POST A COMMENT

49 Comments

Back to Article

  • Rick83 - Friday, February 28, 2014 - link

    It would be interesting to get efficiency not just as a percentage, but also as an absolute value of thermal loss in the PSU. Especially for the idle loads, this is more interesting than the relative efficiency. Reply
  • JarredWalton - Friday, February 28, 2014 - link

    What you're asking for is for us to just provide a second graph showing PSU thermal load (load * (1 - efficiency)), right? If I understand you correctly, we should be able to add such a graph without much difficulty. I'll run that by "E" and see what he thinks. Reply
  • edzieba - Monday, March 03, 2014 - link

    Another option would be a graph of output power vs. power draw (with a 1:1 guide line). This would give a more intuitive view of how much of an effect (or the lack thereof) low efficiencies at low loads have. Reply
  • Rick83 - Friday, February 28, 2014 - link

    Oh, and something I saw other sites do, which might be interesting: Test not only at load levels relative to the absolute performance, but at some "common" load levels, such as 30W for an idling desktop system, 150W for a CPU loaded desktop system, 300W for a gaming system, 500W for a SLI/CF system, and then maybe add 800W and 1000W for the bigger PSUs to simulate tri/quad GPU systems and mining rigs. This way PSUs can be compared how they perform for a fixed load, which is more important when making a buying decision for a certain system. Reply
  • Guspaz - Friday, February 28, 2014 - link

    Great idea. If I'm trying to select a power supply for a system, it's way more interesting to me to know (for example) how loud each of them would be at 300W load, rather than at 50% (which would be different wattages for different PSUs). Percentages are of limited utility for comparing between PSUs unless they all have identical ratings, which isn't what happens in the real world. As in, if a power supply rated 100W higher is much quieter, it might be a better option even if the extra power isn't required. Reply
  • E.Fyll - Friday, February 28, 2014 - link

    If I would replace percentages with wattage, the shape of the graph would remain entirely unchanged. If you do want to compare different units, you only have to check the percentage that corresponds to the load you want to check. It's a simple multiplication. Reply
  • Cellar Door - Sunday, March 02, 2014 - link

    E.Fyll - the seasonic review is one of the best PSU reviews I have read in a while, awesome job. Very clear and easy to read. Could not ask for more! Reply
  • ShieTar - Monday, March 03, 2014 - link

    I think what Guspaz is getting at, and what is not very clear in the above article, is that he wants to see actual testing at other loads than 20%, 50%, 80%. From the actual test you posted in parallel it looks to me like you do indeed test at the entire load range, with much more than only 3 specific load cases. Maybe you just need to make that a bit more clear in the above text? Reply
  • E.Fyll - Friday, February 28, 2014 - link

    That's not really good practice. First of all, none of the systems that you described do reach such power figures. For example, a single gaming system is between 160 W and 220 W at full load while a triple SLI system may reach anywhere between 400 W and 700 W. There are great variances between systems, plus it would not be equal testing conditions for every unit. It is far more reasonable to test the unit at specified intervals. Most charts are provided across the entire load range, so you can easily assess the performance of your unit, assuming that you do know the power of your system. Reply
  • Rick83 - Friday, February 28, 2014 - link

    While it may not give equal testing conditions, selecting a number of fixed points for taking measurements does allow for easier alignment of the different efficiency curves, and for easier readability and comparison between the different units.
    Also, the numbers I gave were just one example of a spread of useful points - any number of reasonably spaced points in the 10-1000W gamut will do the job.

    Alternatively, an absolute power based graph might also work, just to do the alignment, but just 4 load points on a 1000W PSU might not be enough to compare it against a set of 800 and even 600W PSUs, which might be cross-shopped for a system that has a TDP-sum around 550W. (According to the various 50%/80% rules and whatnot that many believe in)
    Reply
  • doctormonroe - Friday, February 28, 2014 - link

    Could you also provide the RPM data in the reviews? Reply
  • E.Fyll - Friday, February 28, 2014 - link

    Although I do not see how RPM alone would be of any use (noise is what matters, no?), yes, I can do that. I will think about it. Reply
  • ThomasS31 - Friday, February 28, 2014 - link

    Well, it could show maybe lifetime and noise/fan-lifetime expectancy relation for a given unit (~suspect :) ). Even if something at first may sound less noisy when new.... could become faster noisier.

    What I mean, if something uses a high RPM, the fan lifetime (especially the bearing/sleeve) could be decreasing faster... also fans tend to get noisier by time, so high RPM usage could cause more noise earlier in a unit/fan's lifetime.

    (Like small motherboard/notebook fans, that are silent in their early life, but "we all know", the high RPM they use make them an issue very soon.)
    Reply
  • bobbozzo - Friday, February 28, 2014 - link

    Hi,

    If you cannot test at 110V/60Hz, how "wrong" would it be to test at 110/50Hz (using your Variac)?

    BTW, nominal Voltage in the US & CAN is 120V (not 110V) +- 5%.
    However, it is apparently 100V in Japan, with some parts 50Hz and some 60Hz.
    https://en.wikipedia.org/wiki/Mains_electricity#St...

    I've seen some reviewers stress-test with under-voltage and over-voltage as well.

    Thank you for the article!
    Reply
  • E.Fyll - Friday, February 28, 2014 - link

    I cannot tell exactly how wrong it would be but it would be wrong, especially because each unit will react differently to the change of frequency. Some units might actually perform (and they should) very similarly between 50 Hz and 60 Hz, while other units could have their conversion efficiency dropped by up to 4-5%. Although the RMS voltage remains the same, the shift of the frequency would affect the PFC circuit and the ripple that will appear on the secondary. The difference would primarily depend on the primary rectifier gate(s) and the ripple suppression design. I cannot foretell how each unit will react without actually testing it. Even though the difference should be small, I find it misleading to add an improper test into my articles. I will be doing such tests (including some stress testing) once I can get a hold of a high power programmable AC source. Reply
  • bobbozzo - Friday, February 28, 2014 - link

    Also, a constant-voltage transformer would be more reliable than a Variac, esp. if the voltage is changing during the testing.
    If it's an isolation transformer as well, it would have the added benefit of reduced chance of electrocution, esp. if you plugged all the test equipment in.

    However, even a 100W constant-voltage transformer can be very large.
    Reply
  • E.Fyll - Friday, February 28, 2014 - link

    Not really. Isolation transformers are 1:1; you cannot adjust them, you are getting exactly what your distribution grid supplies to you. What I wanted to do is get my 225-235 Volts exactly at 230 Volts and the VARIAC works just fine for that. CVT's are ridiculously expensive, I could buy a programmable AC source for that kind of money, so they are out of the question.

    Thank you for being concerned about my safety. I'll be careful, I promise. :)
    Reply
  • Rick83 - Friday, February 28, 2014 - link

    After reading the first of the series of review, there's one more thing, that might be worth testing: Whether the device is actually ErP6 compliant, if it is advertised as such. And I also didn't see a testing of the OPP/OTP/OCP/OVP circuits. Where/when/how these work would be of interest, especially in low-end units. Reply
  • E.Fyll - Friday, February 28, 2014 - link

    Indeed; I do need more equipment to test ErP6 compliance though, so that will be added in the future. I do test the safeties of the PSU and will note it if (when) a PSU fails a specific safety test. Actually, it is the first thing I test; I do not want to plug my testing equipment onto a PSU that fails to meet the basic safety standards to being with, these things cost several thousands of dollars. Reply
  • Mr Perfect - Friday, February 28, 2014 - link

    Sounds like a thorough test. It reminds me of SPCR's PSU testing, though it looks like you might have some better funding for your gear. :) That being said, SPCR is able to separate their load from the PSU somehow and do live acoustic testing(rather then measuring RPM and measuring the fan later). It may have been as simple as passing the PSU leads through a wall to the loading system, but I don't remember now. Any chance you could do live acoustic testing too? Reply
  • E.Fyll - Friday, February 28, 2014 - link

    Even if it would be possible to add 30 centimeters (about a foot) of insulation material between a unit and the loading/testing equipment, you would be getting readings at over 40 dB(A) on the side of the PSU. The equipment generates a lot of noise, it simply is not possible to isolate the load from the PSU entirely, it will destroy the test results even if there is a wall of sound-absorbing material in between those two.

    Plus, in order to do such a noise test, you need to have a clearance of at least two meters from every wall and ceiling, or you will be getting false results due to echo and vibrations. So, you need at least two meters and a wall of sound-insulative material in-between the PSU and the testing equipment. That's simply not possible, even if someone would give me a few millions of dollars in funding.

    Ok, I lied there, I can come up for a solution for that but it would really need about a half a million dollars. That's how much a noise and vibration isolated, pressure and temperature controlled room large enough for a PSU and with a fixture/equipment added to it would cost. Assuming that you have a space to build the room upon and not buy a new piece of land.
    Reply
  • pvdw - Friday, February 28, 2014 - link

    SPCR did a DIY hemi-anechoic in 2008 for just a few thousand dollars. Though that's likely not counting the large amount of labour involved. That's possibly a bit too far to go at the moment, but maybe there's an anechoic chamber not too far away that you could make use of? Reply
  • E.Fyll - Friday, February 28, 2014 - link

    That's not nearly close to what I consider usable, sorry. Extremely far from noise and vibration isolated, no control over ambient conditions. Even if I could make a room capable of meeting proper requirements, the metering equipment alone would cost tens of thousands of dollars. Why on earth would I want to spend such funds on the futile attempt to provide end users with "results" that are of absolutely no use to them? I would rather do testing in a real room and by using a normal SPL meter, displaying real-life conditions as a practical reference to common consumers, than doing tests that are not quite correct and not realistic either.

    Plus, tens of thousands of dollars worth of equipment aside, I am not willing to waste an entire room of my house just for trying to look cool.
    Reply
  • Mr Perfect - Thursday, March 06, 2014 - link

    Hmm, okay, strike that then. Is there some way to test for electronic noise from the PSU then? I've heard PSUs make some irritating coil whine. It's rare of course, but completely infuriating. Reply
  • ssj3gohan - Friday, February 28, 2014 - link

    As a PSU designer I have always cringed at least in some respect at the vast, vast majority of PSU reviewers for basically not understanding what goes into proper PSU testing. It's so refreshing and exciting to finally see a site doing everything *right* for once. I'll still be skeptical until I see more reviews, but this new PSU reviewer person (do you have a first name?) seems to be the best in the reviewing business, surpassing x-bit labs and jonnyguru. Reply
  • E.Fyll - Friday, February 28, 2014 - link

    Better than Jonnyguru? Repent!

    Ok, seriously, he is the only author that I have ever seen who is actually using proper, 100% "guideline-compliant" equipment to test PSUs and an inspiration to me.

    I actually need more and better equipment to truly be able to fully test PSUs but the cost is in the tens of thousands. Such an investment will obviously take time.

    Of course I have a first name. It just is much uglier than my surname. ;) Feel free to call me Fyll.
    Reply
  • extide - Saturday, March 01, 2014 - link

    What do you think or HardOCP's testing method? Reply
  • E.Fyll - Saturday, March 01, 2014 - link

    I do not comment on the work of other people, sorry.

    The required methods and equipment specifications are not hidden at all. Anyone can easily have a look and see who goes anywhere close to them.
    Reply
  • ssj3gohan - Saturday, March 01, 2014 - link

    How about 'Dr. Fyll' ;)

    Jonnyguru is not the only other reviewer with proper test equipment, but the only one who knows how to use it and has enough background in electronics to be able to troubleshoot weirdness. X-bit labs has a proper electrical engineer with (I guess) too little budget to afford a compliant test bench.

    As for compliant hardware for testing; for the love of god, don't get hung up on specs and recommended test equipment lists because you will pay through the nose to get it. It's often much faster and less of an investment to get lesser equipment properly cal'd or even modify it to satisfy requirements. As long as it's electrically equivalent and you are not using different measurement principles this is perfectly fine. And instead of having to throw down 50k$ on a Chroma test rack, you only need to pay a couple hundred dollars a year plus shipping to get it short-term calibrated.

    By the way, are you going to use error bars? Please use error bars. So many sites use 4 significant digits for efficiency with a total error of more than 2%...
    Reply
  • E.Fyll - Saturday, March 01, 2014 - link

    Actually, the recommended specs are there for a reason. There is also a reason why getting something far less expensive and modifying it is "about the same". "About the same" is not the same, especially when we are talking about ultra fast and sensitive equipment.

    For example, Intel recommends at least a 20 MHz oscilloscope with a sampling rate of 200 MSa/s. I am using a 40 MHz scope with a sampling rate of 500 MSa/s and it still is not good enough for some tests. The "about the same" approach of many people is the use of a 0.15 MHz oscilloscope with a maximum single-channel sampling rate of 20 MSa/s. Those that actually do have and use even those, because such a device has a resolution of 2 mV and thus can produce only even numbers (I have seen some people reporting odd numbers too). Oh well. Accuracy aside, this is ten times lower than the minimum recommended spec; it is not inaccurate, it is entirely wrong, the instrument cannot read the correct electrical signal to begin with.

    I do not need a Chroma test rack, because that is just a convenient device to have but not really any better than the electronic loads that I have already purchased. Far cheaper maintainance aside, I can calibrate them on my own too. For some parts however, there is no cheap workaround, you really need to shell a lot of money or you are better off not trying at all.
    Reply
  • DanNeely - Saturday, March 01, 2014 - link

    Before spending lots of money shipping expensive hardware back and forth to maintain calibration, I'd recommend getting quotes on having it done on site. After a several thousand dollars repair bill for a high end oscilloscope that was damaged in transit my employer discovered it was cheaper to pay a company 2 hours away to have someone drive out to cal about a half dozen bits of RF hardware and a few torque wrenches than it was to ship the same hardware to their office and have it done there. Reply
  • pvdw - Friday, February 28, 2014 - link

    Great to see some real study being put into prepping for these reviews!

    Just one thought, your noise floor seems rather high by modern standards. I know Anandtech focuses more on gaming PSUs which are generally noisier, but you're not going to be able to separate quietish PSUs from truly silent ones until you can go much lower.

    I believe these dBA levels are right:

    0          The softest sound a person can hear with normal hearing 
    10         Normal breathing 
    20         Whispering at 5 feet 
    30         Soft whisper
    Reply
  • E.Fyll - Friday, February 28, 2014 - link

    Actually no, these figures are common misinformation going around the internet. The floor noise inside sound-recording studios is 18-22 dB(A); just a hint higher than what the equipment of an ISO testing lab would read with you standing in the middle of the room, just because you are breathing and your heart is beating. The floor noise of a common household room usually is over 30 dB(A), commonly near 40 dB(A). You cannot even measure figures much lower than that without a noise and vibration isolated room, let alone specialized equipment that costs tens of thousands of dollars. Even assuming that you can create a perfect, undisturbed low-noise environment anywhere near a household, it is extremely difficult to acquire such a low power energy signal and convert it to a usable electric signal with any sense of accuracy, especially when even the vibrations of the air's molecules will affect the equipment, let alone things such as electromagnetic fields and wireless signals.

    Also, dB(A) figures are not linear and stand-alone, they are logarithmic and additive. This means that if you have a floor noise of 30 dB(A) and add a 20 dB(A) noise source into the room, the floor noise will not remain at 30 dB(A); 20 dB(A) will be added to it and not linearly, logarithmically. So, a product that tests at 20 dB(A) inside a zero-dB(A) environment of a lab will not read 20 dB(A) when testing it inside a common room but it will bring up the floor noise of the room it is being tested in. Not perfect, but it is realistic testing and all we can provide for reference. Replicating ISO lab conditions in my household and acquiring tens of thousands of dollars worth of equipment = not going to happen (unless of course if I become a billionaire, in which case I promise you that I will do my best to build my own lab in my huge new back yard).

    When a manufacturer says that a fan is rated at "17 dB(A)", it does not mean that you will be hearing 17 dB(A). That fan generates 17 dB(A) in a zero dB(A) environment (ISO test lab). You will be hearing the floor noise of you room *plus* 17 dB(A). Per 3 dB(A) the magnitude of noise pressure doubles.
    Reply
  • Death666Angel - Friday, February 28, 2014 - link

    Which is why dB / dB(A) really suck for measuring loudness for the human hearing. I've liked sone based measurements, though it's just as imperfect, because our hearing is imperfect and varied and subjective and so on. :D Reply
  • E.Fyll - Friday, February 28, 2014 - link

    Actually dB(A) readings are OK for basic reference and comparisons. Sones are somewhat difficult to be perceived as a figure by most people and ridiculously complex to measure at <40dB levels. Not that I do not think of them as a better choice when you want to indicate actual loudness, just that it is unrealistically difficult to introduce them into reviews (with any sense of accuracy, at least). Reply
  • ssj3gohan - Saturday, March 01, 2014 - link

    SPCR - at least the most reputable in noise measurements - also points out that noise bandwidth is a big contributor to perception and it is much better to have up to 6dB higher wide bandwidth noise (whoosh sound) than a concentrated tone (buzz/beep). That's something that definitely needs to be evaluated in a proper fan testing environment.

    Which brings me to something you said earlier: testing the fan in isolation of the PSU. I don't know if that's a better way of doing things, because aerodynamic obstructions (fan grating, heatsinks) as well as reflective or absorptive surfaces are going to change the noise signature quite a bit, especially when dealing with badly designed fan gratings.
    Reply
  • E.Fyll - Saturday, March 01, 2014 - link

    dB(A) are the acoustic frequency; you can hear any frequency that is part of it. Perception however truly is a vital factor; other people are annoyed by high frequency noises but others will not even perceive them. I cannot quantify that however, it is mostly psychological.

    You are absolutely correct, you should not test the fan on its own, otherwise you are not really testing the noise of the PSU. That is why I do not remove the fan from the unit; the unit is just as complete as ever, with its fan hotwired to an external, fanless PSU.
    Reply
  • Death666Angel - Saturday, March 01, 2014 - link

    The only sound measurements I trust in reviews are those done by the German magazine c't which has done sone measurements since I remember. :D Since it is a linear unit it is also much more intuitive to the way most non-science people think. Reply
  • makerofthegames - Friday, February 28, 2014 - link

    This is why I read Anandtech - they actually know what the hell they're talking about, and they test things with actual tests, not just plugging it in and seeing how fast Crysis is. Reply
  • Death666Angel - Friday, February 28, 2014 - link

    Sounds like a good start, will read the full Seasonic review when I find the time. Up to now my primary PSU site was jonnyguru and you are now approaching their level of devotion to PSU tests. :) Reply
  • Death666Angel - Friday, February 28, 2014 - link

    Just wanted to say that the nVidia side ads are really annoying to me. I browse the website in a half-screen format (Win7 feature on a 1440p monitor) and I've click on the "ads" space 3 times today while I tried to get the Firefox window into focus again to scroll down an article. Really annoying. Reply
  • dealcorn - Friday, February 28, 2014 - link

    Over time, Intel may capture some (or most) low end desktop market share with its Silvermont or better Atom core products that have TDP <= 10 watts and SDP <= 5 watts. Assuming 4 GB 1.35v DRAM, a ssd that supports DevSleep or comparable, and a 2.5" spinning drive, a passively cooled system may typically draw 10 watts or less before considering power supply efficiency losses. This is wall wart territory. In the next 3-5 years I expect 100's of millions of power supplies will be sold for such systems. You want your power supply operating at a 20%+ load for efficiency purposes. Who makes the most efficient 25 and 50 watt wall warts? Due to the popularity of P2P daemons, many of these systems are powered on 24/7. While the savings vary depending on local electricity costs, they add up over time. Please provide some testing of wall warts to focus on this emerging market. Reply
  • nurd - Friday, February 28, 2014 - link

    I'd like to see measurements of how much RFI crap the power supply spews back out the AC line. Some of us like having computers running, but still having a noise floor below the ceiling on MF and HF! Reply
  • alexvoda - Saturday, March 01, 2014 - link

    Maybe someone asked this already but I would be interested to see how they perform in a NAS scenario.
    AFAIK many PSUs can not handle the load of many HDDs (e.g.: 8 drives) spinning up simultaneously at start up.
    Reply
  • stlc8tr - Saturday, March 01, 2014 - link

    What would also interesting is if there was an article comparing a "quality" PSU with something of lesser quality. I see a lot of advice that sticking with a quality ODM like Seasonic is advisable but what does that mean in real life terms? What are the implications of using a 650W Seasonic vs the cheapest 650W available from Newegg? What are the performance or safety deltas? Reply
  • A5 - Sunday, March 02, 2014 - link

    Jonnyguru used to do that sometimes. The cheapest of the cheap would tend to, uh, catch fire if loaded them to their labeled wattage. Reply
  • ShieTar - Monday, March 03, 2014 - link

    Sadly, thats very true. German former magazine, how home-page Hardware-Luxx has also done this on some occasions, apparantly the labeled watts on most no-name PSUs are just false advertising. Of course, the situation should be a bit better on the cheapest PSU which does actually hold a 80+ certification. Also those things have the tendency to specifiy half their wattage at 3.3V and 5V, and there seems very little sense in testing 300W loads at 3.3V. Reply
  • jmke - Wednesday, March 05, 2014 - link

    in my experience PSU efficiency and performance is directly related to ambient temperature. Will you test efficiency at different temperatures? Reply
  • jmke - Wednesday, March 05, 2014 - link

    ignore my comment

    *The difference between our cold and hot tests is an increase of over 20°C ambient and, naturally, all of the temperature readings increase significantly.*

    excellent :-)
    Reply

Log in

Don't have an account? Sign up now