Welcome to MilkyWay@home

Posts by Ed.T

1) Message boards : News : ATI application updated to 0.60 (Message 47802)
Posted 13 Apr 2011 by Ed.T
Post:
v.62 on my two 4850s still needed more than one core (on my two processor dual core Operton system) to reach 99% GPU utilization.

Ed.T, your results on that machine show 10 seconds or less CPU/WU with v.62.


I've gotten the notion that if the CPU isn't available when GPU needs it that GPU utilization goes down. CPU time for the job wouldn't increase but PPD would still go down.

Perhaps the combination of BOINC, task manager and MSI Afterburner (plus the WCG work units running) pushed the need for "some of one core" not running a WU to "some of 2 cores" not running a work unit... IIRC, there's a tweak for raising the priority of the CPU side of the app in the works. I'm going to wait and see if that fixes it.

Some folks are saying everything is working fine. I'm saying not for everyone; we still need the additional releases.

- Ed.
2) Message boards : News : ATI application updated to 0.60 (Message 47776)
Posted 13 Apr 2011 by Ed.T
Post:
My two (underclocked) 4850s scored 107K yesterday, down a little bit over 20% compared to before the current mess started. My client has been automatically downloading new apps; last I looked (yesterday) it was v.62. Last I looked (yesterday,) v.62 on my two 4850s still needed more than one core (on my two processor dual core Operton system) to reach 99% GPU utilization. At least now MW is yielding to my preferred non-MW CPU work units.
3) Message boards : News : increased WU limits (Message 47555)
Posted 11 Apr 2011 by Ed.T
Post:
No need to rush this. If the servers are running smoothly then we don't need it. If they aren't then it will only make matters worse.
4) Message boards : News : ATI application updated again (Message 47551)
Posted 11 Apr 2011 by Ed.T
Post:
This is embarassing. I think the project would be better off if they just restored last week's version and our administrators backed off, took a few deep breaths and did some real software engineering and project planning.



5) Message boards : News : reducing max_wus_in_progress (Message 47178)
Posted 9 Apr 2011 by Ed.T
Post:
I don't know what a "feeder" is other than that it seems to be keeping my machines from reporting...

4/8/2011 9:19:27 PM Milkyway@home update requested by user
4/8/2011 9:19:28 PM Milkyway@home Sending scheduler request: Requested by user.
4/8/2011 9:19:28 PM Milkyway@home Reporting 106 completed tasks, not requesting new tasks
4/8/2011 9:19:36 PM Milkyway@home Scheduler request completed
4/8/2011 9:19:36 PM Milkyway@home Message from server: Server error: feeder not running
6) Message boards : News : Server updated (Message 47080)
Posted 6 Apr 2011 by Ed.T
Post:
Are you seeing credit on the MW site? My pending-validation queue is going down but I'm still not seeing points...
7) Message boards : Number crunching : killo-o-watt readings ... I thought they'd be higher.! (Message 46961)
Posted 4 Apr 2011 by Ed.T
Post:
That's nice!

My rig is a couple generations behind both in computational power and power usage.

I have to admit that my main goal is to compute for WCG. Doing that alone, the dual Opteron is competitive with the Phenom II X3 on my desktop (and the server board ready-to-go with CPUs memory cost about the same as I spent on the heatsink for my Phenom chip!) GPUs were and afterthought but I did have a 4850 laying around after I upgraded my desktop with an NVIDA board for 3D gaming.

I fist tried PRIMEGRID but its GPU WUs still used about 25% of a CPU and I didn't want to take that much from WCG. I settled in on MW@H because MW WUs steal very little CPU time. (Though on this system MW WUs seem to take closer to .1 CPU rather than .05 CPU as BOINC manager suggests.)

I'm hoping that Intel Sandy Bridge or capabilities like it will come down in price (significantly) in a year or so and will look at an upgrade then. Maybe by then those 6970s will be $50 on ebay like the 4850s are now and I can put a couple of them in a new system at that time too. ;)

At any rate, I am seeing the expected performance gain from adding the second card. Give or take the servers actually passing out work and validating results, the system has gone from about 75k ppd to 135K ppd. (Not double because I've under-clocked the cards a little to save power and heat.) Combined, the two 4850s average a WU every 2 minutes and cycle (or drain) a 24 job queue in 50 minute to an hour depending.

The system been running for a few days and seems stable so far. I do notice that there's a little droop on the 12v line when the GPUs are going full tilt. 11.91 volts idle drops to 11.72 volts. That is still well above the 11.4 volt minimum in the AT spec. Fine, but I'll continue to watch that voltage over time. I think that if I ever see that going down, that would be the sign that the PSU is failing. No sign of that yet.

Thanks again,
- Ed.T
8) Message boards : Number crunching : killo-o-watt readings ... I thought they'd be higher.! (Message 46851)
Posted 30 Mar 2011 by Ed.T
Post:
Zy,

Thanks for you and the others making a new guy feel welcome. I realize that with two million plus users that there isn't much new that will show up in the forums so I doubly appreciate the comments and feedback.

I've been working on cooling along the way. My two CPUs are running at 54C and the 4850s at 65C and 75C. (The single slot board doesn't cool as well as the dual slot board does.) Those temps are with the side of the case open. Even so, I added air ducts from the outside of the case to the CPU fans (good for a 5C drop) and a point of use fan blowing outside air onto the GPU fans (good for about a 8C drop.) A heat shield under the disk drive lets the PSU draw cool air in around it while hot air from the front CPU blowing near it goes elsewhere. (Before that, the drive was "uncomfortable" to keep a finger on it, not it's merely warm to the touch.) In addition, there's cardboard and duct tape to guide what air flow there is in the (open) case to and from where I guessed it would do the most good.

As is, it's not pretty sight but I'm okay with that since I plan to put it in the basement. Come summer, it's the only place that stays cool.

Thanks again,
-Ed.
9) Message boards : Number crunching : killo-o-watt readings ... I thought they'd be higher.! (Message 46849)
Posted 30 Mar 2011 by Ed.T
Post:
Hi Chris,

Is that measured with a kilowatt meter? The few reviews I've read show about a 250 watt increase in watts "under load" for the 6970 but also I've seen that MW (on my system at least, YMMV) doesn't seem to use all the power eating features available on the GPU. Worse, it seems that even something trivial happening on the screen seems to block GPU resources used by MW. I wrote about the "mystify" screen saver; since then I noticed that the MSI Afterburner partially blocking the BOINC manager drops GPU utilization (and power used) by about 15% as well. (So my observation is that for the most GPU points, I won't assume that even something trivial isn't slowing down the GPU.)
10) Message boards : Number crunching : killo-o-watt readings ... I thought they'd be higher.! (Message 46846)
Posted 30 Mar 2011 by Ed.T
Post:
Sounds like you are enjoying yourself


Yes I am!

I installed the second 4850 came today. I had to hunt down a single width card because my old Tyan server board only has one slot between the two pci express connectors. I made a dummy plug, extended the desktop and clocked down the GPUs before starting MW. The screen went wacky and the system crashed.

At first it seemed like the like Antec couldn't handle the two boards so I took the old one out and left the new one in. The system still crashed; turns out that the new (Diamond) 4850 won't run with the memory clock turned down much. Seeing that, I put the old 4850 back in and I have had them both crunching for a few hours now.

Two 4850s powered by a 430 watt PSU... No complaints here!

I clock both cards at 500/700 to save power and run cooler. The watt meter is reading 395 watts into the power supply. By math, that should be about 335 watts out of the PSU's rated 430 watts, less than 80% utilization.

Well, time will tell if this is "robust" but in the short term I'm hopeful for a little kick upwards in the standing! :)

- Ed.T




11) Message boards : News : bypassing server set cache limits (Message 46843)
Posted 30 Mar 2011 by Ed.T
Post:
Really!? Once it gets to the point where there's no work to fetch then what's the difference between my system being out of work and your system being out of work?

If fact, I bet your (newer) system idles more efficiently than mine so from an energy-use-on-this-planet point of view it seems more efficient for your system to sit idle than mine.

Just thinking...

- Ed.
12) Message boards : Number crunching : killo-o-watt readings ... I thought they'd be higher.! (Message 46817)
Posted 29 Mar 2011 by Ed.T
Post:
Well, I think I have RAC-fever. (I adopt that from my astronomy hobby where I suffer from "aperture fever.") Even if I go over-budget putting a second 4850 in the system, I'm like to up the budget for a while just to see the points pour out of it. I want to be in this of the long run though and am sure that budget will cure my "fever" before too long.

The Antec is running like a champ on that (a Tyan) MB. The Tyan is my favorite MB now because it has two pci express slots. Strange, though, the Antec wouldn't even start the fans on my other (Supermicro) MB. Google-ing, I did find that Antec has trouble with Supermicro boards. I don't know whether that's something do with the power-good logic; I do notice the PSUs that work have a bit more +3v and +5v to give than the Antec offers.

I ended up with the OEM PSU on the Supermicro. WOW IS IT LOUD! It's big enough that it should run two 4850s on the Tyan board if the Antec won't. I can't find an efficiency spec on it but now that I have the Tyan system fairly well categorized, I should be able to plug it in and infer what it is. Sounds like something worth knowing; I think I'll do that...

- Ed.T
13) Message boards : Number crunching : Validate errors (Message 46793)
Posted 28 Mar 2011 by Ed.T
Post:
So, I had a few 10_3 errors, now I've a quadrillion "Completed, validation inconclusive" as punishment. So the system, it seems, sends these to other folks flagged and scores their results as "Completed, validation inconclusive" too and then sends the WU off to another and so and so on ... this one is on its 5th machine -

http://milkyway.cs.rpi.edu/milkyway/workunit.php?wuid=260891871

... will it ever end?

- Ed.T
14) Message boards : Number crunching : killo-o-watt readings ... I thought they'd be higher.! (Message 46786)
Posted 28 Mar 2011 by Ed.T
Post:
Got a new unit, it behaves the same way on my desktop system; I've swapped enough pieces to say that there's something wacky about the "hec-X" PSU... the weirdness followed it when I put in in different systems.

I've taken readings over several days on a different system that seems to report accurately on the watt-meter: a dual dual-core Opteron 280 "T server board has 4 512MB memory sticks and a IDE disk running XP professional. System is powered by a Antec Earthwatts 430 (that claims 85% efficiency.) I run WCG on the CPUs and get give or take 1200 BOINC points a day.

I don't have any CPU power saving enabled so this system runs about 185 Watts constantly whether or not it's running apps. At .13 USD for a kilowatt-hour, that's about $18 USD a month.

I originally got this system with slower Opteron 265s in it, that ran at 165 watts. I have another similar system with 280 "A" chips, it runs about 205 watts.

I put a ATI 4850 in it t run MW WUs. The clocks are set to 600/420 with MSI Afterburner. I under-clock the core slightly and the memory a lot; like others, I don't see a performance difference with a slower memory clock.

Idle, the system power draw rises to 245 watts; that's sixty watts for just plugging in the 4850. I understand that other boards have better idle numbers but, really, who what's their cruncher to be idle?

With the lower clocks, the system crunches just under 300 watts. (With the stock clocks, the system draws a bit over 310 watts.)

The 300 watts is input to the power supply. If I accept the published 85% efficiency of the PSU the output of the PSU is about 255 watts... so I'm thinking maybe if I take two memory sims out (about 15 watts) I can squeeze another 4850 into the system and sill only be at 85% of the PSUs 430 watt continuous rating...

Mystified:
I was playing with screen savers and saw that the power usage dropped 15 watts using the "Mystify" screen saver vs. the WCG screen saver. WOW! Imagine that times all the machines doing GPU processing.... not to be however. Looking at the GPU usage chart, usage drops to 85% and WU run-times increase accordingly. Strange, the few other screen savers I tested (including the similar bezier curves) still allow the GPU to run at full usage. So, the "Mystify" screen saver is not to be used; too bad it's been my favorite since it was first offered way back when.

So, I'm way over my compute budget. This one system spends my $30/month target and the other about 200 watts ($20/mo). Oh well, I'll likely keep it up until I see a few electric bills that hurt then decide what to do.

- Ed.
15) Message boards : Number crunching : killo-o-watt readings ... I thought they'd be higher.! (Message 46349)
Posted 25 Feb 2011 by Ed.T
Post:
Probably your WUs are suspended


No, I cycle them on and off and I do see a small change in readings when I do. I wonder if there's something wacky about the PSU that throws off the reading. It doesn't seem to be just "stitching supplies" because the reading seams reasonable on the other computer... Anyway, back it goes for replacement and I'll write again with readings from another unit.

-Ed.[/i]
16) Message boards : Number crunching : killo-o-watt readings ... I thought they'd be higher.! (Message 46280)
Posted 18 Feb 2011 by Ed.T
Post:
Hi all,

I'm new to crunching and want to stay within budget. I am surprised at some of the numbers I'm getting. I don't want to set a crunching pace that I can't sustain. I can do the math but the usage readings I'm getting from my meter leaves me skepical to say the least...

System: Phenom II X4 at 2.1GHz w/ Nvidia 460. The meter measures 80 watts idle, 86 watts running 3x cpu WUs and 96 watts when a gpu WU is added tp the 460... what!? ??? I wouldn't guess that and can't believe it.

I get reasonable readings from the meter checking plug-in appliances around the house, just not on the computer.

By comparison, my other system a P4 with a ATI 4850 idles at 100 watts, run 2 cpu procs at 140 watts and jumps to 200 watts when a gpu task is added. Even that seems low to me but much closer to what I expected.

My swag was 300 watts from the P4/4850 and 400 watts from the Phenom/460... what do you think about that?

BTW, my budget is 300 watts continuous so I don't expect to be able to maintain my current crunch rate!

TIA, - Ed.T
17) Message boards : Number crunching : Panic Thread (Message 46168)
Posted 11 Feb 2011 by Ed.T
Post:
Einstein is still up and they have GPU work units ready for you


About 35 years ago, we had a "gravity wave antenna" in the lab next to the optics shop. It seems that the key to the optics shop opened that door as well and I got to go in a get a good look at it. IIRC, it looked like a shiny can a couple of times bigger than a 55 gallon drum. I never saw anyone working in the lab; searching for gravity waves was largely unattended work. Every time we dropped something, we'd laugh (or cry depending on what we dropped) and say "there goes a gravity wave!" Gravity waves didn't get a lot of respect in the mid-70's.

A year or two later, whoever was running the project must of got a grant because a new antenna came, somewhat bigger, that looked more like a one of those propane tanks (with rounded instead of flat ends.) We were jealous, of course, because getting grant money was what college was all about and, inconceivable, someone had picked "gravity waves" again... We'd stomp on the floor as we walked down the hall but apparently along with the grant came enough money to change the locks on the door so I never did get too close a look at it. And, IIRC, not to long after that the lab moved somewhere else, I don't know where.

Still, I just can't bring myself to search for gravity waves...

BTW: the other money pit at school at the time was "laser fusion" and I see that's getting renewed interest. So I guess the saying should be "if ou can't fool all the people all the time, just wait for the next generation to come along." :)

Cynic for Science,
- Ed.T




©2021 Astroinformatics Group