As your power supply testing methodology continues to evolve do you see yourself retesting previously reviewed units, if necessary? My concern is that all test results should be directly comparable between all reviewed units.
As an audio guy I feel it's necessary to point out that the *frequency* of the sound generated by a fan has a lot to do with how loud / annoying we might perceive it to be. The human ear, as we know, is not linear and is more sensitive at some frequencies than at others (for those unfamiliar with the concept, Google "Fletcher Munson equal loudness curve" or try the [url=http://en.wikipedia.org/wiki/Fletcher-Munson_curve...">http://en.wikipedia.org/wiki/Fletcher-Munson_curve...]Wikipedia entry[/url]).
For example, given two fans outputting 40dBSPL, one generating a tone centered at 1KHz will be far more annoying than one with a fundamental of 600Hz, simply because the ear is particularly sensitive around 1KHz, and less so at 600Hz.
Given the quality of your audio test equipment (great choice on the MG mic & pre, BTW) you can easily create an accurate frequency plot of the PS's audio output. Perhaps you could include these frequency plots in future reviews, along with sample recordings of the actual fan noise.
Just a thought.
I've been waiting a long time for a review site to pick up some chroma gear and put out consistent PS reviews. Kudos!
It can be problematic picking a high wattage PSU based on fan noise frequency. It is actually better to have a higher fan frequency for PSU lifespan, because that higher frequency is typically caused by use of ball bearings instead of sleeve bearing, which is much more reliable in a horizontally mounted PSU fan.
If the noise is important I suggest you use a system that doesn't consume much power and has the older rearward facing fan (being a quality sleeve bearing, brand like Papst or Panaflo).
this review tells of how this is one of the very few psu's to actually have true independent voltage regulation; not sure how that has any real world effect but it at least looks like one of the most jam packed psu's i've ever seen.
I noticed that the secondary heatsink temperature reached 90 degrees during testing. I've never been a fan of Teapo capacitors, and I suspect (though not stated in the article) that the caps on this PSU are NOT 105 C. caps. This means that given time, the degradation can be significant and the MTBF will rise accordingly. Can anyone justify 170 doolars a new PSU every couple of years? Especially when 35 dollars in better caps would change all that. It would basically future-proof your purchase, though the manufacturer would have to raise the price accordingly as they are not in the business of charity.
I forsee this happening as soon as the majority of home users expect more from a PSU than a shiny case or blinky lights and a wildly opptimistic output rating.
Teapo are not considered the "best" alternative in electrolytics but through several years of use PSU owners have found their performance generally acceptible. They are not one of the typically problematic caps seen to fail frequently in PSU.
It is just unfounded to think they're not 105C caps, odds are very high that they are on the secondary side as practically all very-low ESR caps these days are 105C rated. While the large primary caps aren't always 105C, often 85C, it is seldom we ever see them fail because there is a much lower ripple current ratio on the high (voltage) side, ESR vs physical size means there's very little self heating.
Also, a heatsink is meant to be a higher thermal density having the purpose of spreading out heat conducted from the (diodes, etc) parts on secondary side. Conduction of this heat through air to adjacent parts is low and the caps won't be heated as much as you seem to suspect merely by having the heatsink at 90 degrees.
However, it does not change your point about higher heat levels in the PSU ultimately leading to a shorter life. We would need more measurements of the capacitor tops (and mathematically solve for cap core temp) to arrive at some kind of end of life projection, but even then end of life is considered to be falling below cap manufacturers specs, which in a well engineered PSU with a lot of margin could still mean that even if the cap is below IT'S spec, it is still functionally acceptible for use in the PSU without it falling outside of acceptible ATX spec ranges.
Yes better caps would be nice, yes lowering temps would be good too. Unfortunately too many people act as though a 600W+ PSU is supposed to be quiet when even at today's 80% efficiency target that would mean 150W of heat that must be removed.
I have personally witnessed a UL-certified fibreglass circuit board catch fire due to an inner-layer power-plane to ground short at a feed-through hole, initially stimulated by physical damage to the board by a component overheating and over-currenting the feed-through causing it to burn-out and generate a low-impedance path between the adjacent inner-layer power-planes with a ring of carbonized circuit-board material. (In modern circuit boards, signal feedthroughs often pass through close-tolerance clearance-holes on inner-layer power-planes ) Due to the conduction of the carbon across the edges of the inner-layer planes, the board will actually burn - fire-retardant or not - generating a larger and larger carbon-edged hole, until one of the following happens:-
* The power source is removed and the board instantly self-extinguishes.
* the current density around the hole as it enlarges becomes reduced sufficiently to allow the self-extinguishing property of the UL-certified fibreglass to gain the upper hand.
Unfortunately the carbonized edge of the hole is not a short-circuit and thus the current-limit of the power-supply is likely to be the only limiting protection. So, if 54 amps just happens to be available instead of 20amps, there is a strong possibility that a burning circuit board can damage a bunch of other components and cause a secondary fire before the hole is big enough for the current density to fall low enough for the board to finally self-extinguish.
Intel's per-line current limit recommendation is built on very good safety/reliability science. Buy one of these power-supplies and disable the individual-line current limits at your own risk.
While your comment is interesting, it's non-applicable here because these are only one layer per side boards (if that, many PSU still use pretty poor paperboard and single-sided PCB).
Intel's recommendation is based on amps/rail leaving the PSU casing, it does not stipulate nor limit what the current capability would be between a power and ground plane (if such a board design existed, but none do AFAIK (on a high current PC PSU, I don't rule out such a design in a notebook or other CE devices) as mentioned above.
While I think the Efficiency charts in load percentage are nice, if you included Efficiency charts in watts it would be more informative from a buying perspective.
The reason? I know the sum of system load is 375watts normally. The way it is now, if power supply XYZ has a max load of 650watts, I have to calculate where my ~375watts falls into that load chart (~57% load). But if the next power supply has 1000 watts max, then I have to yet again calculate what load percentage that may be for *that* power supply. If the Efficiency charts were in watts, instead of load, no calculations would be necessary. If I could look at your charts and see that XYZ power supply provided those watts the most efficiently, that would be the power supply I would get.
Maybe you could just provide a second X axis on the chart that included the watts.
I picked up one of these OP650 units. Worked well in my system, but emitted a buzzing/chirping noise with any graphic change 2D and 3D.
I was in immediate need for a presentation so bought my friends Corsair 620HX...which met my needs.
The Corsair is queiter at idle and load in my system, and has modular cabling as better power effeciencicy if I remember right.
Kudos to Silverstone for good tech support and willing-ness to replace my unit....I ended up returning to the on-line place and they paid for all shipping charges and no fees. Silverstone said if I had any problems such as restocking fees or shipping fees they'd help out.
Good job Silverstone...stand behind your products! I will likely buy from them next time I need a PS
Thank you all for the feedback again. I think I can answer most of you with saying that we already work on the additonal tests. Some additional tests do need more time but will follow. PCP&C Silencer will follow next and it's well worth waiting for it.
Sounds like a nicely built power supply that has problems with voltage drops. I've had a few hard drives fail because of voltage drop in the 5V rails. The 4 rail design puts the graphics cards on rail4, so does that mean if you run a 8800GTS/GTX SLI setup, it will overload the 4th rail when power is available on rail 3? I would like to see a test of the PC P&C 750W Silencer Quad, which claims to be a single rail 60A design. I'm considering getting the silencer unit to replace my OCZ GameXtreme 600W which suffers ripple voltage problems.
quote: This might be a valid way to design a power supply, but, even though we don't know what it is, Intel probably had a reason for designing the specifications around a 20 amp per 12V rail limit.
Intel considers more then 20 amps in a rail to be a fire hazard. If you short a 12v rail to a grounded case, you can pump more amps into your ground line then your house's wiring can handle, but since your circuit breaker isn't on the ground line, you'll never trip the breaker. In that case the only thing between you and burning your house down is the limiter on the power supply, hence the 20 amp requirement.
I don't know how big of a risk this is in practice, but thats the official explanation.
quote: you can pump more amps into your ground line then your house's wiring can handle
Not if the power supply is constructed correctly. ATX power supplies are required to be SELV systems, which means that crowbaring and output to an external ground will simply result in a shift of the internal ground level and no current will flow. This is actually done for electrocution safety reasons - even if ATX PSUs had 500V outputs, you'd still be pretty much safe from electrocution.
Additionally, the restriction is in power, not current. It's 240 VA limited, so 12 V at 20 A, or 5 V at 48 A, or 3.3 V at 73 A.
The real reason for the restriction is to limit the danger if an output gets tied to an internal ground. The resulting power being dissipated in a screwdriver (for example) can result in bits of the screwdriver getting turned into plasma, blasting bits of hot metal everywhere. Even if you don't get such an explosive response, you still have a fire danger (inside the case, not in your house).
In a prior article you showed us the box you use for testing power supplies. I don't remember reading how the heat generated by the power supplies is removed from the sealed box. Does the temp of the air in the box rise with time? Would the drop off in voltage be less with full ventilation?
Good job. Clearly a lot of effort was put behind putting the methodology and this review together.
Couple questions..
1. How much of the voltage drop shown would you contribute to the resistance created by the interface board and wires located between the end of the PSU's connector and the actual load. This isn't my favorite power supply, but I know the voltage regulation isn't nearly as bad as you have it graphed. And naturallly, since load also creates resistance, your voltage is going to decrease exponentially if you're not loading and measuring at the end of the connector. Since you're using a Chroma, I know you're not loading and measuring at the connector. Perhaps you could rig a DMM to measure voltages at the end of the connector instead of reporting what the Chroma is telling you since the Chroma is going to be incorrect since it doesn't take into consideration this added resistance.
2. Do you think it's possible to define "10%, 20%, etc." as it pertains to your loads. Naturally, you can't load all rails by 80%, 90%, 100%" because you'd easily exceed the power supplies capabilities. And certainly you're not loading the +12V with 80%, 90% 100% and then filling up the remainder with a 3.3V and 5V load because then you would crossload the PSU with an unusually high +12V load.
3. And, at risk of sounding like I'm beating a dead horse, ripple and noise results. We all know spec is 1%, and ripple and noise does typically increase with load. It would be nice to see if the power supplies being tested stay within that spec. through out it's total advertised capability.
Another issue pertinent to the extreme sag in voltage lines and perhaps to the soaring temperatures is to ask how long the PSU was run at each load level (and in total) and what was the air temperature at the PSU intake? Maybe the test box is just letting the PSU get way too unnaturally hot, and this has a deleterious effect on performance. This test box is not the same as thermal control chamber where you can dial in a specific target temperature, which will then be held constant by the chamber throughout the testing.
Since the test box is well insulated and sealed, there is a direct correlation between time spent with the PSU on and internal air temperature. No matter how the PSU is cooled, at any load, the temperature of the test box should rise continuously (even if only incrementally at very low load) as long as the PSU is running. This is a significant uncontrolled and unreported factor. That PSU intake temperature will be dictated by a complex mix of factors, by too many variables --
-starting ambient air temp in the box
-how long the PSU is kept running in the box and at which loads
-heat loss out of the box
-efficiency curve of the PSU
-fan and fan controller characteristics
If the time spent at each load and the starting ambient temp is kept the same for each and all PSUs, then the variables are correctly contained to the PSUs being tested. Then differences in the air temperature at the intake at any given power level for each PSU would reflect real differences in the cooling system and efficiency of each PSU. It would ensure a more level thermal playing field.
It was a nice article. I know not to buy this PSU now when I build my roommates computer later this month. I don't like how it falls out of spec at high loads. I would like to see a review on the PC P&C 750W Silencer Quad, as that was what I was planning on using for his computer.
Looking at that board frightened me, seeing as how much power was in that supply, and how close together the components were. I hope that they can increase the size of the standard power supply to help alleviate this problem now that we are building computers that have such high loads. I used to design and test power supplies (albeit for automotive conponents) and seeing how they crammed those parts so close together was scary. That is an easy way to kill the reliability and life of your supply. The heat just kills the board and components. Although it does reduce problems like parasitic capacitance. Maybe that is why many manufactures are avoiding using the top mounted 120mm fan; to keep from having to package the component like that.
quote: If you leave the PC off for a long period of time, you should still always hit the switch or unplug the cable.
But doesn't that mean the motherboard will need to use its little battery backup to keep the BIOS settings? Turning off the PSU switch and/or unplugging the cable to fully remove power sounds like a way to kill your motherboard's battery quickly.
It is only a "suggestion", there is just as valid an argument to not unplug it unless you're on a quest to save every last bit of power possible which is a nobile goal but put in perspective, a bit of a band-aid since anyone using a modern computer to access webpages is wasting orders of magnitude more power, even ignoring the typical products with large power consumption.
It might be said that unplugging also provides some protection against surges, limiting exposure to them, but it's really something that would have to be considered on a per-site basis, remembering that most people don't unplug their computers any day of the year and seldom is surge damage a recurring problem. IOW, a matter of how much extra effort to put forth to guard against something that, statistically, isn't likely to happen.
iirc, your mobo battery is in use when the computer is off and the PSU is on anyway. i could be wrong ... its been like a decade since i paid attention to that.
But either way, mobo batteries last years even when their not powered up.
Very nice work on AT's first PSU review with such detail. One question though, what happened to the Ripple and Noise results from the PSU? In the methodology they were mentioned to be tested, yet not in this review?
and yet, we don't really need to know the ripple values so long as they stay within ATX specs at the max rated loads and all crossloading combinations possible. Within these limits, lower ripple is not necessarily "better" per se, if it were important to have lowest possible ripple we wouldn't be using switching PSU at all or they'd at least have an addt'l stage of LC filtration before the output.
A PSU having multiple rails just means that a single rail in the PSU runs through a number of parallel current limiters - so all lines on the 12V1 rail go through one 20A current limiter, all lines on 12V2 go through another, etc. This is done as IEC safety requirements (and consequently ATX PSU requirements) say that "operator accessible" connections must not be able to deliver more than 240 VA (ie: 20 A at 12 V).
In older PSUs, only a single current limiter was used as there was no requirement for maximum current per line. In many recent PSUs, the single current maximum is starting to come back as well due to the high current requirements of modern GPUs and motherboards.
While it is common (because it's cheaper to implement) for a 12V multi-rail PSU to use parallel current limiters, it is not necessarily true that all are designed this way, typically only those built towards lower component cost instead of higher sustainable current. Other options include having separate capacitance after the current limiter (resistor), or a second inductor-cap LC stage, or additionally a separate rectifier stage, or additionally a separate transformer (essentially going backwards towards building in a 2nd supply until available space and budget limit it).
That makes sense, and makes me think my guess about the 20A limit is possibly a contributing factor - It would be a safety issue if someone hung 40A worth of fans, lights, motorized case windows, whatever you want.. off of one pair of wires (IE, one molex connector feeding into the mass of extenders and passthrough connectors that most fans and lights I've seen use.).
You'd likely overheat the wires carrying all that power, if not the connectors as well, which could cause fire or electrocution hazards.
While a GPU may draw significantly more than 20A, they are also using 3 pairs now, so the actual power draw will be closer to 20A per pair.
The PCIe V2.0 PSUs I've seen suggest only using connectors from the same 12V rail for PCIe graphics cards -- because if you don't, you'll be connecting the common from two different 12V rails together.
This can cause issues.
If a graphics card has one 4pin and 1 8pin connector, like the HD 2900 XT, the GPU can potentially draw up to 225W from a single 12V rail through 2x PCIe graphics power connectors (3 pairs). That's about 19 amps through one rail for one PSU, but not over each pair.
They explained it.
The Intel ATX standard calls for no more than 20 Amps per 12V rail. So in order to avoid maxing out a single 12V rail at 20 Amps, PSUs have multiple rails support up to 20 Amps each.
If you use a single rail that can max out at 54 Amps as stated here, then you do not need the additional rails, but you are going against the ATX standard.
That confused me as well. I think they mentioned that the PSU supposedly includes an ability to turn the other rails off, but it doesn't work, and it always has 4. They did state the PCB was originally designed for 4 rails.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
46 Comments
Back to Article
meeshu - Thursday, July 19, 2007 - link
I would have liked to have seen results of ripple tests, regardless of whether they are within spec or not.Piyono - Sunday, July 15, 2007 - link
As your power supply testing methodology continues to evolve do you see yourself retesting previously reviewed units, if necessary? My concern is that all test results should be directly comparable between all reviewed units.Thanks,
Piyono
Christoph Katzer - Monday, July 16, 2007 - link
If it's possible (a matter of time) we will do it.Piyono - Sunday, July 15, 2007 - link
Very good first PS review.As an audio guy I feel it's necessary to point out that the *frequency* of the sound generated by a fan has a lot to do with how loud / annoying we might perceive it to be. The human ear, as we know, is not linear and is more sensitive at some frequencies than at others (for those unfamiliar with the concept, Google "Fletcher Munson equal loudness curve" or try the [url=http://en.wikipedia.org/wiki/Fletcher-Munson_curve...">http://en.wikipedia.org/wiki/Fletcher-Munson_curve...]Wikipedia entry[/url]).
For example, given two fans outputting 40dBSPL, one generating a tone centered at 1KHz will be far more annoying than one with a fundamental of 600Hz, simply because the ear is particularly sensitive around 1KHz, and less so at 600Hz.
Given the quality of your audio test equipment (great choice on the MG mic & pre, BTW) you can easily create an accurate frequency plot of the PS's audio output. Perhaps you could include these frequency plots in future reviews, along with sample recordings of the actual fan noise.
Just a thought.
I've been waiting a long time for a review site to pick up some chroma gear and put out consistent PS reviews. Kudos!
Piyono
mindless1 - Wednesday, July 18, 2007 - link
It can be problematic picking a high wattage PSU based on fan noise frequency. It is actually better to have a higher fan frequency for PSU lifespan, because that higher frequency is typically caused by use of ball bearings instead of sleeve bearing, which is much more reliable in a horizontally mounted PSU fan.If the noise is important I suggest you use a system that doesn't consume much power and has the older rearward facing fan (being a quality sleeve bearing, brand like Papst or Panaflo).
Christoph Katzer - Sunday, July 15, 2007 - link
Thanks, actually we are still working on the audio-equipment. When it's ready we will have quite some data to show.Piyono - Sunday, July 15, 2007 - link
Hey, that's good to know.I'm eager to see how this develops!
Piyono
xsilver - Saturday, July 14, 2007 - link
do silverstone manufacture this psu themselves or do they subcontract it to an OEM company like etasis?im most intrigued because they make the ZF series of psu's which have a dual PCB design
http://www.xbitlabs.com/articles/other/display/atx...">http://www.xbitlabs.com/articles/other/display/atx...
this review tells of how this is one of the very few psu's to actually have true independent voltage regulation; not sure how that has any real world effect but it at least looks like one of the most jam packed psu's i've ever seen.
Operandi - Saturday, July 14, 2007 - link
A good PSU review the proper way isn't an easy thing but it looks like you guys have an excellent handle on it, great work.maluckey - Saturday, July 14, 2007 - link
I noticed that the secondary heatsink temperature reached 90 degrees during testing. I've never been a fan of Teapo capacitors, and I suspect (though not stated in the article) that the caps on this PSU are NOT 105 C. caps. This means that given time, the degradation can be significant and the MTBF will rise accordingly. Can anyone justify 170 doolars a new PSU every couple of years? Especially when 35 dollars in better caps would change all that. It would basically future-proof your purchase, though the manufacturer would have to raise the price accordingly as they are not in the business of charity.I forsee this happening as soon as the majority of home users expect more from a PSU than a shiny case or blinky lights and a wildly opptimistic output rating.
mindless1 - Wednesday, July 18, 2007 - link
Teapo are not considered the "best" alternative in electrolytics but through several years of use PSU owners have found their performance generally acceptible. They are not one of the typically problematic caps seen to fail frequently in PSU.It is just unfounded to think they're not 105C caps, odds are very high that they are on the secondary side as practically all very-low ESR caps these days are 105C rated. While the large primary caps aren't always 105C, often 85C, it is seldom we ever see them fail because there is a much lower ripple current ratio on the high (voltage) side, ESR vs physical size means there's very little self heating.
Also, a heatsink is meant to be a higher thermal density having the purpose of spreading out heat conducted from the (diodes, etc) parts on secondary side. Conduction of this heat through air to adjacent parts is low and the caps won't be heated as much as you seem to suspect merely by having the heatsink at 90 degrees.
However, it does not change your point about higher heat levels in the PSU ultimately leading to a shorter life. We would need more measurements of the capacitor tops (and mathematically solve for cap core temp) to arrive at some kind of end of life projection, but even then end of life is considered to be falling below cap manufacturers specs, which in a well engineered PSU with a lot of margin could still mean that even if the cap is below IT'S spec, it is still functionally acceptible for use in the PSU without it falling outside of acceptible ATX spec ranges.
Yes better caps would be nice, yes lowering temps would be good too. Unfortunately too many people act as though a 600W+ PSU is supposed to be quiet when even at today's 80% efficiency target that would mean 150W of heat that must be removed.
kilkennycat - Friday, July 13, 2007 - link
I have personally witnessed a UL-certified fibreglass circuit board catch fire due to an inner-layer power-plane to ground short at a feed-through hole, initially stimulated by physical damage to the board by a component overheating and over-currenting the feed-through causing it to burn-out and generate a low-impedance path between the adjacent inner-layer power-planes with a ring of carbonized circuit-board material. (In modern circuit boards, signal feedthroughs often pass through close-tolerance clearance-holes on inner-layer power-planes ) Due to the conduction of the carbon across the edges of the inner-layer planes, the board will actually burn - fire-retardant or not - generating a larger and larger carbon-edged hole, until one of the following happens:-* The power source is removed and the board instantly self-extinguishes.
* the current density around the hole as it enlarges becomes reduced sufficiently to allow the self-extinguishing property of the UL-certified fibreglass to gain the upper hand.
Unfortunately the carbonized edge of the hole is not a short-circuit and thus the current-limit of the power-supply is likely to be the only limiting protection. So, if 54 amps just happens to be available instead of 20amps, there is a strong possibility that a burning circuit board can damage a bunch of other components and cause a secondary fire before the hole is big enough for the current density to fall low enough for the board to finally self-extinguish.
Intel's per-line current limit recommendation is built on very good safety/reliability science. Buy one of these power-supplies and disable the individual-line current limits at your own risk.
mindless1 - Wednesday, July 18, 2007 - link
While your comment is interesting, it's non-applicable here because these are only one layer per side boards (if that, many PSU still use pretty poor paperboard and single-sided PCB).Intel's recommendation is based on amps/rail leaving the PSU casing, it does not stipulate nor limit what the current capability would be between a power and ground plane (if such a board design existed, but none do AFAIK (on a high current PC PSU, I don't rule out such a design in a notebook or other CE devices) as mentioned above.
Vidmar - Friday, July 13, 2007 - link
While I think the Efficiency charts in load percentage are nice, if you included Efficiency charts in watts it would be more informative from a buying perspective.The reason? I know the sum of system load is 375watts normally. The way it is now, if power supply XYZ has a max load of 650watts, I have to calculate where my ~375watts falls into that load chart (~57% load). But if the next power supply has 1000 watts max, then I have to yet again calculate what load percentage that may be for *that* power supply. If the Efficiency charts were in watts, instead of load, no calculations would be necessary. If I could look at your charts and see that XYZ power supply provided those watts the most efficiently, that would be the power supply I would get.
Maybe you could just provide a second X axis on the chart that included the watts.
Thanks!
Souka - Friday, July 13, 2007 - link
I picked up one of these OP650 units. Worked well in my system, but emitted a buzzing/chirping noise with any graphic change 2D and 3D.I was in immediate need for a presentation so bought my friends Corsair 620HX...which met my needs.
The Corsair is queiter at idle and load in my system, and has modular cabling as better power effeciencicy if I remember right.
Kudos to Silverstone for good tech support and willing-ness to replace my unit....I ended up returning to the on-line place and they paid for all shipping charges and no fees. Silverstone said if I had any problems such as restocking fees or shipping fees they'd help out.
Good job Silverstone...stand behind your products! I will likely buy from them next time I need a PS
Christoph Katzer - Friday, July 13, 2007 - link
Thank you all for the feedback again. I think I can answer most of you with saying that we already work on the additonal tests. Some additional tests do need more time but will follow. PCP&C Silencer will follow next and it's well worth waiting for it.Belldandy - Friday, July 13, 2007 - link
Sounds like a nicely built power supply that has problems with voltage drops. I've had a few hard drives fail because of voltage drop in the 5V rails. The 4 rail design puts the graphics cards on rail4, so does that mean if you run a 8800GTS/GTX SLI setup, it will overload the 4th rail when power is available on rail 3? I would like to see a test of the PC P&C 750W Silencer Quad, which claims to be a single rail 60A design. I'm considering getting the silencer unit to replace my OCZ GameXtreme 600W which suffers ripple voltage problems.saratoga - Friday, July 13, 2007 - link
Intel considers more then 20 amps in a rail to be a fire hazard. If you short a 12v rail to a grounded case, you can pump more amps into your ground line then your house's wiring can handle, but since your circuit breaker isn't on the ground line, you'll never trip the breaker. In that case the only thing between you and burning your house down is the limiter on the power supply, hence the 20 amp requirement.
I don't know how big of a risk this is in practice, but thats the official explanation.
qpwoei - Saturday, July 14, 2007 - link
Not if the power supply is constructed correctly. ATX power supplies are required to be SELV systems, which means that crowbaring and output to an external ground will simply result in a shift of the internal ground level and no current will flow. This is actually done for electrocution safety reasons - even if ATX PSUs had 500V outputs, you'd still be pretty much safe from electrocution.
Additionally, the restriction is in power, not current. It's 240 VA limited, so 12 V at 20 A, or 5 V at 48 A, or 3.3 V at 73 A.
The real reason for the restriction is to limit the danger if an output gets tied to an internal ground. The resulting power being dissipated in a screwdriver (for example) can result in bits of the screwdriver getting turned into plasma, blasting bits of hot metal everywhere. Even if you don't get such an explosive response, you still have a fire danger (inside the case, not in your house).
LTG - Friday, July 13, 2007 - link
This is exactly what what is missing from the first page of the article.I think the author said Intel wanted it for "security" reasons when he meant "safety" reasons which made it hard to infer the fire issue.
Or maybe it is security and 4 rails makes it harder to hack into your system.
YellowWing - Friday, July 13, 2007 - link
In a prior article you showed us the box you use for testing power supplies. I don't remember reading how the heat generated by the power supplies is removed from the sealed box. Does the temp of the air in the box rise with time? Would the drop off in voltage be less with full ventilation?YellowWing - Friday, July 13, 2007 - link
Brownouts have caused power supply failures for me in the past. How about dialing the input voltage down to 90 or so and see what happens.From experience I can say that some supplies don't survive at lower input voltages.
DerekWilson - Friday, July 13, 2007 - link
we are considering looking at 90V input. Thanks for the suggestion.bpt8056 - Friday, July 13, 2007 - link
I just want to say that I thought AT did a great job on the review. Keep up the great work!puffpio - Friday, July 13, 2007 - link
Great article..looking forward to many many moreOn one of the graphs of the 230V tests, it is labeled as 115V
jonnyGURU - Friday, July 13, 2007 - link
Good job. Clearly a lot of effort was put behind putting the methodology and this review together.Couple questions..
1. How much of the voltage drop shown would you contribute to the resistance created by the interface board and wires located between the end of the PSU's connector and the actual load. This isn't my favorite power supply, but I know the voltage regulation isn't nearly as bad as you have it graphed. And naturallly, since load also creates resistance, your voltage is going to decrease exponentially if you're not loading and measuring at the end of the connector. Since you're using a Chroma, I know you're not loading and measuring at the connector. Perhaps you could rig a DMM to measure voltages at the end of the connector instead of reporting what the Chroma is telling you since the Chroma is going to be incorrect since it doesn't take into consideration this added resistance.
2. Do you think it's possible to define "10%, 20%, etc." as it pertains to your loads. Naturally, you can't load all rails by 80%, 90%, 100%" because you'd easily exceed the power supplies capabilities. And certainly you're not loading the +12V with 80%, 90% 100% and then filling up the remainder with a 3.3V and 5V load because then you would crossload the PSU with an unusually high +12V load.
3. And, at risk of sounding like I'm beating a dead horse, ripple and noise results. We all know spec is 1%, and ripple and noise does typically increase with load. It would be nice to see if the power supplies being tested stay within that spec. through out it's total advertised capability.
Thanks and I look forward to the next PSU review!
mcvan - Saturday, July 14, 2007 - link
Like Johnny said. ;)Another issue pertinent to the extreme sag in voltage lines and perhaps to the soaring temperatures is to ask how long the PSU was run at each load level (and in total) and what was the air temperature at the PSU intake? Maybe the test box is just letting the PSU get way too unnaturally hot, and this has a deleterious effect on performance. This test box is not the same as thermal control chamber where you can dial in a specific target temperature, which will then be held constant by the chamber throughout the testing.
Since the test box is well insulated and sealed, there is a direct correlation between time spent with the PSU on and internal air temperature. No matter how the PSU is cooled, at any load, the temperature of the test box should rise continuously (even if only incrementally at very low load) as long as the PSU is running. This is a significant uncontrolled and unreported factor. That PSU intake temperature will be dictated by a complex mix of factors, by too many variables --
-starting ambient air temp in the box
-how long the PSU is kept running in the box and at which loads
-heat loss out of the box
-efficiency curve of the PSU
-fan and fan controller characteristics
If the time spent at each load and the starting ambient temp is kept the same for each and all PSUs, then the variables are correctly contained to the PSUs being tested. Then differences in the air temperature at the intake at any given power level for each PSU would reflect real differences in the cooling system and efficiency of each PSU. It would ensure a more level thermal playing field.
qpwoei - Friday, July 13, 2007 - link
While we're beating dead horses, I'll land a few blows on the dynamic load response horse as well :)
SilthDraeth - Friday, July 13, 2007 - link
Seems like you know what you are talking about. I hope they take some of your points into consideration, since they all seem very valid.jonnyGURU - Friday, July 13, 2007 - link
Thanks. But I'm just an old hack. ;)Martimus - Friday, July 13, 2007 - link
It was a nice article. I know not to buy this PSU now when I build my roommates computer later this month. I don't like how it falls out of spec at high loads. I would like to see a review on the PC P&C 750W Silencer Quad, as that was what I was planning on using for his computer.Looking at that board frightened me, seeing as how much power was in that supply, and how close together the components were. I hope that they can increase the size of the standard power supply to help alleviate this problem now that we are building computers that have such high loads. I used to design and test power supplies (albeit for automotive conponents) and seeing how they crammed those parts so close together was scary. That is an easy way to kill the reliability and life of your supply. The heat just kills the board and components. Although it does reduce problems like parasitic capacitance. Maybe that is why many manufactures are avoiding using the top mounted 120mm fan; to keep from having to package the component like that.
yacoub - Friday, July 13, 2007 - link
But doesn't that mean the motherboard will need to use its little battery backup to keep the BIOS settings? Turning off the PSU switch and/or unplugging the cable to fully remove power sounds like a way to kill your motherboard's battery quickly.
mindless1 - Wednesday, July 18, 2007 - link
It is only a "suggestion", there is just as valid an argument to not unplug it unless you're on a quest to save every last bit of power possible which is a nobile goal but put in perspective, a bit of a band-aid since anyone using a modern computer to access webpages is wasting orders of magnitude more power, even ignoring the typical products with large power consumption.It might be said that unplugging also provides some protection against surges, limiting exposure to them, but it's really something that would have to be considered on a per-site basis, remembering that most people don't unplug their computers any day of the year and seldom is surge damage a recurring problem. IOW, a matter of how much extra effort to put forth to guard against something that, statistically, isn't likely to happen.
DerekWilson - Friday, July 13, 2007 - link
iirc, your mobo battery is in use when the computer is off and the PSU is on anyway. i could be wrong ... its been like a decade since i paid attention to that.But either way, mobo batteries last years even when their not powered up.
SpaceRanger - Friday, July 13, 2007 - link
Very nice work on AT's first PSU review with such detail. One question though, what happened to the Ripple and Noise results from the PSU? In the methodology they were mentioned to be tested, yet not in this review?Looking forward to more PSU reviews..
Shadowmage - Friday, July 13, 2007 - link
I agree. The reviews must have ripple. That's why Jonnyguru's reviews are so highly regarded.mindless1 - Wednesday, July 18, 2007 - link
and yet, we don't really need to know the ripple values so long as they stay within ATX specs at the max rated loads and all crossloading combinations possible. Within these limits, lower ripple is not necessarily "better" per se, if it were important to have lowest possible ripple we wouldn't be using switching PSU at all or they'd at least have an addt'l stage of LC filtration before the output.LTG - Friday, July 13, 2007 - link
The first page leaves out an explanation of why multiple rails are used in the first place.I'm sure many technical software people, who don't know hardware, wonder like I do, why wasn't it always just one rail?
Just a couple sentences would probably be helpful.
thanks.
qpwoei - Friday, July 13, 2007 - link
A PSU having multiple rails just means that a single rail in the PSU runs through a number of parallel current limiters - so all lines on the 12V1 rail go through one 20A current limiter, all lines on 12V2 go through another, etc. This is done as IEC safety requirements (and consequently ATX PSU requirements) say that "operator accessible" connections must not be able to deliver more than 240 VA (ie: 20 A at 12 V).In older PSUs, only a single current limiter was used as there was no requirement for maximum current per line. In many recent PSUs, the single current maximum is starting to come back as well due to the high current requirements of modern GPUs and motherboards.
mindless1 - Wednesday, July 18, 2007 - link
While it is common (because it's cheaper to implement) for a 12V multi-rail PSU to use parallel current limiters, it is not necessarily true that all are designed this way, typically only those built towards lower component cost instead of higher sustainable current. Other options include having separate capacitance after the current limiter (resistor), or a second inductor-cap LC stage, or additionally a separate rectifier stage, or additionally a separate transformer (essentially going backwards towards building in a 2nd supply until available space and budget limit it).Araemo - Friday, July 13, 2007 - link
That makes sense, and makes me think my guess about the 20A limit is possibly a contributing factor - It would be a safety issue if someone hung 40A worth of fans, lights, motorized case windows, whatever you want.. off of one pair of wires (IE, one molex connector feeding into the mass of extenders and passthrough connectors that most fans and lights I've seen use.).You'd likely overheat the wires carrying all that power, if not the connectors as well, which could cause fire or electrocution hazards.
While a GPU may draw significantly more than 20A, they are also using 3 pairs now, so the actual power draw will be closer to 20A per pair.
DerekWilson - Friday, July 13, 2007 - link
The PCIe V2.0 PSUs I've seen suggest only using connectors from the same 12V rail for PCIe graphics cards -- because if you don't, you'll be connecting the common from two different 12V rails together.This can cause issues.
If a graphics card has one 4pin and 1 8pin connector, like the HD 2900 XT, the GPU can potentially draw up to 225W from a single 12V rail through 2x PCIe graphics power connectors (3 pairs). That's about 19 amps through one rail for one PSU, but not over each pair.
SilthDraeth - Friday, July 13, 2007 - link
They explained it.The Intel ATX standard calls for no more than 20 Amps per 12V rail. So in order to avoid maxing out a single 12V rail at 20 Amps, PSUs have multiple rails support up to 20 Amps each.
If you use a single rail that can max out at 54 Amps as stated here, then you do not need the additional rails, but you are going against the ATX standard.
Duraz0rz - Friday, July 13, 2007 - link
Also, I didn't see if there was a reason that it was advertised as a single rail, yet you have 4 12V rails.Nice article...really love the line curves for the load outputs. One thing I noticed missing is ripple testing. Any reason why it's not here?
SilthDraeth - Friday, July 13, 2007 - link
That confused me as well. I think they mentioned that the PSU supposedly includes an ability to turn the other rails off, but it doesn't work, and it always has 4. They did state the PCB was originally designed for 4 rails.Duraz0rz - Friday, July 13, 2007 - link
Nevermind...disregard my statement about the ripple testing. I probably just missed it in the original article after skimming the comments from it :)