Dead Seasonic PSU! Please help!

[wedge]

I have no idea what's going on here.  I'm running two 4x7950 mining rigs.  I'm running a Seasonic X-1050 psu for each of them.  These rigs have been very unreliable.  I've had them almost a month, and they've only been running for half that time, because they are constantly locking up or shutting down on a daily basis.  It's taking all my time to babysit these things, it's not worth what I'm getting out of it.

If I ran even a small overclock, they would usually shut down in under 5 minutes.  If I ran the cards at stock, the system seemed stable.  But they still would shut down eventually, it just took longer.

The first psu completely died a couple weeks ago, and the second one just died this morning.  This is going to be the third one that I RMA.


What am I doing wrong?  These are supposed to be the best power supplies available, and they are MORE than big enough to handle 5 cards.  Right?  So what's going on?  Why is this happening?

[desired_username]

1000W seems a little low to me. I would've used 1200-1500W for piece of mind.

Can you test them with a bigger PSU?

Also, How's your airflow?

[wedge]

I didn't just pick any random power supply.  I did plenty of research in advance and did the math to figure out what size to get.
Here's what I calculated as a worst case scenario: 188W per gpu + 100W the rest of the system + 20% headroom, that totals 1022W.  So I figured a 1050W should be perfect.  Where did I go wrong?  I even checked a few different online psu calculators.  Most seemed to say around 950W.  Which is the same as what I calculated, but without the 20% safety margin.



Airflow is good.  Running them open air, using 20cm risers.  The cards are sitting with almost 3-1/2" spacing between them.  So there's plenty of airflow.  They are averaging between 75 - 77 degrees.


Here is a screenshot of what it looked like the moment it froze this morning.  Running very nicely, nothing overheating.  Nothing to indicate there might be a problem.

[zackclark70]

4 7950 vaporx cards pull under 950w from the wall getting 630kh each not sure about other brands though if you want help with setings give me a shout 

[wedge]

4 7950 vaporx cards pull under 950w from the wall getting 630kh each not sure about other brands though if you want help with setings give me a shout 

My cards are Gigabyte WindForce 3.


Also, I should mention, that when I ran 3 cards, everything seemed to run rock solid stable, no issues for days. 

[zackclark70]

what are your settings ? / what program are you using ? Edit >are you running ccc in the background or msi afterburner ?

[wedge]

Running BAMT as the OS.  So there is NOTHING else happening, except for mining, no other programs running.

I've tried a lot of different settings, nothing seems to make any difference.  I was originally overclocking to 1085core, 1375memory.  But it's been running at stock clock speeds for the past few days, trying to find if its stable.  I've never done any overvolting.

[zackclark70]

Running BAMT as the OS.  So there is NOTHING else happening, except for mining, no other programs running.

I've tried a lot of different settings, nothing seems to make any difference.  I was originally overclocking to 1085core, 1375memory.  But it's been running at stock clock speeds for the past few days, trying to find if its stable.  I've never done any overvolting.

my setings are 1050mhz core 1250mhz ram 1.050-1.060v using windows and reaper on windows

[candoo]

Dude those PSUs are not made to run at 100 or 110% all the time.

If your system needs 1200 watt then you buy 1500 watt psu. Give him some air

[zackclark70]

Dude those PSUs are not made to run at 100 or 110% all the time.

If your system needs 1200 watt then you buy 1500 watt psu. Give him some air

i run 4 cards and pull less than 950w from the wall so the rig is not even using 820w from a 1250 psu 

ever a 1050w psu would be ok for it 

[wedge]

Dude those PSUs are not made to run at 100 or 110% all the time.

If your system needs 1200 watt then you buy 1500 watt psu. Give him some air

That's the thing, the system should only need around 900.  I did leave some headroom with a 1050.  That's why I'm so confused.

Even if the current draw is more than I'm expecting, it shouldn't be enough to blow the power supply.  Especially one of these Seasonic's that are supposed to be the best of the best.



I guess I'm going to have to get something larger.  But I'd like to understand why....

[zackclark70]

what voltage do you have the vddc set at ?

[wedge]

what voltage do you have the vddc set at ?

I'm not specifically setting voltage, just letting it default to whatever it wants.
One card runs at 1.169, the other three run at 1.25, IIRC

[wedge]

I was about to buy a killawatt meter.  But then I remembered that I can check kW measured from the meter outside the house.

So, I just measured the wattage from the rig that's still working.  I turned off the rig, and pretty much everything in the house, and made sure no appliances were on or anything like that.  I turned on the rig, and waited for it to start mining, then read the increase in kW from the meter. 

That rig is apparently drawing ~1300W.  This is measured from the meter outside the house, so it's not 100% accurate.  But that's waaay more than I expected.
A 1050W psu, that is 85% efficient, should never draw more than ~1200W at max load.

So that certainly confirms what I already knew, this psu isn't enough. 


But there's still more questions here.  Why is the power draw so high?
zackclark70, you say you've got essentially the same setup I do, but you're only drawing 950W from the wall.  That is exactly what I expected from mine before I started.  I must be doing something else wrong here, or there's something I'm not aware of. 

Where does the extra 400 Watts come from?

[crazyates]

Your PSUs were prolly stressed beyond what they were designed for. This would account for the instability, as well as the pre-mature death.

First of all, you're not going to be pulling 188W per GPU at 1.25V. You're looking at more like 250W, depending on your OC. That's at least 1000W just for the GPUs, add 100W for the system, and then you've got no headroom.

My recommendation? RMA them and get new ones back. Then flash those Gigabyte 7950s to the F43 BIOS, which lowers the voltage down to 1.09, which will lower your power consumption by quite a lot. Then you might get it down to <200W per GPU. If you do that, your PSU will have it's breathing room back, and you should be a lot more stable.

[zackclark70]

are you cards v unlocked ? as lowering the vddc down to 1.050 will save you a lot of power

( 1050mhz core 1250mhz vrm and 1.050v gets me 630kh per card and 900-950w from the wall depending on the room temp)

Edit you can also under clock and undervolt the cpu

[fractalbc]

I was about to buy a killawatt meter. 

That would be a wise investment. 

4x200 + 100 is within the rating of that supply but those are for stock cards.

A simple guideline is to keep the AC reading on the kill-a-watt less than the supply DC rating.  This leaves you with a bit of margin.

You may have to limit your overclock to stay within that capacity.

[wedge]

My recommendation? RMA them and get new ones back. Then flash those Gigabyte 7950s to the F43 BIOS, which lowers the voltage down to 1.09, which will lower your power consumption by quite a lot. Then you might get it down to <200W per GPU. If you do that, your PSU will have it's breathing room back, and you should be a lot more stable.

That's it!

I just flashed the F43 BIOS (followed this guide)
Now running at 1.09v
Power measured at the meter has dropped to 900W
Rig is running fine so far at my previous o/c without any shutdown issues.
Cards are running about 3 -5 degrees cooler than before, at lower fan speeds too
The rig is overall running at 2.5Mh/s

That bios must be magical or holy or something.  I'm going to print out the code of that bios, and frame it, and mount it on my wall, and give worship to it on a daily basis.


I bet I could push the overclock further now.  But I think I've done enough for one day.  I'll let it run, and make sure it's 24/7 stable.

This is what I needed all along.  I never realized that two stock bios's could be so drastically different.  And what's really funny is that the older one is the better one.


Thanks for the help guys!



Edit: Let's have a moment of silence for the pair of poor psu's that I killed.  They were so young and strong and beautiful.  Taken much before their time...

[crazyates]

My recommendation? RMA them and get new ones back. Then flash those Gigabyte 7950s to the F43 BIOS, which lowers the voltage down to 1.09, which will lower your power consumption by quite a lot. Then you might get it down to <200W per GPU. If you do that, your PSU will have it's breathing room back, and you should be a lot more stable.

That's it!

I just flashed the F43 BIOS (followed this guide)
Now running at 1.09v
Power measured at the meter has dropped to 900W
Rig is running fine so far at my previous o/c without any shutdown issues.
Cards are running about 3 -5 degrees cooler than before, at lower fan speeds too
The rig is overall running at 2.5Mh/s

That bios must be magical or holy or something.  I'm going to print out the code of that bios, and frame it, and mount it on my wall, and give worship to it on a daily basis.

Glad I could help! The newer BIOS should help keep the VRM temps down a little bit as well, which you DEFINITELY WANT!

[wedge]

Glad I could help! The newer BIOS should help keep the VRM temps down a little bit as well, which you DEFINITELY WANT!

Thanks for the pre-emptive warning!  I could've used that about the bios...
Anyway, yeah I've already noticed how hot the VRM's are.  I've already got a fan blowing on the back of the first card.  Not any of the others cause they already have air moving behind them from the card behind them.  The first card doesn't have that, hence the added fan.  I'll probably stick on some heatsinks on the back of each of them.  Something like this:

http://img297.imagevenue.com/img.php?image=523696347_DSC01701_122_419lo.JPG

But I can do better.  Much bigger heatsinks than that!  I'm thinking I can pick up a box of old cpu heatsinks, like socket 7 old, for a couple bucks.  I can hacksaw each of them down the middle, and fit one or two just like in that picture.  That should do the trick.