Message boards :
Number crunching :
New PSU, capable of feeding O.C:d CPU and 4 GTX780's -- oppinions, hard facts. Please anyone?
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
@link: from nvidia 780 specs (just for one 780 no OC): Minimum of a 600 Watt power supply. (Minimum recommended power supply with +12 Volt current rating of 42 Amps.) Be sure you understand the word: minimum @batter up - I'm sure it´s works for now. Let´s wait a year or 2 and we talk about again... waiting... time bomb deployed... For each pair of 780's I personally wouldn't be putting anything less than a really good 100Amp single rail 1200W PSU 1200 W just for the 2x780 that´s what i sugest from the begining... Finaly Jason arrives to the thread, so now it´s with him, he is our master guru. My advice: allways follow his lead. :) <edit> from my personal experience... i blow at the time a expensive 850-950 (not sure)W gold TT PSU with about a year in a host who runs 24/7 with just 3x560 GPU´s (not ti) runnig at stock speed. |
tbret Send message Joined: 28 May 99 Posts: 3380 Credit: 296,162,071 RAC: 40 |
As a long-time audiophile, I get it. You are "listening" to something typically in the 1w range. However, transients can create a terrific demand, suddenly, and for an extremely short amount of time (milliseconds). BUT, that causes the power supply to do "funky things" and results in s slew/skew rate that you wouldn't want your expensive speakers dealing-with (compression and clipping). On the other hand, your woofer is sucking-up a tremendous amount of the total current draw from the amplifier. Your RMS amplifier rating may be 100x your "average" listening level, but if you want your amplifier to never "clip" or compress, you need abundant headroom in your amplifier's power. Lots of people "bi-amp" for that reason (the stuff you are listening-to never gets CLOSE to sucking-up the power the woofers <which you feel more than hear> are demanding). That's why I have put a 750w single rail power supply driving two GTX 560Ti video cards. The "TDP" on them is something like 170w, so I'm using 340w of 750w IF you believe that the cards are full-on and up against the TDP rating when crunching. I don't. I'm guessing (and it is a guess) that the cards are pulling something less than 125w each when they are "full on" crunching. So, I've got approximately three times the power I'm guessing I use available for "transients". If you translate that to audiophile talk, I've got somewhere between 3db and 6db of "headroom." I'd never run a stereo system very hard with so little headroom. I might be listening at 1w, but I want 500w "available" should the material call for it, in milliseconds. Unfortunately, I can't afford a 500:1 computer power supply and if it were ever called-on to produce that much power it would fry my household wiring. My point is only this: There is a point at which you have to stop. For me, having 1,500 to 1,800 watts available to the system is all I can reasonably do, even if the ideal is to get 100:1 or 500:1 available compared to RMS. Surely, Jason, sticking meters to it all you have a good idea of what the maximum millisecond draw of a video card might be compared to it's "average" maximum draw. If you tell me it's 1000:1, I'd believe you. I'd also just give-up because nobody sells a PSU that big. |
petri33 Send message Joined: 6 Jun 02 Posts: 1668 Credit: 623,086,772 RAC: 156 |
The audioohile example and Jasons AC reference got me to think of a 1F car audio capacitor providig amps during the transients.. To overcome Heisenbergs: "You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13855 Credit: 208,696,464 RAC: 304 |
... of what the maximum millisecond draw of a video card might be compared to it's "average" maximum draw. If you tell me it's 1000:1, I'd believe you. I wouldn't- the maximum a video card can draw it what it's rated for. It will draw more than that if it is being overclocked, but there will still be a hard limit on what it can draw. Unless of course something dies & shorts out. Grant Darwin NT |
Link Send message Joined: 18 Sep 03 Posts: 834 Credit: 1,807,369 RAC: 0 |
@link: from nvidia 780 specs (just for one 780 no OC): Minimum of a 600 Watt power supply. Yes, for the whole system, not just for the GPU itself. So for 2 GPUs you don't need 2x42A. BTW, "minimum recommended" != "minimum". nVidia for sure has there good safety margin in their recommendation since they have to expect, that many people will use cheap PSUs which also might be not that new anymore. tbret wrote: As a long-time audiophile, I get it. Are you sure you want to compare the "needs" of audiophiles with those of computers? You don't need 500:1 PSU for the computer, computers don't need it. A PSU can be safely run at 70% load if it's a good one. So even 2:1 would be more than enough. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Surely, Jason, sticking meters to it all you have a good idea of what the maximum millisecond draw of a video card might be compared to it's "average" maximum draw. If you tell me it's 1000:1, I'd believe you. I'd also just give-up because nobody sells a PSU that big. That's right. Instantaneous peaks demand for a single GPU, specifically on a PCIe Rail, will mostly depend on, and hopefully be smoothed by, the individual converter phases on the Card. For the ATX12V specification, under transient load of 1A/usec, the PSUs must slew to 50% of their rated rail current load -> in standard terms this 'max' rate is a minimum performance spec for the design of the PSU. The problem with this picture is specifically with adding multiple high power GPUs, and coincident peak draws. Multiple cards multiply that load transient rate intermittently. If on the same big rail (easier, why most experienced recommended them), This similarly divides the effective current slew on the PSU rail, the min spec requirement to reach 50% current now is down to 25% specced minimum. If on multiple rails then the rails might remain 'stiffer' but only still slew at 50% rate of a lower peak as well, so at first appearing simpler, getting a multiple rail rig stable ends up needing even more overkill and complexity. Now for the average single GPU customer, within reference specs you're fairly safe within the margins of specs and reference designs (barring extremely shoddy PSUs and bad cards etc). Why do you see power supply units sold as SLi and/or crossfire Certified ?, and extreme Overclockers giving a whole supply per GPU, going to the extent of hard modifying power circuitry ( e.g. KINGPIN and TIN ) ? Is it pure marketing ? No. It's because multiple GPUs have special demands over and above atandard PSU specs and reference designs. Don't get me wrong, marketing forces are always at play too, but for the most part if you use multiple GPUs you need 'better' &/or multiple PSUs, with overkill, and the PSU industry has enough competition to ensure the companies have to keep things fairly real, providing relatively honest value for money (in the mid to high end anyway). Making recommendations by minimum becomes risky, and overkill wise. A 42 Amp (peak) per *single* card recommendation, quickly becomes 84 Amps with two, but with some 10% regulation allowance in the standards, we then get up to ~92 Amps, and require double the rate of change. Now start factoring in component aging / expected lifespan, some safety margin, and the risk to take is your own. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Batter Up Send message Joined: 5 May 99 Posts: 1946 Credit: 24,860,347 RAC: 0 |
Exactly; have to stop. My limit is one consumer grade computer with a 1500W limit; the maximum watts a 120V outlet can deliver. It really isn't that difficult to build a supercomputer. Cray did with the Titan supercomputer what the top crunchers do; add more GPUs, with the Titan there are 18,000 GPUs. Sad, SETI will only accept 64 GPUs. |
ivan Send message Joined: 5 Mar 01 Posts: 783 Credit: 348,560,338 RAC: 223 |
@link: from nvidia 780 specs (just for one 780 no OC): Minimum of a 600 Watt power supply. Thermal and Power Specs: 95 C Maximum GPU Temperature (in C) 250 W Graphics Card Power (W) 600 W Minimum Recommended System Power (W) One 8-pin and one 6-pinSupplementary Power ConnectorsFrom http://www.geforce.co.uk/hardware/desktop-gpus/geforce-gtx-780/specifications My take on that is minumum recommended power for a dual-GPU rig would be 850 W, or 1350 W for a quad. |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
For each pair of 780's I personally wouldn't be putting anything less than a really good 100Amp single rail 1200W PSU, even though the RMS draw would likely be less than half that. It's the coincident peak current draws that aren't specified for multi-gpu, and resulting required effective slew. I could only add: Follow Jason´s lead he know what his talking, from years of good/bad experiences and sure a lot of burned PSU´s. I use a 1350W TT in my 2x690+670(770W), 1000W coolermaster in the 690+770(470W), a 750W-C3 in the 2x670(340W) or 780(250W) hosts and works perfect for years even in my hot tropical enviroment with NO AC. Do not make savings on the PSU´s, there is nothing worse than wierd defects caused by problematic PSU´s or without current capacity. |
Wes Kay Send message Joined: 12 Jul 04 Posts: 45 Credit: 3,125,395 RAC: 0 |
url[/http://www.extreme.outervision.com/PSUEngine]/url The link is to a PS calculator, two versions available...Lite(free) and Pro. It's a general guide of course but is reasonably thorough with component add-ons and such. IMHO I think you might want to go with two PS, especially if going above 90% recommended system load but I always leave a 25%-30% safe haven when calculating load requirements for a psu purchase. Regards, Wes Damm, I can never get the URL thing to work...:( |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
[/http://www.extreme.outervision.com/PSUEngineurl] http://www.extreme.outervision.com/PSUEngineurl You just need to add "url#.........."/url# (change the " for [ and # for ] to work) |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
[/http://www.extreme.outervision.com/PSUEngineurl] I always loved that calculator. If you build a (now old) modest Core2 system, mild overclock, with decent fans, 4 x HDDs, and a single GTX 480, then allow for 3 years capacitor aging (all available in the options) then you come to about 850W. That pretty quickly explains the quick demise of my Seasonic Gold 700W, a venerable PSU. Again it all comes down to personal risk choices. That's not something I'd be comfortable cutting corners on for the sakes of $50 or so. If you must, Cut corners on the case or the flashy light thingies. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
SciManStev Send message Joined: 20 Jun 99 Posts: 6658 Credit: 121,090,076 RAC: 0 |
i have to agree with Jason. Over kill, over kill, over kill. That is the safest way. I ran a dedicated 30 amp circuit to run 1 computer with an AX1200 Gold PSU. I am having issues with it at the moment, but it can take all the power it wants without me worrying at all. I usually run 850 watts continuous just from the tower, and more from the support equipment. That is with two againg GTX 480's. Newer GPU's are much more efficient, but I over clock my system to the maximum possible. Steve Warning, addicted to SETI crunching! Crunching as a member of GPU Users Group. GPUUG Website |
Iona Send message Joined: 12 Jul 07 Posts: 790 Credit: 22,438,118 RAC: 0 |
Have to agree. Even with my (now) mere Core 2 (soon to be i5) and the power-crazy HD4890 I used at the time (running a 6870 these days), I upgraded to a Corsair HX 850, even though the PSU I was using at the time was, on paper, quite adequate. Why change an adequate PSU? The deciding factor, was, as the 4890 was a bit of a power guzzler and having noticed a 'drop off' in voltage, it then followed that as it produced quite a bit of heat (they were known for that), so the PSU also needed to do it's job, at fairly high temperatures, especially given that the room the PCs are in, is often at 30+c, during a standard British summer. Essentially, it was the HXs ability to keep producing maximum power at 50c (I think) that was the real clincher. Operating temperature is something that is often overlooked and it is something that needs to be factored in....every component in any PC is affected by temperature and in extreme cases individual components could be trying to run outside their tolerances, which could have considerable negative effects on everything else. Don't take life too seriously, as you'll never come out of it alive! |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.