wants later. What should not happen is the media forcing their biases on the entire public.

  38. My links to those sites was once again to show the difference between max loaad and idle which cannot be affected too much by whatever PSU they were using. And I also stated that no site was doing a proper power consumption test.

    It’s good that you’ve corrected your results, which are becoming more inline with all the specs. As Thorne was saying, you maybe maxing the cpu fine but that’s not the component that draws the most power. We all agree that it’s a 4-5W device. Problem is the northbridge, and you need to apply proper loads on that, including the graphics. I think if you did that, you would get closer and closer to the 45-50W range.

    My comparison of the Intel 46 W number was for the board. not board+harddrive, for which I took used your 9W for the harddrive number. It was 55W total.

    And lastly, yes manufacturers give wiggle room in their recommendations, but not by 2x.. If they did that, than everybody would be running power supplies much more than required and would be hit by the power supply inefficiencies you cite.

    I’m not trying to be a wise guy.. Just pointing out that we don’t need to run "tests" to see problems with someone who has.

  39. It depends on what your application is. I would not use the 945GC/Atom as a graphics platform, I use it as a server or low-end computer which almost certainly consumes less power than the updated numbers OCCT 2.0 numbers I quoted.

    The board + memory has an output load of ~20W using an OCCT 2.0 dual-thread load and we can probably get it up to ~30W or possibly higher output load if we cranked the 945GC graphics engine. I really doubt you can get a graphics load to push the board up another 20W above or beyond my updated numbers and I think there’s a LOT of headroom in that 46W number. Maybe you could find some kind of synthetic load that stresses the North Bridge, GPU, and CPU all at the same time and get close but that’s a full 26W above the 20W measurement which is close to the TDP of the entire 945GC chipset so I doubt it.

    All of those tests you quoted on LinuxDevices.com were non graphics intensive and my point was that all of them except for HardOCP botched it badly using horribly inefficient power supply. My original point was that Tomshardware’s test was a failure due to the inefficient and poorly sized power supply. Thorn criticised my critisism of Tomshardware without running any tests and that’s what I objected to.

  40. Thorne criticized your numbers, and he also criticized tomshardware. That wasn’t the issue. He saw some unrealistic numbers on your tests based on known specs of the hardware. And you kept on defending yourself and throwing away his valid suggestions without giving any valid reason besides "I performed many tests, you didn’t perform any tests, so I know more" logic.

    Now a month later your results are almost 2x what they were before, and instead of saying, thanks to valid posts that made me realize I should look back at my numbers, you’re still saying you know it all and others don’t have a clue, and they should not speak since they didn’t run any tests.

    I didn’t run any tests, but that didn’t prevent me from seeing holes in yours. So stop trying to come out on top. You went through so many rants trying to defend your original numbers and outright calling Thorne an imbecile. Apologize to him, and let it be.

    It’s also hilarious that your last post is now throwing TDP numbers into the mix :)

    This will be my last post.

  41. Not entirely true E. Thorn insisted that the Tomshardware numbers aren’t that wrong when they were that wrong and continue to be wrong.

    The TDP numbers I’m throwing is jut to show that the board/CPU still don’t add up to the 46W that you’re suggesting. You threw out a lot of numbers without testing that clearly don’t look right. I never said I knew it up; just that I knew more because I ran tests. What I objected to is that I’m talking measurements while Thorn is talking TDP which are not the same thing. I never called Thorn an imbecile or any names, I only rejected his comparisons of TDP.

    I’ve fixed my mistake and apologized for posting it.

  42. Been reading these articles – it seems like only the total power consumption figures are being accused of innaccuracy. This blog post doesn’t seem to question the relative power consumption between the two cpu/mobo’s on an identical platform.

    Also, just commenting – Tom’s posted their specification. Technically, as long as it’s reproducible, doesn’t that mean that their numbers are accurate? The real premise seems to be a preference to use a smaller PSU.

    I’m just not making the connection that this is even relevant – unless there’s an assertion that the relative performance/power consumption results would be changed.

  43. FYI. My MSI Wind PC barebone system with 2GB RAM, a 750GB HD, no DVD, OS installed on CF card, eats only 20W while downloading to USB flash drive (bittorrent). Peak is 30W while HD is reading/writing.

Comments are closed.