TPU's WCG/BOINC Team

[[Ion]]

I just made an inventory on my three mayor rigs:
2x7970 - 1.25 days left of GPU work
2x7770 - 4 days left
1x7950 - 2 days left

4275 WUs ready to either upload or report from these three rigs.

Ion: what went wrong with your buffer? Did you crash just before they increase the buffer?

How did you manage to score four days of work on the HD7770s?

My main issue is that I blew the breaker on Sunday; that caused all of my rigs to turn off. A couple of them had errors resulting from the sudden power failure, so WCG wouldn't send them their usual quota of WUs (even when I got them back up on another circuit). But I don't think that explains everything--part of it was that I had a bunch going on and forgot to increase the buffers on some of the systems. I'm really not sure about the Xeon though---it's been running fine, but it would never get the 3 days of WUs it should have gotten (more like 12 hours). I'm going to fight with things once the project comes back.

 

[ThE_MaD_ShOt]

Some good news and some bad news: the power issue I had was only because I threw the breaker--so the repair man was able to fix it in thirty seconds. However, it does mean that, somehow, I managed to throw a 20 amp breaker,

Like I said you drawing to much Powah cowboy. I say 2 to 3 rigs max on a single line while running gpu work unit's. Under full load they draw more amps then one would think.

I keep it down to 2 rigs per single 20 amp breaker to allow room. Also you need to take in account for other things that may be on that line.

How did you manage to score four days of work on the HD7770s

The way i did it was that i increased my buffer by a day at a time. I let it download the extra units. When it was done I bumped another day and let it download again. Keep doing that until I had a full 5 day buffer.

 

[[Ion]]

Like I said you drawing to much Powah cowboy. I say 2 to 3 rigs max on a single line while running gpu work unit's. Under full load they draw more amps then one would think.

I keep it down to 2 rigs per single 20 amp breaker to allow room. Also you need to take in account for other things that may be on that line.

The way i did it was that i increased my buffer by a day at a time. I let it download the extra units. When it was done I bumped another day and let it download again. Keep doing that until I had a full 5 day buffer.

I've checked, and the 20 amp breaker just goes to my half of the room. I guess that twelve systems on 20 amps might have been too many. I think if I can put an i7 and the Xeon DP on my roommate's breaker I'll be fine. That and I'm likely selling two of the HPs tonight (-250W).

I'm surprised it gave you that much work--now I'm curious. The problem then is that WCG also tries to send me 5 days of CPU work too, even on rigs where the CPU is only used to power the GPU(s).

 

[mstenholm]

How did you manage to score four days of work on the HD7770s?

It could be because I had 2x7970 in the rig just before....

No I think that the limit for number of WUs in the buffer were increase to 3000 per GPU and with around 1500 done per day I have plenty.

When it comes to CPU work I think you have to earn the right to a big buffer over a certain time and you ran it on and off and hence proofed to be a somewhat unstable cruncher. The rigs that I have running 24/7/365 get nothing but rush jobs. New rigs never seen one.

 

[[Ion]]

It could be because I had 2x7970 in the rig just before....

No I think that the limit for number of WUs in the buffer were increase to 3000 per GPU and with around 1500 done per day I have plenty.

When it comes to CPU work I think you have to earn the right to a big buffer over a certain time and you ran it on and off and hence proofed to be a somewhat unstable cruncher. The rigs that I have running 24/7/365 get nothing but rush jobs. New rigs never seen one.

That's possible--since the BOINC Data folder has now been copied across about half a dozen different systems, so it probably appears a bit strange to WCG. I've noticed that they've been gradually giving it a bit more work at once...

 

[Jstn7477]

Depends on what machines you're running, but I keep my P55 rig in its own room because it has 3x GTX 460s and takes 600-650w usually on my LEPA G900 Gold PSU. I have 8 machines and 2 laptops on a 20A in the main office but there are only 2 folding cards in there atm (used to run 4). The Kill a Watt is an awesome device and like you said, the older dual core rigs usually take around 130w each without a substantial video card being used. This is why I love Ivy Bridge i7 chips, as the chip itself does 8K PPD, only eats ~60w @ 4.3/1.175v and there's an ~8w PCH and the rest of your stuff that probably adds up to about the same wattage as an old Core 2 Duo or Athlon 64 X2.

 

[ThE_MaD_ShOt]

I'm surprised it gave you that much work--now I'm curious. The problem then is that WCG also tries to send me 5 days of CPU work too, even on rigs where the CPU is only used to power the GPU(s).

My one rig that does gpu units is only set for Gpu unit's. I do not receive and cpu units on it. The other 2 rigs that do both received a ton of gpu units, but also got a bunch of cpu units also when I did the buffer change. Which is fine. If this outage last for any amount of time they might run out of gpu units but will still have cpu's units to fall back on to keep crunching. the way it sets I have about a 5 day or so buffer on all rigs. I will cut it back when the outage is over and let them clear out the buffer and reset it.

 

[[Ion]]

Depends on what machines you're running, but I keep my P55 rig in its own room because it has 3x GTX 460s and takes 600-650w usually on my LEPA G900 Gold PSU. I have 8 machines and 2 laptops on a 20A in the main office but there are only 2 folding cards in there atm (used to run 4). The Kill a Watt is an awesome device and like you said, the older dual core rigs usually take around 130w each without a substantial video card being used. This is why I love Ivy Bridge i7 chips, as the chip itself does 8K PPD, only eats ~60w @ 4.3/1.175v and there's an ~8w PCH and the rest of your stuff that probably adds up to about the same wattage as an old Core 2 Duo or Athlon 64 X2.

Three laptops (2x 35W C2D, 1x 35W i7), one Atom, three i7/GPU systems, the Xeon DP, and four or five C2D systems (one with an 8800GTS, the others GPU-less).
I'd estimate that the HPs I have draw maybe 110w each (65W CPU and a single HDD on a G31 board), the Atom ~35W, and each laptop no more than 75W. Xeon DP is probably just under 200w--not sure exactly about the i7s.

I still can't imagine all of this being more than 2400w--but apparently it must have been. I'll get it spread out as much as possible (but I'd imagine still having two i7s and the duals on one 20A breaker)

My one rig that does gpu units is only set for Gpu unit's. I do not receive and cpu units on it. The other 2 rigs that do both received a ton of gpu units, but also got a bunch of cpu units also when I did the buffer change. Which is fine. If this outage last for any amount of time they might run out of gpu units but will still have cpu's units to fall back on to keep crunching. the way it sets I have about a 5 day or so buffer on all rigs. I will cut it back when the outage is over and let them clear out the buffer and reset it.

That is true...too late now to help, but I appreciate it

 

[mstenholm]

My one rig that does gpu units is only set for Gpu unit's. I do not receive and cpu units on it.

The problem then is that WCG also tries to send me 5 days of CPU work too, even on rigs where the CPU is only used to power the GPU(s).

Hmm, 6.10.58 and only one GPU WU at the time (The_MaD_ShOt) does only get GPU work when asked for only GPU. Version 7.0.42 (and 7.0.44) disregard that and send you CPU work as well. I have detached at least 1000 CPU units in the past week and I still have about the same in my buffer on my GPU only rigs. I'm now keeping them for a rainy day (aka no new GPU work due to the update took 48 hours, not 24)

 

[ThE_MaD_ShOt]

Hmm, 6.10.58 and only one GPU WU at the time (The_MaD_ShOt) does only get GPU work when asked for only GPU. Version 7.0.42 (and 7.0.44) disregard that and send you CPU work as well. I have detached at least 1000 CPU units in the past week and I still have about the same in my buffer on my GPU only rigs. I'm now keeping them for a rainy day (aka no new GPU work due to the update took 48 hours, not 24)

My gpu only rig is running whatever version that is on Wcg's site. It also runs 5 wu's at a time. "toast" Here's a screen shot.

 

[mstenholm]

My gpu only rig is running whatever version that is on Wcg's site. It also runs 5 wu's at a time. "toast" Here's a screen shot.
http://img.techpowerup.org/130112/screenshot.jpg

Sorry if I suggested that you ran low numbers. I only look at RAC (19K) and the version number. I can see that you just only got it up and running and it can do so much more. However my point was that the older version apparently accepts if you ask for only GPU jobs whereas the new ones feed you the lot. If you want a newer version then go here http://boincstats.com/en/page/downloadBOINC

Sooner or later we all have to update to at least 7.0.44 if we want to run multiple GPU units....

 

[mstenholm]

OK start uploading. We are back on.

 

[johnspack]

Cool, got my wus uped. I notice I only had gpu wus in my cache, but in device manager I set it to no cpu wus and no other projects. Maybe that helps? Oh, also using 7.0.44 x64.

 

[mstenholm]

Cool, got my wus uped. I notice I only had gpu wus in my cache, but in device manager I set it to no cpu wus and no other projects. Maybe that helps? Oh, also using 7.0.44 x64.

You could have a point there. My device manager allowed CPU work but my local setup file asked for only GPU work. That is now fixed. Let's see if it works.

 

[Norton]

All 3 of my rigs here are pushing in the wu's like crazy!!!!

My points shot up over 300k over FreeDC's last report and I've lots more work to get in


@Ion- assume about 10% for loss from cords/extensions, voltage fluctuations, wiring and breaker degradation, etc... don't run at the bleeding edge of 20 Amps or that breaker will likely pop again

 

[[Ion]]

All of my AMD card are now crunching again

I unfortunately still have the Xeon DP and GTX470s offline (they're in my dorm off, pending rewiring), but I have most of my production back online

300k Norton?
I wouldn't be surprised if that was enough to push you ahead of me again

And that's a good point about the power setup---I think if I budget a max of 1800-2000W/circuit it's probably reasonable

 

[Norton]

All of my AMD card are now crunching again

I unfortunately still have the Xeon DP and GTX470s offline (they're in my dorm off, pending rewiring), but I have most of my production back online

300k Norton?
I wouldn't be surprised if that was enough to push you ahead of me again

And that's a good point about the power setup---I think if I budget a max of 1800-2000W/circuit it's probably reasonable

About 90k away from 17 million atm and still have a whole bunch to report from the Xeon/7770x3 rig!

1800w should be safe- what is the incoming voltage? Remember that the power companies can supply lower than 120v and sometimes line voltage can drop as low as 100v or less from time to time.... which will send your amps even higher

 

[[Ion]]

About 90k away from 17 million atm and still have a whole bunch to report from the Xeon/7770x3 rig!

1800w should be safe- what is the incoming voltage? Remember that the power companies can supply lower than 120v and sometimes line voltage can drop as low as 100v or less from time to time.... which will send your amps even higher

Currently just about 80k away from 17m--and no more to report still (at least IIRC).
Sounds like I'll get to overtake you again soon enough p

I assume 120V, but I'm honestly not sure...

 

[Jstn7477]

If the circuit is well loaded, you could see up to around a 10V drop on the circuit. With the secondary 14K BTU/Hr window unit going in the office, the voltage on that circuit drops to around 113V from 120-122v measured on the other side of the room.

 

[ThE_MaD_ShOt]

Sorry if I suggested that you ran low numbers. I only look at RAC (19K) and the version number. I can see that you just only got it up and running and it can do so much more. However my point was that the older version apparently accepts if you ask for only GPU jobs whereas the new ones feed you the lot. If you want a newer version then go here http://boincstats.com/en/page/downloadBOINC

Sooner or later we all have to update to at least 7.0.44 if we want to run multiple GPU units....

Oh I understand And yes I just got that rig running within the last couple weeks. But the gpu has been crunching in another rig for awhile now for me. The top rig under my freedc stats list is the one it was in, that now has a gtx280 in it.

It usually puts out 40 to 45k ppd. I have no ideal what is going on with the points over the last few days. All my rigs are down on points.

 

[mstenholm]

I have no ideal what is going on with the points over the last few days. All my rigs are down on points.

The past 10 days we all experienced big fluctuations and for most a build up of WUs that was waiting for validation and hence a decline in daily production. Let's hope that this upgrade on the server side makes things a bit more stable.

On a side note my wireless connection was a bit slow so I still have around 2000 WUs spill over for tomorrow.

 

[stinger608]

My gpu only rig is running whatever version that is on Wcg's site. It also runs 5 wu's at a time. "toast" Here's a screen shot.
http://img.techpowerup.org/130112/screenshot.jpg

Hey man, what 7800 series of card are you running? Reason that I ask, is the 7870 that I have will run 8 WU's without issue. Hell, I can still surf the net and play solitaire.

 

[ThE_MaD_ShOt]

Hey man, what 7800 series of card are you running? Reason that I ask, is the 7870 that I have will run 8 WU's without issue. Hell, I can still surf the net and play solitaire.

It's a Visiontek 2gb 7850. It's a temporary card for that rig until I decide what I am going to run full time in it. I need to make that decision very soon and get the card or cards ordered. I may pick up 2 more 2gb 7850's since they seem to just sip power.

 

[HammerON]

Just got home from work and updated BOINC on both rigs. Reporting like crazy but it is going to take some time
This is the main rig. See where the page slider is? All the way to the top is "Uploading" WU's:

 

[[Ion]]

Everything is going again

I've moved the i7-2700k and i7-860 to my roommate's circuit to ease up the load a bit on mine. Hopefully this proves more stable