What can a 350W PSU power?

[Pandamonium]

So I'm on a quest to build a "$30" computer.

I just realized that I have a spare 350W PSU lying around. I didn't actually buy this one. One of the labs across the hall from mine has been pitching a lot of gear recently. Among the things they left up for grabs was this PSU. It's brand new- the original rubber band is still around all the wires, so I'm assuming it functions. (They've also thrown out two broken but under warranty Samsung 214Ts (21.4' LCDs), a ton of working but dusty UPS units, a laptop or two, etc. Anyway, I have a free PSU, and I'd like to use it for my replacement computer.

It's an Antec SL350, with 5V/35A 12V/21A, 3.3V/28A maximums. My planned final build will have some dual core CPU, one video card (I'm eyeing the 8800GT, but might end up getting an 8600GT/GTS), one optical drive, one hard drive. If my motherboard doesn't have eSATA ports, I'll also have that in there. No overclocking. So the question is, will I be pushing it with a 350W PSU, or is this unit quality enough for my purposes?

 

[MarcVenice]

21a on the 12v rail isn't going to make for a stable system with a dualcore and a 8800gt, perhaps not even a 8600gt(s) ... Shop around, sometimes theres a 30$ deal on the antec earthwatts 500w psu at frys for example. Or look for a coolermaster extreme power 600-650w for 30$ or less ...

 

[MDE]

21A on the 12V rail should be fine for that system, especially with one HDD and one optical and a modest video card, as long as you're not overclocking.

 

[MarcVenice]

So nvidia recommending a minimum of 28a on the 12v rail for a 8800gt is just a fairy tale ? It's not the wattage we have to be concerned about, it's the amps ... The maximum amount the 12v rail could put out is 252w, 125 for the GPU/CPU each... Leaves no room for anything else being powered of the 12v rail. I'd say it's nuts ...

 

[Lunyone]

21A on the 12V rail isn't too bad, but a 8800gt? Not too sure about that combo. You might be okay witha 3850, but 8800gt is probably too much for 21A on the 12v rail. Don't know for sure, but if nVidia is saying a 28a on 12v rail than I don't think so. If you had 25a on the 12v rail, than I'd say you'd have a much better chance.

 

[dclive]

Originally posted by: Pandamonium
So I'm on a quest to build a "$30" computer.

I just realized that I have a spare 350W PSU lying around. I didn't actually buy this one. One of the labs across the hall from mine has been pitching a lot of gear recently. Among the things they left up for grabs was this PSU. It's brand new- the original rubber band is still around all the wires, so I'm assuming it functions. (They've also thrown out two broken but under warranty Samsung 214Ts (21.4' LCDs), a ton of working but dusty UPS units, a laptop or two, etc. Anyway, I have a free PSU, and I'd like to use it for my replacement computer.

It's an Antec SL350, with 5V/35A 12V/21A, 3.3V/28A maximums. My planned final build will have some dual core CPU, one video card (I'm eyeing the 8800GT, but might end up getting an 8600GT/GTS), one optical drive, one hard drive. If my motherboard doesn't have eSATA ports, I'll also have that in there. No overclocking. So the question is, will I be pushing it with a 350W PSU, or is this unit quality enough for my purposes?

With my Acer E700's 300W PSU, I powered the following:

Quad Core Q6600
P965 chipset Acer motherboard
4 500GB HDDs
2 TV Tuners, one a dual-tuner model
4 x 1 GB sticks of RAM
8800GTS/320 graphics card

Worked great.... my Kill-A-Watt reader said I needed high-200s power while at max load, so the 300W PSU was able to fully supply power. (that was an at-the-wall reading; the PSU actually was supplying about 75% of that to the computer, so there was plenty of headroom there - about 70W or so.)

You might read Anandtech's reviews to see what kind of power requirements there are for graphics cards and newer CPUs - in all the reviews they spell out power supply requirements. Most of the systems hover around 200W - 225W, just like mine. (75% x 275 = 206W actually used by the PC).

 

[MarcVenice]

And yet again I disagree. It's nice to see your 300w PSU managed to power all of that. How long did you have it up and running though, and what did you do with it ? Office work all the time ?

If you look at these charts for example: http://www.extremetech.com/art.../0,1697,2217143,00.asp

Perhaps running 3dmark06 and 2 instances of orthos give unrealistic figures, it be nice to see the power consumption when they run Crysis for example, but still, I doubt it will be that far off.

For all I care people use crappy PSU's which come with some crappy cases, but if they want advice, I would feel bad if they followed my advice, and had their PSU die on them. Wouldn't you ?

 

[dclive]

Originally posted by: MarcVenice
And yet again I disagree. It's nice to see your 300w PSU managed to power all of that. How long did you have it up and running though, and what did you do with it ? Office work all the time ?

If you look at these charts for example: http://www.extremetech.com/art.../0,1697,2217143,00.asp

Perhaps running 3dmark06 and 2 instances of orthos give unrealistic figures, it be nice to see the power consumption when they run Crysis for example, but still, I doubt it will be that far off.

For all I care people use crappy PSU's which come with some crappy cases, but if they want advice, I would feel bad if they followed my advice, and had their PSU die on them. Wouldn't you ?

I did 3D games (CoH) and VMWare work - keeping 3 VMs in memory and working at a time. It worked great. Given Anandtech's numbers, and the numbers posted above (in your previous post it was clear that a 350W PSU would be plenty) it's pretty clear that suggestions for 520W and 620W PSUs are a bit silly, wouldn't you agree?

 

[MarcVenice]

No I don't agree. The numbers I gave showed maximum load being 302w for a HD3850 and 317w for a 8800GT. Unless you fancy running your PSU at maximum capacity with all risks that come with that, you should in fact get a 500w PSU, or perhaps a qaulity 430-450w PSU, like Seasonics S12 II 430w, or a Corsair 450vx.

 

[dclive]

Bear in mind you're in a forum where people regularly overclock their CPUs by 50 and 100 percent.

Bearing that in mind, please tell me exactly what the issue is with running a PSU at *under* it's rated load capacity?

 

[MarcVenice]

I'm not quite getting you. Nothing wrong with running it at under it's rated load capacity, definately something wrong with running it NEAR it's rated load capacity. Most people will agree with me that you don't want to stress your PSU more then 70% or so of it's maximum. Some other people even recommend that number be 50-60%.

 

[dclive]

Originally posted by: MarcVenice
I'm not quite getting you. Nothing wrong with running it at under it's rated load capacity, definately something wrong with running it NEAR it's rated load capacity. Most people will agree with me that you don't want to stress your PSU more then 70% or so of it's maximum. Some other people even recommend that number be 50-60%.

Why? How is it stressed? We don't worry about over 'stressing' our CPUs by 50 to 100% over Intel's rated spec - why are we worried about running our PSUs at their rated spec?

 

[MarcVenice]

Okay, then bear in mind overclocking is not recommended by Intel or AMD. Then bear in mind PSU manufacturers admit that their units are suspect to degradation, for example, they rate units at a certain temperature to output it's maximum capacity, for an antec earthwatts that would be 50c continous. For a coolermaster that might be 30c continous, with a loss of 3-4w per each degree the unit goes up. Running a PSU near it's maximum capacity will surely heat it up. Also keep in mind that Nvidia recommend 28a for a 8800gt, and it used to be 30a for a 8800gts IIRC. And, like I said, these people are looking for advice, can you, honestly, tell them to buy a 300w PSU, and are 100% sure they can run a heavy rig on it, for let's say 2-3 years in a row with 'normal' usage, whatever that would be, 3-4 hours of gaming a day, few hours of browsing the internet ?

 

[dclive]

For the listed machine, yes, it should be fine. On the off chance it's not, do you feel it's appropriate to have people spend hundreds of dollars (and mountains of time) fiddling with the PSU? I don't. For the vast, vast, vast majority, the smaller PSUs will be fine. Note I'm not saying a 300W PSU works for everyone (you can clearly see in the other post today you posted to where I spoke of a 350W PSU) but I think this 520W PSU / 620W PSU (not to mention 800W PSU) business is for the birds.

 

[MarcVenice]

I think it's simply very reasonable to spend 50-70$ on a 400-500w PSU, if you are spending 1000$ on all other components. And once again, it's not so much the wattage, it's the amps that are important to power those high-end videocards and qaudcore CPU's.

 

[dclive]

$1000 on other components? Take a look at the OP's original post.

Even then, it's silly. There's just no need for it. I advocate buying what people can actually use, and the benchmarks all prove that for most people and with most machines, they'll never approach the 520/620W levels - ever. Most won't get to half that.

I choose to believe Anand's ratings more than ExtremeTech's. My reasoning is my Kill-A-Watt results more closely match Anand's, and I cannot see how E.T. did their power measurements. (Do you know?)

 

[MarcVenice]

This isn't the only thread you posted on, advocating the use of 300-350w PSU's for builds with a decent CPU and a high-end videocard... And doh, they won't approach 500w, good thing they don't, because we don't want that !!! And I think E.T. might been using a kill-a-wat too, since they were measuring from the wall.

"Power Utilization: We'll test the power draw from the wall on our test machine in three different scenarios. First, we measure how many watts the system draws when idle at the Vista desktop. Then we measure peak power during 3DMark06, running at 1920x1200 with 4x anti-aliasing enabled. Finally, we stress the whole system by running the 3DMark06 test again with two instances of Prime95 running in the background."

 

[dclive]

Then that explains it - drop the numbers by 20-25% since the power the computer needs from the PSU (ie supplied by the PSU) isn't the same as the power the PSU requires from the wall.

That makes the need for the 350W PSU for the 3870 and 3850 models unnecessary - the 300W PSU would supply enough power.

 

[Aluvus]

Originally posted by: dclive

Why? How is it stressed? We don't worry about over 'stressing' our CPUs by 50 to 100% over Intel's rated spec - why are we worried about running our PSUs at their rated spec?

How are those things related in any way?

 

[dclive]

Originally posted by: Aluvus

Originally posted by: dclive

Why? How is it stressed? We don't worry about over 'stressing' our CPUs by 50 to 100% over Intel's rated spec - why are we worried about running our PSUs at their rated spec?

How are those things related in any way?

I find it amazing that we're all fully willing to overclock a CPU, but we're not willing to overclock - in fact, we talk of drastically 'underclocking', so to speak - the PSU.

The PSU is made to output a rated capacity. If we don't surpass that capacity, I don't see the problem. People act as if 'stressing' the PSU (by using a 620W PSU to deliver 400W of power rather than 300W of power) is going to hurt it in some way. I find that ... odd.

 

[Aluvus]

Originally posted by: dclive
I find it amazing that we're all fully willing to overclock a CPU, but we're not willing to overclock - in fact, we talk of drastically 'underclocking', so to speak - the PSU.

The PSU is made to output a rated capacity. If we don't surpass that capacity, I don't see the problem. People act as if 'stressing' the PSU (by using a 620W PSU to deliver 400W of power rather than 300W of power) is going to hurt it in some way. I find that ... odd.

Do you think it is possible that the factors that limit a processor's clock frequency might be completely different from and unrelated to the factors that limit the current a power supply can deliver? And that those differences might indicate that your analogy is invalid?

 

[dclive]

Originally posted by: Aluvus

Originally posted by: dclive
I find it amazing that we're all fully willing to overclock a CPU, but we're not willing to overclock - in fact, we talk of drastically 'underclocking', so to speak - the PSU.

The PSU is made to output a rated capacity. If we don't surpass that capacity, I don't see the problem. People act as if 'stressing' the PSU (by using a 620W PSU to deliver 400W of power rather than 300W of power) is going to hurt it in some way. I find that ... odd.

Do you think it is possible that the factors that limit a processor's clock frequency might be completely different from and unrelated to the factors that limit the current a power supply can deliver? And that those differences might indicate that your analogy is invalid?

Can you directly address the comment I wrote?

 

[Aluvus]

Originally posted by: dclive
Can you directly address the comment I wrote?

Sure. Your analogy is completely and totally wrong. The claims that you have built on it are logically invalid. Several of your other claims are either based on misrepresenting what other people have said or on simplifying things incorrectly. Is that sufficiently direct?

When people run processors above their "official" frequency, the actual impact on the processor's lifespan is very small. Running transistors at a higher frequency does not age them that much faster; it mostly means they need better cooling. The analogy you should be building is to voltage. Increasing supply voltage does age the transistors faster, and can induce early failure (see Sudden Northwood Death Syndrome). For that reason, reasonable people are very careful about increasing voltages above spec, and do so with the knowledge that there is a greater risk involved.

When you run a power supply at or near its ratings, you are physically stressing it, sometimes quite a bit. Power supply ratings tend to be fairly to very optimistic (unlike processor frequency and voltage ratings, which are fairly to very conservative) to begin with. Asking the supply to deliver more power means it must dispose of more heat (which, even in an "enthusiast" system, may be difficult) and will age the capacitors faster. But they will age anyway, over time. This is the principle reason that good power supply recommendations "de-rate" the supply some; a power supply that can comfortably deliver 22 A @ 12 V today may not be able to comfortably deliver that same amount of current a year from now, and in 2 years it will be in worse shape yet. There is no good purely-mathematical de-rating curve that will work in all circumstances, unfortunately, which means this becomes something of a black art.

Aging is especially problematic for supplies like the SmartPower, that used bad caps. These supplies work just fine if cooled adequately and if they do not have to supply a great deal of power. But the more you ask of them, the higher the odds of failure go, and not in a linear fashion. Running a SmartPower at its rated limits for an extended period of time, unless you have spectacular cooling, is unwise.

There are a great deal of other differences that you have overlooked. Processors, for instance, are individually tested. Every single part went through a testing system and was confirmed to be able to operate within official specs. Power supplies are not generally subject to such scrutiny; many vendors are quite happy to test a small sample of their product (sometimes with rather heroic cooling measures) and call that good enough.

Further, you have failed to present any compelling actual evidence that even an ideal power supply with the ratings of the one in the OP could drive a system with an 8800GT etc. You've provided us with one piece of anecdotal evidence, but even that is not meaningful because you have failed to identify any of the rail ratings for the supply that you used. From your description, I am not sure that you understand why that information would be important. Even ignoring that, a single data point doesn't mean much.

And since no one here has posted a direct measurement of the power drawn by a GeForce 8800GT under load, and the OP hasn't even decided on a processor yet, your conclusion that the power supply in the OP "should be fine" is shaky at best. The OP might be able to get by with the sort of configuration he is considering; plenty of people tempt fate and do OK for a year or so. But there is not enough information here to reasonably say that that is so.

In the future, it would be wise to gather more information before drawing conclusions.

 

[dclive]

Even if I don't agree with your conclusions, I admire your thoughts.

Perhaps you could tell us why you believe an 8800GT with a processor as what the OP has listed so far, in a system defined by what he's listed so far, would have a problem with 350W. Derek Wilson from Anandtech says about 210W: http://www.anandtech.com/video/showdoc.aspx?i=3140&p=13 ... which suggests a decent 350W PSU would be plenty. Those numbers do track with Anand's (from Anandtech, obviously) previous tests.

Without better numbers, and with historical agreement from Anandtech of these numbers, I think they're credible and valid. From my read of the OP's interest, assuming no wild overclocks, he'll easily (easily!) be able to satisfy his needs and have mountains of headroom.

Are there any studies of this black art, as you mention, or is this discussion hypothetical? How do you know the 'half life', so to speak, is 2 years and not 20, for example?

 

[natrap]

Does anyone think this is of any value http://www.extreme.outervision.../psucalculatorlite.jsp ?