500W to 550W: 12 Power Supplies Compared

62 Comments

View All Comments

• mindless1 - Monday, February 23, 2009 - link

  Changing everything for the sake of change, thinking "present" SATA connectors would have any usefulness, is a bad idea. Change things when there is a good reason, only.
  
  How is it annoying to use a connector when you still have to with (what, different connector is no less?) any option, or did you plan on transmitting power with a laser beam? It makes no sense, this is very basic easy inexpensive standard blah blah blah.
  
  Case fans often use the large molex connectors because a case manufacturer can't be assured how many fan headers a board may have or where those are. They could just put very long cables on the fans, but then somebody comes along and says long cables are untidy.
  
  Ultimately you seem to be opposed to seeing wires, want everything invisible. Esthetically I agree but practically speaking none of these issues really matter to the operation of the system, just be sure the wires are secured out of the way of fan blades.

• strikeback03 - Tuesday, February 24, 2009 - link

  I was pointing out that with the exception of fans, pretty much everything in the system which used to use Molex uses the SATA power connector now. No laser beams involved, Sata power cables for HDD and optical. The only place I have seen a Molex recently is on the occasional motherboard that uses it for additional power. I suggested that rather than maintain two connectors indefinitely, they migrate everything to one of them.
  
  And for the case fans, check a Cooler Master (others might do this, but CM is where I have seen it). They come with fans with motherboard fan header connections, and an adapter to use Molex instead of a header.

• JPForums - Monday, February 23, 2009 - link

  I like the delta K chart, but I must ask, does it represent the temperature difference in Kelvins. Why not just use delta C since it is the same number and would be more consistent with how people (including yourself) typically measure temperature in computer systems. Also, while I do like the delta chart, I can always calculate my own deltas if I have the absolute temperatures. I would prefer some absolute temperature figures and then a delta chart if desired. It doesn't really matter if the exhaust temperature and the internal temperature are similar if the PSU puts out enough heat to boil water. Given the efficiency of these units, I doubt this is the case, but it would still be nice to know what the absolute temperatures of the exhaust and each heat sink is. Oh, and where did you measure the internal heat from?
  
  "The efficiency charts make one thing very clear: Seasonic and Enhance deliver kickass efficiency with their products."
  
  I'm not sure where this fits in. There is no Enhance unit on the chart in question. Further, the Seasonic unit is not exactly at the upper end of efficiency in this chart. One might say it got its ass kicked by some of the units here. I ask that you avoid comments that seem this out of place in the future. We already have enough people spamming "Anandtech reviews are just advertisements for Company X" as it is. Don't encourage them.
  
  Now that the negatives are out of the way, I have to say, this one of the best review format I've seen for a large number of PSUs. Comparison is pretty easy and intuitive. I still prefer the power efficiency curves used in single supply reviews, but for a large number of PSUs like in this round up, the chart you put up is much better.
  
  On a side note:
  Looking at the various amounts of ripple and noise on the rails of each power supply, I asked myself a few questions:
  1) How much ripple can I take before if effects my system?
  I think the spec is 12v < 120mVpp ripple and 5V/3.3V < 50mVpp ripple, but I'm wondering about differences in system stability if I decide to overclock my CPU, GPU, RAM, etc.
  
  2) Does the quality of ripple make a difference? I.E. would you rather have sinusoidal ripple (Like BFG's 5V rail) or thin spikes (like Arctic Cooling's 5V rail)?
  
  3) Assuming the noise is at most half the ripple, does it make a difference to system stability?
  
  4) Assuming they are implemented correctly, can more phases on a motherboards VRMs make up for a PSU with higher ripple in comparison to a 3 phase board with a PSU that has low ripple? Alternately, how much voltage ripple actually gets to the processor, north bridge, etc.?
  
  I'm not expecting a separate review or anything, but I'd be interested in your thoughts on these topics (objective or subjective).
  
  

• Christoph Katzer - Tuesday, February 24, 2009 - link

  Thanks Mindless1 for that great explanation!
  
  As for the rest of the questions JPForums, I can use C next time of course. I just thought of using this to save space. I could include a plain table with the results of each heatsink and other temps but I am a fan of graphics you can understand instantly. Maybe I can think of another way when the tested PSUs are less. There will be three more roundups coming in spring: 620-650W, 850W, and 1200-1250W. Plenty occasions to make it better each time ;)
  
  Enhance fits in that particular sentence because four of the units are made by Enhance and four by Seasonic. The meaning was that if you are looking for high efficiency you should look out for products that are made by one of those two companies regardless of what name is written on it...

• JPForums - Tuesday, February 24, 2009 - link

  To be fair, I like the graphics as well and wouldn't want you to get rid of them as yours are very well done. I'd be happy if the hard data was in an excel sheet link in a sentence like "For absolute values, see this spreedsheet".
  
  I must have had a brain fart. I can see where you mention the units that use Enhance and Seasonic in the article. Sorry about that.
  
  Can't wait for the upcoming articles.

• mindless1 - Monday, February 23, 2009 - link

  1) Spec is (+-) 5%? 12V = 1.2Vpp, 5V = 0.5Vpp. Seems like a lot doesn't it? The thing is, any properly designed hardware being powered is designed to run stably within this range.
  
  Parts effected when overclocking are using (re)regulated power on their respective parts or motherboard, ripple from the PSU makes far less difference than you might suspect, so long as the minimal voltage isn't excessively low (excessive response time with changes in load, power state changes from CPU or GPU, etc, as often the case when a PSU barely has enough capacity for the load).
  
  2) Sinusoidal or spikey don't in themselves matter, you still have an impedance and capacitance downstream at the powered parts, reregulation for most critical voltages, decoupling at the chips, etc.
  
  3) What are you calling noise? There's switching noise, environmental/EMI/RF, digital switching noise, etc. The noise that effects a part the most is the noise that part produces itself. Given a theoretically perfect PSU with 0.00V ripple, you still have noise created in the system from the powered parts.
  
  4) More phases in the VRM would reduce PSU ripple more, but these phases themselves create noise. Practically speaking, the fewer stages or slower each switches, the more capacitance you'll need per current. Either increases costs, has different layout/space factors, and a tradeoff of efficiency versus max power handling capability.
  
  Until the PSU is performing much more poorly than you seem to be thinking about, the design of the VRM itself causes the issues, not minor differences from PSU supplying power to it (within the context mentioned previously, that the PSU at least doesn't have very slow response time).
  
  If your concern is what to do to get the best o'c, the answer is better heatsinking to keep parts cooler, then with this addt'l thermal margin you have less restraints in giving the part a higher (average) voltage, meaning whatever the largest depressed voltage is, stays high enough to retain stability at the operational frequency. To this extent, a shorter spike depression would be better than a longer period sinusoidal depression at same voltage, merely because the shorter period means downstream capacitance is buffering better.

• JPForums - Tuesday, February 24, 2009 - link

  Thanks for the explanations. I didn't realize that the VRMs would create so much noise. Since noise adds in quadrature and the oscilloscope graphs don't show nearly as much noise as ripple, the VRMs must put out nearly as much noise as the PSUs ripple to be comparable to the ripple.
  
  Does it really matter what noise I was talking about? I was mostly thinking about switching noise. Since the PSUs are fully encased, they shouldn't be very susceptible to background noise (I.E. there should be far more noise in the PSU case). The oscilloscope graphs show the noise from all the sources you mentioned. All of that noise is still on the line when it reaches what is powering it, so it doesn't make sense to me to distinguish between them.
  
  Given what you were saying about critical circuits being reregulated and excessive response time with changes in load being a problem, I think more phases (faster response) would be of benefit up to a certain point. As you eluded to, fewer phases require more capacitance to make sure the voltage doesn't droop while switching between phases. Which leads to what you mentioned about fewer and/or slower stages requiring more capacitance per current. More capacitance slows the response of the circuit to load changes. However, switching from more phases causes more noise (apparently a lot). It would be important to make sure that this noise is kept as low as the voltage droop between changes. Do you agree with this assessment?
  
  With modern boards I have been avoiding 3 phase VRMs, but I also don't really see the merits of the boards with more than 8 phases. I know this depends heavily only switching rate and capacitance as well, but I'd be interested in how many phases you think would be best (assume the switching rates, capacitances, and topologies currently implemented by manufacturers).
  
  Thanks

• Nickel020 - Sunday, February 22, 2009 - link

  It seems that you missed posting the 12V Ripple & Noise screenshot for the Corsair HX520, any chance you could post it?
  
  From your graph it seems that the Corsair is quite a bit worse in the 12V department than many other PSUs here, would you say that this has a negative impact on overclocking?

• just4U - Saturday, February 21, 2009 - link

  Hate to nitpick, but I really do wish one of the more common Antec's was thrown into the mix (500Watt earthwatts ect) Moreso because they are considered common in many builds due to casing/psu combo's and it wouldn't be surprising to see them in many standard builds.
  
  

• Christoph Katzer - Saturday, February 21, 2009 - link

  I'll have an Antec case review soon which came with an Earthwatts. Will test that one too of course.