I figure on boot it draws at most 700w and on running it should draw about 550w at most.
Averaging 550 W does not mean the peaks are 550 watts and it is the peaks that the power supply has to cope with gracefully. As a retired electrical engineer, I used to believe that the folks urging amazingly high capacity rated power supplies for GPU's were exaggerating grossly.
Then I happened to buy two UPSes of a different model than I had used previously which showed an on-screen power consumption number as seen by the UPS, which updated somewhat faster than my Kill-a-Watt does. I was shocked to see much more time variation on the scale of a few hundred milliseconds than I had expected. And that is all the way out at the power cord of the machine, where the real short-term variation has been considerably buffered by the bypass capacitors in the PC power supply. So I don't know what the real variation in consumption at the GPU is, but currently believe it to be hundreds of watts (and I have a mere GTX 460).
It well may not be an issue for you, but I post this as a warning to others reading the thread that it may be an issue for them.
LOL...
You thought the kittyman was exaggerating when I recommended high end single rail PSUs with overkill capacity, eh?
I have seen some amazing spikes on my Killawatts.....and like you said, they are somewhat slow to respond.
A premium PSU like the PC Power and Cooling Silencer series that I run will handle such spikes far better than many lesser models of the same wattage rating.
Sometimes you DO get what you pay for.
In my years of crunching i never see one who say: "My PSU is to big"...
Double rails could be teoricaly a good and cheap solution... but for crunching... go to single rail heavy eficience PSU with at least 30% or more spare power than you "nominaly needs"... just in case...
In my years of crunching i never see one who say: "My PSU is to big"...
Double rails could be teoricaly a good and cheap solution... but for crunching... go to single rail heavy eficience PSU with at least 30% or more spare power than you "nominaly needs"... just in case...
I'd rather see every (High-End)GPU getting his 'own juice', but 30% 'overhead should take care of this also.
Looking at 2 Kill-a-Watt's I see some change in Amp's or Watts.
I7-2600 + 2 HD5870, uses ~500Watts when running 2 Collatz C. and 8 Einsteins Gravitationel Wave Search, or LHC work[/i], Docking, SETI.
doing 4 BRP4ATI and 6 GWSearch it's using ~450Watt/hour. (PSU = 1KWatt)
Both NVidia (CPU Q6600+ X9650 {@3600MHz), GTX470 + 480, respectively,
use ~320Watt/h and ~375Watt/h for the Extreme CPU. PSUs= 650Watt/h.
All GPUs run at about 85C CPUs at 88C(I7-2600), 66C (X9650)and 80C(Q6600).
A high Power consumption leads to big PSUs and lots of heat.(OHMs Law)P=VxI (or
P=I^2 xR or P=V^2 /R).
And PSUs above 600Watt have a higher efficiency, compaired to lower rated ones.
One component, which has never failed, is the P.S.U.
They have multiple detectors for 'shorcut' , 'overload'and 'heat', also
voltage range in which they shouls operate, error %.
The HX1000W is packed with the latest technology and features:
• Guaranteed to deliver 1000W at 50°C
• 80%+ energy efficient under real world load conditions
• True dual rail design for a total of 80A on +12V
• Multi-GPU ready
• 105°C solid state capacitors
etc...I dont think PSU is causing any problems
That machine used to have 2x 560s in it and 12 drives (9x 2tb's and 2 SSD and optical) I moved 4 of the 2tb ones to a NAS about 2 months ago so compared to 2x 560s and 12 drives or 560 + 660 and 8 drives power use is probably slightly less now overall but HDDs use bugger all power once theyre spun up.
My plan is to repaste the 660ti today and run the GPUs in different machines for a few days to see if its one or the other causing the problems or is it in cc_config possible to tell the machine which GPU to use?
You have an excelent PSU (i have one very similar and drives 2x670 on a single I5 host) but take a Look in the PSU manual page #5, each rail have a combined max capacity of 500W (12V+5V). So depends on who you split the power cables and the devices you feed by each rail you could be easy overpass the safe zone. Don´t forget that is a Nominal capacity not the surge capacity, each time one WU start it´s drain a spike of power from the PSU (I don´t know why)
If you go to the Nvidia site and see even for the 660TI they call a minimum 450W is required. You say you have some HD and other stuff on your host (plus the MB and processor itself). All uses the 12V rail for power so try to balance the load by spliting all devices on the 2 rails.
My sugestion put you MB and one GPU (the less power hungry one) only on one rail, and the HD´s, the other GPU and whatever else you have on the 2 rail and look, you have nothing to loose if you try.
Of course that could be another problem but for now, everything points to that direction.
Yes I looked at that but the 450w they say is for system power i.e. gpu cpu hdds etc the whole lot. they say there that the gpu itself useds max 150w.
Based on that I figure I need max 300w for gpus, about 160w for hdds on spin up only then a little for fans (it has 7 ranging from 80mm to 140mm including the cpu HS fan).
As I said it ran fine for like a year with 2x 560s in it with 4 more HDDs to run as well.
On a side note I added a old 8800gtx to one of the other boxes to try to pick up some of the slack while I dick around with this one.
Anyone know of a 12 step plan to deal with this RAC obsession :)
Lack of power on one rail might be a problem and am looking at how its all plugged in now. (pic coming)
The HX1000W is packed with the latest technology and features:
• Guaranteed to deliver 1000W at 50°C
• 80%+ energy efficient under real world load conditions
• True dual rail design for a total of 80A on +12V
• Multi-GPU ready
• 105°C solid state capacitors
etc...I dont think PSU is causing any problems
That machine used to have 2x 560s in it and 12 drives (9x 2tb's and 2 SSD and optical) I moved 4 of the 2tb ones to a NAS about 2 months ago so compared to 2x 560s and 12 drives or 560 + 660 and 8 drives power use is probably slightly less now overall but HDDs use bugger all power once theyre spun up.
My plan is to repaste the 660ti today and run the GPUs in different machines for a few days to see if its one or the other causing the problems or is it in cc_config possible to tell the machine which GPU to use?
Forget the PSU, and may be forget about repasting the 660... Ive been looking at your invalid tasks and all the tasks Ive seen (Ive not looked every one of them but the first ten and some others randomly choosen) show this:
[08:44:41][5756][INFO ] Using CUDA device #1 "GeForce GTX 560" (336 CUDA cores / 1209.60 GFLOPS)
I realy understand what you say and seems logic but the 560 uses less power than the 660TI and works diferent, you are at the edge of the capacity of the rail, any small changes could cause the problem. Try to split and load balancing the power in the 2 rails you have nothing to loose and a lot to gain. That could not help and fix the problem, but certainly will extend your PSU life.
FYI i literaly melt in just 6 months a corsair 900W gold PSU just with 3x560 on a I7 Host with 3 HD... by power ratings that will never could be happening... that was the "bad way" i learn about the 2 rails PSU limits. And i see now you have 12HDs ??? at nominal 15W per HD thats a lot of power drain in the 12V rail!
Still mantain my sugestion:
Put you MB and one GPU only on one rail, and the HD´s and the other GPU on the 2 rail and look.
Hmm seeing now Horacio finding that the 560 is causing the problems I'll pull that gpu out for the time being then look at adding another psu into the mix.
The way its wired up now shouldnt cause any problems.
If one has any question about how much power their computer is drawing compared to the capacity of their PSU, a Kill-a-Watt is a rather modest investment and will display the actual power being drawn from the mains by the PSU.
I have 4 of them.
RE: RE: I figure on boot
)
LOL...
You thought the kittyman was exaggerating when I recommended high end single rail PSUs with overkill capacity, eh?
I have seen some amazing spikes on my Killawatts.....and like you said, they are somewhat slow to respond.
A premium PSU like the PC Power and Cooling Silencer series that I run will handle such spikes far better than many lesser models of the same wattage rating.
Sometimes you DO get what you pay for.
In my years of crunching i
)
In my years of crunching i never see one who say: "My PSU is to big"...
Double rails could be teoricaly a good and cheap solution... but for crunching... go to single rail heavy eficience PSU with at least 30% or more spare power than you "nominaly needs"... just in case...
RE: In my years of
)
I'd rather see every (High-End)GPU getting his 'own juice', but 30% 'overhead should take care of this also.
Looking at 2 Kill-a-Watt's I see some change in Amp's or Watts.
I7-2600 + 2 HD5870, uses ~500Watts when running 2 Collatz C. and 8 Einsteins
Gravitationel Wave Search, or LHC work[/i], Docking, SETI.
doing 4 BRP4ATI and 6 GWSearch it's using ~450Watt/hour. (PSU = 1KWatt)
Both NVidia (CPU Q6600+ X9650 {@3600MHz), GTX470 + 480, respectively,
use ~320Watt/h and ~375Watt/h for the Extreme CPU. PSUs= 650Watt/h.
All GPUs run at about 85C CPUs at 88C(I7-2600), 66C (X9650)and 80C(Q6600).
A high Power consumption leads to big PSUs and lots of heat.(OHMs Law)P=VxI (or
P=I^2 xR or P=V^2 /R).
And PSUs above 600Watt have a higher efficiency, compaired to lower rated ones.
One component, which has never failed, is the P.S.U.
They have multiple detectors for 'shorcut' , 'overload'and 'heat', also
voltage range in which they shouls operate, error %.
All these opinions on PSUs
)
All these opinions on PSUs :)
This is the PSU I have in that box.
http://www.corsair.com/professional-series-hx1000-80-plus-certified-modular-power-supply.html
The HX1000W is packed with the latest technology and features:
• Guaranteed to deliver 1000W at 50°C
• 80%+ energy efficient under real world load conditions
• True dual rail design for a total of 80A on +12V
• Multi-GPU ready
• 105°C solid state capacitors
etc...I dont think PSU is causing any problems
That machine used to have 2x 560s in it and 12 drives (9x 2tb's and 2 SSD and optical) I moved 4 of the 2tb ones to a NAS about 2 months ago so compared to 2x 560s and 12 drives or 560 + 660 and 8 drives power use is probably slightly less now overall but HDDs use bugger all power once theyre spun up.
My plan is to repaste the 660ti today and run the GPUs in different machines for a few days to see if its one or the other causing the problems or is it in cc_config possible to tell the machine which GPU to use?
Still could be a long shoot
)
Still could be a long shoot but:
You have an excelent PSU (i have one very similar and drives 2x670 on a single I5 host) but take a Look in the PSU manual page #5, each rail have a combined max capacity of 500W (12V+5V). So depends on who you split the power cables and the devices you feed by each rail you could be easy overpass the safe zone. Don´t forget that is a Nominal capacity not the surge capacity, each time one WU start it´s drain a spike of power from the PSU (I don´t know why)
If you go to the Nvidia site and see even for the 660TI they call a minimum 450W is required. You say you have some HD and other stuff on your host (plus the MB and processor itself). All uses the 12V rail for power so try to balance the load by spliting all devices on the 2 rails.
My sugestion put you MB and one GPU (the less power hungry one) only on one rail, and the HD´s, the other GPU and whatever else you have on the 2 rail and look, you have nothing to loose if you try.
Of course that could be another problem but for now, everything points to that direction.
Yes I looked at that but the
)
Yes I looked at that but the 450w they say is for system power i.e. gpu cpu hdds etc the whole lot. they say there that the gpu itself useds max 150w.
Based on that I figure I need max 300w for gpus, about 160w for hdds on spin up only then a little for fans (it has 7 ranging from 80mm to 140mm including the cpu HS fan).
As I said it ran fine for like a year with 2x 560s in it with 4 more HDDs to run as well.
On a side note I added a old 8800gtx to one of the other boxes to try to pick up some of the slack while I dick around with this one.
Anyone know of a 12 step plan to deal with this RAC obsession :)
Lack of power on one rail might be a problem and am looking at how its all plugged in now. (pic coming)
RE: All these opinions on
)
Forget the PSU, and may be forget about repasting the 660... Ive been looking at your invalid tasks and all the tasks Ive seen (Ive not looked every one of them but the first ten and some others randomly choosen) show this:
[08:44:41][5756][INFO ] Using CUDA device #1 "GeForce GTX 560" (336 CUDA cores / 1209.60 GFLOPS)I realy understand what you
)
I realy understand what you say and seems logic but the 560 uses less power than the 660TI and works diferent, you are at the edge of the capacity of the rail, any small changes could cause the problem. Try to split and load balancing the power in the 2 rails you have nothing to loose and a lot to gain. That could not help and fix the problem, but certainly will extend your PSU life.
FYI i literaly melt in just 6 months a corsair 900W gold PSU just with 3x560 on a I7 Host with 3 HD... by power ratings that will never could be happening... that was the "bad way" i learn about the 2 rails PSU limits. And i see now you have 12HDs ??? at nominal 15W per HD thats a lot of power drain in the 12V rail!
Still mantain my sugestion:
Put you MB and one GPU only on one rail, and the HD´s and the other GPU on the 2 rail and look.
Hmm seeing now Horacio
)
Hmm seeing now Horacio finding that the 560 is causing the problems I'll pull that gpu out for the time being then look at adding another psu into the mix.
The way its wired up now shouldnt cause any problems.
Any opinions on http://www.add2psu.com/ ?
It may well be the psu cant handle the 560 to 660 change and was at the edge of supply with 2x 560s.
If one has any question about
)
If one has any question about how much power their computer is drawing compared to the capacity of their PSU, a Kill-a-Watt is a rather modest investment and will display the actual power being drawn from the mains by the PSU.
I have 4 of them.