Starting a push on MW@home

[VirtualLarry]

Just installed/signed up for MW@home, on one of my refurbished Q9300 @ 3.0Ghz rigs, with a HD4850 card installed.

Wonder how well the points will rack up with one of those video cards?

That rig is running ONLY MW@home.

Edit: I busted out GPU-Z to monitor temps. They are running 83-85C. Is that too high? Fan speed is 34-40%.

Edit: Unlocked CCC Overdrive tab so I could manually adjust fan speed.
First I tried 66% fan speed, and it sounded like a blender. Temps went down from 84C to 70C.
I then adjusted it to 50% fan speed (4800-5000RPM), and then it was much quieter. Temps still around 70C.

Who knew that the fans on these cards could go to like 9000RPM?

 

[Sunny129]

Just installed/signed up for MW@home, on one of my refurbished Q9300 @ 3.0Ghz rigs, with a HD4850 card installed.

Wonder how well the points will rack up with one of those video cards?

That rig is running ONLY MW@home.

Welcome to the party!

i'm currently running MW@H on an HD 5870 2GB, and soon to join the party is an MSI 6950 Twin Frozr II w/ dual BIOS and unlocked shaders ...well it may actually shares its resources w/ SETI@Home too.

anyways, its tough to say what kind of PPD you'll see with that thing - when i look at people's MW@H machines on the MW@H forums, HD 47xx/48xx task run times are all over the place, ranging from as much as 500s. to as little as 160s. if i had to guess at the run time distribution, i'd say that the HD 4890's/4870's are in the 160-220s range, while the HD 4850's/4770's are closer to the 4-minute mark, and HD 4830's are near the 5-minute mark.

my HD 5870 averages ~225,000 PPD, completing tasks in ~2:04, or 124s. and that's running 2 tasks simultaneously, so theoretically that's ~1:02, or 62s per task. suppose your HD 4850 averages ~3:45, or 225s per task running one at a time. then some rough math gives us 62s / 225s * ~225,000 PPD = ~62,000 PPD. i believe Assimilator1 uses an HD 4850, though the last i talked to him he was having trouble w/ the new OpenCL-based MW Separation app. but he would have a better idea than i. either way, your HD 4850 will do very well. its one of the few projects where the AMD VLIW5 and VLIW4 architectures absolutely put nVidia's CUDA architecture to shame.

 

[VirtualLarry]

My GPU usage under GPU-Z is like 66-68%. Is that low?

 

[Sunny129]

My GPU usage under GPU-Z is like 66-68%. Is that low?

sounds like it. a single MW@H task puts my HD 5870 at 96-97% load. 2 simultaneous tasks pegs it at 99-100%. i'd start searching the MW@H forums for possible causes and solutions. i'll search around too and see what i can dig up.

btw i noticed that your GPU task run times seem to average about ~3:10, or 190s. that may change once you get the GPU usage thing figured out, but for now these run times should equate to just under 75,000 PPD.

 

[wayliff]

I also have an HD4850 1GB.

Here are my readings:
GPU Load: ~60% - 65%. (Sometimes hangs around high 30s and low 40s)
GPU Temps: 66C
Fan readings are not present\unavailable! I wonder why. I am using the latest GPU-Z version. I need to inspect my box but based on the temps, I think the fun is running.

As far as production...check out my page on free dc stats.
http://stats.free-dc.org/stats.php?page=user&proj=mil&name=184885

About a week ago, perhaps a bit more, I stopped my HD3850 because the room was getting too hot.

 

[Assimilator1]

Fan speed isn't shown in GPU-Z for me either, shows in CCC though.

Just installed/signed up for MW@home, on one of my refurbished Q9300 @ 3.0Ghz rigs, with a HD4850 card installed.

Edit: I busted out GPU-Z to monitor temps. They are running 83-85C. Is that too high? Fan speed is 34-40%.

Edit: Unlocked CCC Overdrive tab so I could manually adjust fan speed.
First I tried 66% fan speed, and it sounded like a blender. Temps went down from 84C to 70C.
I then adjusted it to 50% fan speed (4800-5000RPM), and then it was much quieter. Temps still around 70C.

Who knew that the fans on these cards could go to like 9000RPM?

I've got a HD4830, I set its fan @ 55%, close to yours .
GPU temp running MW atm is ~70C, 97-98% load.

I get really bad GUI lag with MW running so I only have it on when I'm not using it.

9k!? didn't know that, that's insane! lol

 

[Sunny129]

well i took delivery of the MSI 6950 TFII GPU and the new Corsair 500R computer case yesterday. hopefully i'll get to transfer it all into the new case tomorrow or this weekend. and on another positive note, if i play my cards right, i could be trading my HD 5870 + cash for another unlocked MSI 6950 TFII . now if only they'll at least overclock to HD 6970 speeds (880mhz) with little or no voltage increase...overclock or not, MW@H stands to benefit from my investment ().

 

[dajeepster]

i'd hope in, but it runs me $500 a month to run my systems..

 

[blckgrffn]

FWIW, as soon as I break 1M in Collatz I'll be putting my shiny new 4770 on SETI and MW full time.

Funny (not really?) story there - I have this 1045t machine in my sig. I had the 6570 in there (which is currently hiding in my HTPC, I need to get it back to work... anyway...) and it ran fine. It seemed like everytime I would ignore it for a few days though I would come back to find it rebooted a day or two into my attention hiatus.

Well, I put the 4770 in there and try to fold with it (curious about ppd) and I get driver failures, furmark bombs out, etc. Since this is an "OEM" new card from geeks.com, I immediately begin to dread the RMA process. I thought about it a bit, and resolved to try my new shiny backup PSU since PSU's have become my go-to fix it for bizarre/erratic PC behavior.

Pop that Antec Eco 620W in, no problems now, by golly. Crushing out Collatz WU's like that's its job

Moral of my too long story - check the PSU - and have a spare! Now I need a new spare

RIP Rosewill RP500-2, http://www.newegg.com/Product/Produc...82E16817182016 You served me well despite your less than desirable heritage.

 

[Sunny129]

RIP Rosewill RP500-2, http://www.newegg.com/Product/Produc...82E16817182016 You served me well despite your less than desirable heritage.

glad to hear you solved the issues w/ a new PSU. its funny how cheap/inexpensive PSUs can serve end users just fine for normal, non-stressful PC activities, and you'd never know their flaws if you didn't significantly load the CPU or GPU occasionally. needless to say, the Antec Neo Eco 620C (SeaSonic-built) is a rock solid PSU.

 

[blckgrffn]

glad to hear you solved the issues w/ a new PSU. its funny how cheap/inexpensive PSUs can serve end users just fine for normal, non-stressful PC activities, and you'd never know their flaws if you didn't significantly load the CPU or GPU occasionally. needless to say, the Antec Neo Eco 620C (SeaSonic-built) is a rock solid PSU.

Yep.

I am kicking myself for passing on the 400W Eco that was a shell shocker in the last couple weeks... $30 shipped. That would have worked really well for this PC.

 

[Sunny129]

well this weekend was a let-down in terms of parts upgrades. the MSI R6950 TFII turned out to run far hotter than i ever expected, and required fairly high fan speeds to control the temps, thus making it quite a loud card. so that card will be back up for sale in the next few days...

also, my new Corsair Carbide 500R case, which i already RMA'd once due to a broken foot and front bezel, is having front panel problems...specifically, the following things don't work: the reset switch, the HDD LED, the fan LED switch, and the fan controller. so i'll have to RMA this for the 2nd time now. fortunately i got it through amazon, who cross ships, pays for my return shipping, and gives me plenty of time to return the damaged/broken part. that makes things easier, since i can keep the computer running in the semi-working case until the 3rd replacement arrives.

 

[blckgrffn]

well this weekend was a let-down in terms of parts upgrades. the MSI R6950 TFII turned out to run far hotter than i ever expected, and required fairly high fan speeds to control the temps, thus making it quite a loud card. so that card will be back up for sale in the next few days...

also, my new Corsair Carbide 500R case, which i already RMA'd once due to a broken foot and front bezel, is having front panel problems...specifically, the following things don't work: the reset switch, the HDD LED, the fan LED switch, and the fan controller. so i'll have to RMA this for the 2nd time now. fortunately i got it through amazon, who cross ships, pays for my return shipping, and gives me plenty of time to return the damaged/broken part. that makes things easier, since i can keep the computer running in the semi-working case until the 3rd replacement arrives.

Jeeze Sorry to hear about all that.

Does that mean I should open up the box on my Carbide 400 and check it out? I was just assuming that buying a relatively high end case (for me, my typical budget is ~ $40-$60) would mean it was all good. Boo for bad quality control!!!

With any luck that 6950 should be fairly easy to unload...

 

[Sunny129]

Jeeze Sorry to hear about all that.

Does that mean I should open up the box on my Carbide 400 and check it out? I was just assuming that buying a relatively high end case (for me, my typical budget is ~ $40-$60) would mean it was all good. Boo for bad quality control!!!

With any luck that 6950 should be fairly easy to unload...

i don't know man...i'm no Corsair whore by any means, but i've used various Corsair products over the years, from memory to PSU's, and was never disappointed...until now. though i will say this is my first experience with a Corsair case. if i were you, i'd check it. your front panel is identical to mine, minus the fan speed controller. it really is a shame, b/c i love everything else about the case. we'll see what happens when i get my 3rd replacement. if i have anymore issues after that, its going back for a refund.

 

[wayliff]

My GPU usage under GPU-Z is like 66-68%. Is that low?

I have done some reading regarding this question and what I've found is that many recommend to stick to the 11.9 drivers unless it is absolutely necessary to use anything newer.

I've installed 11.9 on one machine but I am yet to monitor it while it runs.
I did this on a machine that runs the HD3850. Will monitor sometime and post back.

The HD4850 machine somehow became unstable with 11.9 and had trouble even running Windows so I had to install the latest and it is fine now.

 

[Assimilator1]

I was about to roll back to 11.9, until I saw your last sentence .


On a different note, I blew out a load of dust out of my PC earlier (looked like the Red Arrows had flown by! ), CPU's now running 6-7C cooler! So MW will be able to run more now .

 

[wayliff]

So I did a bit of playing with my machines to see why GPU load is not near 100% most of the time.

I have a hexa core with the HD4850 and a quad core with the HD3850.
Both machines are fully crunching WCG with the CPUs and MW with the GPUs.

Once I changed the quad core to use three CPU cores and the hexa core to use five CPU cores for WCG...both GPUs were maintaining 98-99% loads.
I simply tried the 75% and 84% respectively, I did not try to bring the CPU use closer to 100% to see what would happen.

I guess the reason is that MW requires at least one CPU, or a portion of it, to be feeding it constantly. However for what I read on the BOINC tasks lists MW should only consume 0.05 of one CPU for 'something'.

Thoughts?

 

[Sunny129]

So I did a bit of playing with my machines to see why GPU load is not near 100% most of the time.

I have a hexa core with the HD4850 and a quad core with the HD3850.
Both machines are fully crunching WCG with the CPUs and MW with the GPUs.

Once I changed the quad core to use three CPU cores and the hexa core to use five CPU cores for WCG...both GPUs were maintaining 98-99% loads.
I simply tried the 75% and 84% respectively, I did not try to bring the CPU use closer to 100% to see what would happen.

I guess the reason is that MW requires at least one CPU, or a portion of it, to be feeding it constantly. However for what I read on the BOINC tasks lists MW should only consume 0.05 of one CPU for 'something'.

Thoughts?

the "0.05 CPUs + x.xx ATI GPUs" you see in the status column of the BOINC manager is really only a CPU consumption estimate for that task. in reality, a MW@H task takes up a full core for a few seconds both at the very beginning and the very end of the task. for the few minutes between the beginning and the end of the task, CPU consumption should be zero. so the value of 0.05 CPUs given by BOINC is only accurate if you average the CPU usage over the life of the task itself.

it sounds to me like something else if affecting your GPU utilization. i say this b/c my MW@H machine has an X6 1090T CPU that crunches Einstein@Home on all 6 cores AND crunches 2 MW@H tasks simultaneously on the GPU...and my GPU utilization is always pegged at 99%. then again, perhaps it has something to do with the type of project running on the CPU. i'm running Einstein on the CPU, whereas you're running WCG.

 

[wayliff]

I am suspecting it might be the type of project.
These two boxes are doing nothing else other than crunch most of the time.

I may have to give that a try and see how another project behaves. The only other project I am subscribed to right now is Correlizer.

I will test some time, no rush, and report back.

 

[blckgrffn]

On my Collatz (GPU) +WCG (CPU) box, they seem to play very nicely. The 1045t is running flat out and so is the 4770.

Six CPU WU's and one GPU WU. I can't really tell how big of a negative effect it is having on WCG WU completion times, but it seems minimal. At least its fully loaded, that was the goal

I only bring it up to point out a functional pairing of GPU and CPU projects.

Also, I am running the latest Catalyst drivers in which CPU overhead for GPU compute is supposedly significantly reduced...

 

[Sunny129]

i should have specified my driver version as well - i'm using Catalyst 12.1 on a 5870 right now...