Posted by
Hemos
on from the lets'-get-'em-together dept.
CainX writes "A comparison of five power supply units was posted recently. Anandtech has done a more complete review and examined 18 of the newest models in the plus 300 watt range." Power supplies tend to be one of those forgotten, but ever so important parts of machine construction.
The results of the memory tests are amasing. The MTBF is about an hour on some of the power supplys. I'm not sure If I understand the setup but that is appauling. I expect a MTBF of about 100+ years not an hour.
Re:Power Supplys
by
jpmkm
·
· Score: 4, Informative
I'm thinking that's a UPS. Servers usually have redundant power supplies, not one for multiple machines.
Not quite as amusing as dansdata.com, but...
by
Junior+J.+Junior+III
·
· Score: 5, Interesting
I really liked Anandtech's article. It actually had a lot of information and was fairly thorough -- measuring performance in a variety of ways, including not just output wattages but also noise levels, heat buildup, and cost.
A bit more explaining the basics of what each different voltage rail is for and why x-level of performance is important would have been helpful. Along with some more basic stats, such as how long the power cables actually are -- surely people still build full-tower PCs, don't they?
-- You see? You see? Your stupid minds! Stupid! Stupid!
Re:How much power is drawn from the wall?
by
wirelessbuzzers
·
· Score: 5, Informative
No, that's a maximum rating. It doesn't draw nearly 400W if you aren't going to use all that, otherwise it would have to dissipate 200W on its own, which it certainly cannot.
-- I hereby place the above post in the public domain.
28 pages? Give me a break.
by
Anonymous Coward
·
· Score: 5, Informative
Re:What's the deal on the Antec 330W and P4?
by
drwtsn32
·
· Score: 4, Informative
Agreed. IMO Antec makes some of the best power supplies. I use several Antec 330W ones at work in P4 systems.
PC Power and Cooling is very good, but their supplies are more expensive than Antec and are usually noisier.
Not wanting to put a downer on things...
by
WIAKywbfatw
·
· Score: 5, Insightful
...but it strikes me that these reviews of PSUs aren't as accurate as they should be. I'm not wanting to run the guys at AnandTech or elsewhere down (because, most of the time, they do a great job) but it strikes me that, when you look at PSUs (as opposed to CPUs, graphics cards or HDDs) then perhaps testing just one sample of each product is flawed.
After all, some of the measurements taken to distinguish good from bad were to the fifth significant figure. It strikes me that if you have to be that precise to differentiate between the winners and the also-rans then you've got to test more than one of each PSU - three would be a minimum, five or more would be better - and average out the test results to give you figures that are more representative of the quality of these products.
After all, not every Zalman ZM400A-APF is going to have a 12V min/max fluctuation of only 0.005V, and not every Enermax EG651P-VE FMA 550W is going to have a fluctuaction of 0.65V. Who knows, perhaps this was just a particularly good Zalman and a particularly bad Enermax? Testing more units means accurate results, which is a good thing.
I appreciate that testing three (or five, or however many) of each PSU means more work - you have to get x many more of each unit, test x many more times, process x much more data before averaging out your results - but, sometimes, I think it's warranted. Without wanting to get down on anyone, I'd like to suggest that, where called for, they try to source more units and test more thoroughly.
And, before people start flaming me for not knowing what I'm talking about, how much work is involved, etc, let me just say that I've run a review lab and I do know what I'm talking about, how much work is involved, etc. It's not a trivial amount but, sometimes, it is worth it.
(No doubt that's just a cue for half a dozen people to tell me where I'm wrong. I welcome objective criticisms but you can keep any childish flames.)
--
"Accept that some days you are the pigeon, and some days you are the statue." - David Brent, Wernham Hogg
Where's the EE beef?
by
AragornCG
·
· Score: 5, Interesting
Great, there are more power supplies, but where are the oscilloscopes? Where is the detailed methodology for testing the cleanliness of the resultant power? They used some "stuff we found in our local university basement" to satisfy "you EE types". Oh, great. LIKE WHAT?
I'll take a smaller review with decent scientific methodologies, thank you very much. There's more to stability of a power supply than "interference testing" which reports an integer number of errors that could be caused by anything... What about the thermal testing... where was it measured? by what, the onboard mainboard thermistor, which is notably unreliable? Which "industry standard Chenming case" was it?
I really, really, really wish that these hardware sites would hire people with a decent understanding of the discipline of science - let alone engineering! - to write reviews. I could make MUCH more informed decisions.
-Ben
No dB ratings?
by
Anonymous Coward
·
· Score: 4, Insightful
I'd expect anyone reviewing power supplies to have provided some analysis of the noise generated by them. Since that isn't the case here, I'll stick with the original source material (read "manufacturer's promotional literature") instead of Anandtech's promotional review.
Less clicking, at any rate.
As a requisite disclaimer for the "It's Just Fine For Me" folks: While it may be true that certain electronic components are being engineered to be quieter, it's still the case that manufacturers rarely advertise/publish noise levels. The fact that there is a booming cottage industry devoted to quieter replacement components is evidence enough that manufacturers haven't yet fully addressed the needs of computer owners who are starting to discover that their nagging headaches and frequent innability to concentrate is due in large part to that box under their desks.
Re:How much power is drawn from the wall?
by
Detritus
·
· Score: 4, Informative
Your friend's parents are mistaken. Just because the computer has a 400 Watt power supply doesn't mean that it is actually supplying that much power. It is probably considerably less. Think of it like the engine in a car. It may be capable of 150 horsepower but the average power output is much less.
-- Mea navis aericumbens anguillis abundat
For anyone who thinks PS don't matter
by
plcurechax
·
· Score: 4, Informative
I just built a new Athlon XP (2600+) based system, and the power supply was the most frustrated part to get working correctly. I bought a generic case with a 300W Pro-Power ATX power supply. It had useless regulations, the 12V rail measured at 12.8V with the system on and idle, and jumpted to 13.2V when running some math calculations that I use for stress testing (www.mersenne.org).
So I spent a Saturday trying to find a local computer dealer open that had any power supplies in stock. Most were sold out because of recent lightning storms (note: most didn't fail right away), anyhow about 2 hours of looking and driving I found a Codegen 300W ATX supply. It didn't setup off the SilentPC power alarms, but it failed the stress testing, with errors in about 1 hour of testing. The CPU temperature was fine (40-42 C) but I suspect the power wasn't clean which introduced memory or logic glitches.
Finally after a week, I got a ThermalTake 360W power supply and my system works fine.
So anyone who has an unstable system, it may not be all the fault of the OS, but poor or underpowered power supply.
Re:External supplies
by
ZorinLynx
·
· Score: 4, Informative
No, this would be a bad idea. The cables from the power supply to the system would be carrying low voltages, and to provide power to the system at those voltages would require high current. This would cause voltage drop (and thus loss) in the cables.
When wiring up electrical systems, it is important to remember that your long runs should always be done with high voltage cable. For instance, from the wall to the power supply is 120V AC. Once the voltage is reduced, the runs need to be kept as short as possible, since every foot of cable loses substantially more power at the increased current required.
50 watts at 120 volts is less than half an amp. 50 watts at 12V is about 4A. Big difference. }:)
Quality power can not be stressed enough...
by
Awptimus+Prime
·
· Score: 4, Interesting
I used to buy whatever 450 watt CompUsa model I could get for $19.00. That was a big mistake.
After about 6 months use, I had one make a popping sound, then erupt in smoke. All I had in the box was dual 700 cpus, 6 scsi drives, 2 ide drives, cdrw, GF3, and a bunch of neon lights. During that time, the system would freeze up a lot, usually while compiling.
Basically 450 watts doesn't mean as much as the quality of components and how cloesly the manufacturer was willing to run them to the point of breakage.
I swapped out my burnt 450 watt for a Antec cool blue running at around 400 watts and my system is quieter and never locks up anymore.
When reading wattages of cheap PSU's remind yourself of audio equipment. Like when a brand like Sparkomatic or Jensen promises 1000 watts out of an amp only for it to sound like shiat and blow speakers. Meanwhile, a Bose amp of 75 watts would be generating more power overall and not be tearing your speakers up. The same game is played with PSU promises. They, the generic cheapies, never planned on anyone using all 450 watts of their equipment. These big number, low performance parts are meant for a consumer with a very basic system seeking comfort big specs.
I guess it reinforces 'You get what you pay for'.
Design Engineer, Power Supplies, Computers
by
lcsjk
·
· Score: 5, Informative
As one who has spent some 16 years designing power supplies for computers and other devices, I found the article to be very weak in factual information. Power supplies have an actual efficiency of about 75%, and no better than 90%, no matter what brand, Efficiency varies with loading. Quick spin-up on hard drives takes about 5 times as much 12V power as is required for running, but little 5V change. The actual voltage needed by components is not nearly so critical as the amount of noise (ripple and "switching" noise. (These are switching power supplies.) The nearest thing to a noise measurement test was the memory test.
Output regulation is also dependent on the loading of the 5 volts or 3.3 volt outputs. Expect the 12 volts to follow the 5 volts as the processor and MB requires more or less computing power during operation. Some are independent, but most are not.
Things like temperature testing can be very meaningless. Most components are designed for operation at an ambient of more than 70 degrees Celcius. The switching transistors will produce the most heat, but a small capacitor setting downstream in the airflow may be the first long term failure. Probably input to output is the best indication.
Note also that the power supply fan sucks hot air from the PC and sends it over power supply components before exhausting to the outside air.
The one test I found most interesting was the memory test, and although I am not sure of the methods used, the results do indicate overall performance. (Yes, I have also designed memorys and been involved it testing of memory chips with walking ones and zeros and checkerboard patterns etc. Without knowing which row of memory transistors is adjacent to the one being used, pattern testing is basically meaningless except for one word at a time.
A few years ago I took a Sparkle (SPI) power supply and a Fortran power supply apart and the insides were identical - completely. Fortran was about $3 more in hundreds at the time. My guess is that they were made on the same production line. Are they the same or had one just been copied.
The power supply continuous output rating should be the one used when talking about a 300 watt power supply, not the power line rating.
Finally, audible noise. Bigger fans, lower speed, less noise. Fans mounted on the inside instead of near the back panel produce less noise. Fans blowing onto holes instead of through a circular guard produce more noise.
Bad power supply tests
by
Animats
·
· Score: 4, Informative
That article was very lightweight. They didn't do tests that any competent QA shop would do.
Tests they didn't do, and should have:
Look up the UL certification for the unit, and check that the unit actually passed UL testing. Many power supplies, especially from China, have forged UL certifications. (UL had about 80 shipments seized at U.S. Customs last year for this.) UL's test include surviving a dead short and not catching fire for any single component failure. Uncertified power supplies are almost always the ones that catch fire.
Checking the UL database is easy. Check any power supplies you have, and if the E-number isn't in the database or doesn't match the unit, report it to UL and have a serious talk with the seller.
Test the power supply with input voltages at the bottom and top of the rated range, and measure the regulated outputs. This is basic stuff.
Test the power supply in a hot box, with temperature maintained at the high end of the specified temperature range. This should be done with the unit loaded up to its rated load, and should continue for at least 72 hours. This, again, is a basic QA test.
Those are all basic. The "memory test" was interesting. That's an indication that these power supplies differ in their ability to sink RF noise coming in on the power rails. That, too, needs to be tested.
Nothing Anandtech did seemed to use a 'scope, let alone a spectrum analyzer.
Slashdot Mirror
DOH! NOW I know why my home build PC isn't working. I also made the "often made" mistake of forgetting the Power Supply.
I'm glad to know this is a common mistake and not just me being stupid.
An Eye for an Eye will make the whole world blind - Gandhi
The results of the memory tests are amasing. The MTBF is about an hour on some of the power supplys. I'm not sure If I understand the setup but that is appauling. I expect a MTBF of about 100+ years not an hour.
Mouse powered Chips, Open source Processors and Lego
I'm thinking that's a UPS. Servers usually have redundant power supplies, not one for multiple machines.
I really liked Anandtech's article. It actually had a lot of information and was fairly thorough -- measuring performance in a variety of ways, including not just output wattages but also noise levels, heat buildup, and cost.
A bit more explaining the basics of what each different voltage rail is for and why x-level of performance is important would have been helpful. Along with some more basic stats, such as how long the power cables actually are -- surely people still build full-tower PCs, don't they?
You see? You see? Your stupid minds! Stupid! Stupid!
No, that's a maximum rating. It doesn't draw nearly 400W if you aren't going to use all that, otherwise it would have to dissipate 200W on its own, which it certainly cannot.
I hereby place the above post in the public domain.
Single-page link.
Agreed. IMO Antec makes some of the best power supplies. I use several Antec 330W ones at work in P4 systems.
PC Power and Cooling is very good, but their supplies are more expensive than Antec and are usually noisier.
...but it strikes me that these reviews of PSUs aren't as accurate as they should be. I'm not wanting to run the guys at AnandTech or elsewhere down (because, most of the time, they do a great job) but it strikes me that, when you look at PSUs (as opposed to CPUs, graphics cards or HDDs) then perhaps testing just one sample of each product is flawed.
After all, some of the measurements taken to distinguish good from bad were to the fifth significant figure. It strikes me that if you have to be that precise to differentiate between the winners and the also-rans then you've got to test more than one of each PSU - three would be a minimum, five or more would be better - and average out the test results to give you figures that are more representative of the quality of these products.
After all, not every Zalman ZM400A-APF is going to have a 12V min/max fluctuation of only 0.005V, and not every Enermax EG651P-VE FMA 550W is going to have a fluctuaction of 0.65V. Who knows, perhaps this was just a particularly good Zalman and a particularly bad Enermax? Testing more units means accurate results, which is a good thing.
I appreciate that testing three (or five, or however many) of each PSU means more work - you have to get x many more of each unit, test x many more times, process x much more data before averaging out your results - but, sometimes, I think it's warranted. Without wanting to get down on anyone, I'd like to suggest that, where called for, they try to source more units and test more thoroughly.
And, before people start flaming me for not knowing what I'm talking about, how much work is involved, etc, let me just say that I've run a review lab and I do know what I'm talking about, how much work is involved, etc. It's not a trivial amount but, sometimes, it is worth it.
(No doubt that's just a cue for half a dozen people to tell me where I'm wrong. I welcome objective criticisms but you can keep any childish flames.)
"Accept that some days you are the pigeon, and some days you are the statue." - David Brent, Wernham Hogg
Great, there are more power supplies, but where are the oscilloscopes? Where is the detailed methodology for testing the cleanliness of the resultant power? They used some "stuff we found in our local university basement" to satisfy "you EE types". Oh, great. LIKE WHAT?
I'll take a smaller review with decent scientific methodologies, thank you very much. There's more to stability of a power supply than "interference testing" which reports an integer number of errors that could be caused by anything... What about the thermal testing... where was it measured? by what, the onboard mainboard thermistor, which is notably unreliable? Which "industry standard Chenming case" was it?
I really, really, really wish that these hardware sites would hire people with a decent understanding of the discipline of science - let alone engineering! - to write reviews. I could make MUCH more informed decisions.
-Ben
I'd expect anyone reviewing power supplies to have provided some analysis of the noise generated by them. Since that isn't the case here, I'll stick with the original source material (read "manufacturer's promotional literature") instead of Anandtech's promotional review.
Less clicking, at any rate.
As a requisite disclaimer for the "It's Just Fine For Me" folks: While it may be true that certain electronic components are being engineered to be quieter, it's still the case that manufacturers rarely advertise/publish noise levels. The fact that there is a booming cottage industry devoted to quieter replacement components is evidence enough that manufacturers haven't yet fully addressed the needs of computer owners who are starting to discover that their nagging headaches and frequent innability to concentrate is due in large part to that box under their desks.
Your friend's parents are mistaken. Just because the computer has a 400 Watt power supply doesn't mean that it is actually supplying that much power. It is probably considerably less. Think of it like the engine in a car. It may be capable of 150 horsepower but the average power output is much less.
Mea navis aericumbens anguillis abundat
I just built a new Athlon XP (2600+) based system, and the power supply was the most frustrated part to get working correctly. I bought a generic case with a 300W Pro-Power ATX power supply. It had useless regulations, the 12V rail measured at 12.8V with the system on and idle, and jumpted to 13.2V when running some math calculations that I use for stress testing (www.mersenne.org).
So I spent a Saturday trying to find a local computer dealer open that had any power supplies in stock. Most were sold out because of recent lightning storms (note: most didn't fail right away), anyhow about 2 hours of looking and driving I found a Codegen 300W ATX supply. It didn't setup off the SilentPC power alarms, but it failed the stress testing, with errors in about 1 hour of testing. The CPU temperature was fine (40-42 C) but I suspect the power wasn't clean which introduced memory or logic glitches.
Finally after a week, I got a ThermalTake 360W power supply and my system works fine.
So anyone who has an unstable system, it may not be all the fault of the OS, but poor or underpowered power supply.
No, this would be a bad idea. The cables from the power supply to the system would be carrying low voltages, and to provide power to the system at those voltages would require high current. This would cause voltage drop (and thus loss) in the cables.
When wiring up electrical systems, it is important to remember that your long runs should always be done with high voltage cable. For instance, from the wall to the power supply is 120V AC. Once the voltage is reduced, the runs need to be kept as short as possible, since every foot of cable loses substantially more power at the increased current required.
50 watts at 120 volts is less than half an amp. 50 watts at 12V is about 4A. Big difference. }:)
I used to buy whatever 450 watt CompUsa model I could get for $19.00. That was a big mistake.
After about 6 months use, I had one make a popping sound, then erupt in smoke. All I had in the box was dual 700 cpus, 6 scsi drives, 2 ide drives, cdrw, GF3, and a bunch of neon lights. During that time, the system would freeze up a lot, usually while compiling.
Basically 450 watts doesn't mean as much as the quality of components and how cloesly the manufacturer was willing to run them to the point of breakage.
I swapped out my burnt 450 watt for a Antec cool blue running at around 400 watts and my system is quieter and never locks up anymore.
When reading wattages of cheap PSU's remind yourself of audio equipment. Like when a brand like Sparkomatic or Jensen promises 1000 watts out of an amp only for it to sound like shiat and blow speakers. Meanwhile, a Bose amp of 75 watts would be generating more power overall and not be tearing your speakers up. The same game is played with PSU promises. They, the generic cheapies, never planned on anyone using all 450 watts of their equipment. These big number, low performance parts are meant for a consumer with a very basic system seeking comfort big specs.
I guess it reinforces 'You get what you pay for'.
As one who has spent some 16 years designing power supplies for computers and other devices, I found the article to be very weak in factual information.
Power supplies have an actual efficiency of about 75%, and no better than 90%, no matter what brand, Efficiency varies with loading.
Quick spin-up on hard drives takes about 5 times as much 12V power as is required for running, but little 5V change.
The actual voltage needed by components is not nearly so critical as the amount of noise (ripple and "switching" noise. (These are switching power supplies.) The nearest thing to a noise measurement test was the memory test.
Output regulation is also dependent on the loading of the 5 volts or 3.3 volt outputs. Expect the 12 volts to follow the 5 volts as the processor and MB requires more or less computing power during operation. Some are independent, but most are not.
Things like temperature testing can be very meaningless. Most components are designed for operation at an ambient of more than 70 degrees Celcius. The switching transistors will produce the most heat, but a small capacitor setting downstream in the airflow may be the first long term failure. Probably input to output is the best indication.
Note also that the power supply fan sucks hot air from the PC and sends it over power supply components before exhausting to the outside air.
The one test I found most interesting was the memory test, and although I am not sure of the methods used, the results do indicate overall performance. (Yes, I have also designed memorys and been involved it testing of memory chips with walking ones and zeros and checkerboard patterns etc. Without knowing which row of memory transistors is adjacent to the one being used, pattern testing is basically meaningless except for one word at a time.
A few years ago I took a Sparkle (SPI) power supply and a Fortran power supply apart and the insides were identical - completely. Fortran was about $3 more in hundreds at the time. My guess is that they were made on the same production line. Are they the same or had one just been copied.
The power supply continuous output rating should be the one used when talking about a 300 watt power supply, not the power line rating.
Finally, audible noise. Bigger fans, lower speed, less noise. Fans mounted on the inside instead of near the back panel produce less noise.
Fans blowing onto holes instead of through a circular guard produce more noise.
Tests they didn't do, and should have:
-
Look up the UL certification for the unit, and check that the unit actually passed UL testing. Many power supplies, especially from China, have forged UL certifications. (UL had about 80 shipments seized at U.S. Customs last year for this.) UL's test include surviving a dead short and not catching fire for any single component failure. Uncertified power supplies are almost always the ones that catch fire.
-
Test the power supply with input voltages at the bottom and top of the rated range, and measure the regulated outputs. This is basic stuff.
-
Test the power supply in a hot box, with temperature maintained at the high end of the specified temperature range. This should be done with the unit loaded up to its rated load, and should continue for at least 72 hours. This, again, is a basic QA test.
Those are all basic. The "memory test" was interesting. That's an indication that these power supplies differ in their ability to sink RF noise coming in on the power rails. That, too, needs to be tested.Checking the UL database is easy. Check any power supplies you have, and if the E-number isn't in the database or doesn't match the unit, report it to UL and have a serious talk with the seller.
Nothing Anandtech did seemed to use a 'scope, let alone a spectrum analyzer.