Just joined the project, with one quad, from DNETC.

Looking at the top computer lists I see PCs with ATI HD5850 GPUs crunching these huge WUs in about 1,900 seconds (or less).

My HD5850 has crunched 6 WUs now and all of them are crunched in the time range 4,150 seconds to 4,420 seconds.

My speed seems to be more than twice as long as equivalent PCs. Any thoughts as to why?

What might I be able to do to get to the normal speeds, if anything?

Are there any GPU optimised clients, or is the normal download process OK?

Try to up the Processor Usage priority from 4 to 9 in "dnetc-1.00.ini" (in the Moo folder)
CPU hogging projects starve Moo and it is not getting CPU power to do it's GPU thing
Also look for screen savers etc
Look at my PC "Floor" with an OC'd 5850
It's doing about 300k / day
Hope it helps

Thanks Copycat.

I've upped the processor usage from 4 to 9 as suggested, now we will see how it goes.

What is the maximum setting recommended for this figure compatible withnormal use of that specific PC?

I have no screen savers running, but I do run the following on CPUs set to 100% -


FreeHAL
PrimeGrid
WUProp@Home

Are these helping throttle Moo!?

That small change knocked 2,000 seconds off the crunch times per WU. Just a bit more adjustment will see my GPU in line with others.

Adjustment still at 9 but crunch times have suddenly gone back up to over 4,100.

I wonder why?

It could be down to the CPUs running 4 x PrimeGrid @ 100%. I've backed off to 90% CPU and the PC is now running 3 x PrimeGrid. Let's see if this reops the crunch time?

My HD5850 is OC'd to GPU 855MHz / Mem 1180MHz
Another thing I forgot to mention is the ATI/AMD's CAL version
I found with Win 32 bit anything over CAL 11.5 runs at a very high CPU (25%) where it should run at only 2%
I use CAL 11.4 on all my PCs
This happens only here at Moo!
DNETC is OK with the later CAL releases

Keep in mind WU size vary - look at the last 5 digets of the WU name

...
I found with Win 32 bit anything over CAL 11.5 runs at a very high CPU (25%) where it should run at only 2%
...

AMD Catalyst 11.9 drivers have a fix for the high cpu usage.
I'm using the 11.9RCx myself on bunch of rigs since a few weeks and everything works fine(cpu load == 0%)

Download -> http://downloads.guru3d.com/download.php?det=2797

---

Join BOINC United now!

My HD5850 is OC'd to GPU 855MHz / Mem 1180MHz
Another thing I forgot to mention is the ATI/AMD's CAL version
I found with Win 32 bit anything over CAL 11.5 runs at a very high CPU (25%) where it should run at only 2%
I use CAL 11.4 on all my PCs
This happens only here at Moo!
DNETC is OK with the later CAL releases

Keep in mind WU size vary - look at the last 5 digets of the WU name

If you are running only Moo! on that HD5850 then you should consider downclocking the mem speed as Moo! doesn't depend on a high memory bandwidth. Downclocking mem speed to e.g. 500 or even 300 MHz will keep your card much cooler without impact on your RAC.

After running for a long time with the HD 5850 crunch times in the region of 2,150 seconds to 2,350 seconds, I am suddenly punching a bunch of WUs taking 4,350 seconds.

I looked at the dnetc-1.ini file and see processor usage is still set to 9 (no change).

I aborted all the cache and re-loaded with new work, no change and RAC taking a hit.

Processor usage still at 75% (local preferences - no change).

I am now running down the cache.

Looks like I may need to let the work crunch and rectify itself.

NOTE: Crunching with a additional 2,000 seconds per WU has not increased the credit per individual WU - just slowed the number of WUs done per day to less than half.

This dows not affect the other quad and its HD58xx GPUs

The HD5850 is still taking 4,000+ seconds to crunch the huge WUs.

I think a project reset is now in order, and setting the processor usage back to 9 in the dnetc-i.ini file

Don't know if the project reset, BOINC Manager close down and changing processor usage from 4 to 9 in the dnetc-1.ini file was the cause. But, the 6 WUs crunched since showed 5 done in 2,155 seconds and 1 in 4,000+. The WU currently being crunched looks to be finishing in 2,250 seconds ... so a return to normal I hope?

I'm having a bunch of 4000s units as well ....

---

I iz also got icons!

Chris

My project reset, then BM shutdown to change the dnetc-1,ini file from 4 to 9 and restart BM seems to have worked. If you have yet to do that then it seemed to solve the issue.

What's the advantage of changing the CPU priority from 4 to 9.
Mine is still set to 4 and other than using more CPU than DNETC (between 20 and 40 sec compared to just 1 to 5 sec), the system is running OK.

Thanks
Conan

---

What's the advantage of changing the CPU priority from 4 to 9.
Mine is still set to 4 and other than using more CPU than DNETC (between 20 and 40 sec compared to just 1 to 5 sec), the system is running OK.

Thanks
Conan

The HD5850s used by Chris and myself, certainly myself, seem to encounter WUs requiring 4,200 seconds to crunch. Normally they take between 2,100 and 2,400 seconds, which means a tremendous difference in the number of WUs completed every 24 hours.

The 4K WUs seem to happen as a norm with the CPU priority at 4, and run the faster times when CPU priority is to 9.

I have no problem with this on the other GPU, but have upped the priority as a matter of habit.

Been watching this one with interest. I have a 5850 running in a Phenom II quad 3.2Ghz, 8Gb memory, operating at 775/500 on Win7 Home Premium - I have found no difference whether its at 4 or 9. I usually keep it at the default 4 now.

No idea why I dont see a difference and you have, its a strange one. There must be other factors coming into play here somewhere ... no idea what they might be though.

I am sure the upping of crunch times from 2.25K to 4.25K is not related to CPU availability.

Although the quad running the HD5850 is on an XP pro 32bit OS, and I am crunching PrimeGrid on the CPUs (as well ad FreeHAL and WUProp non-intensive) the CPU usage is limited to 75% and Speed fan shows the CPU usage never exceeding 95%. Also, the quads have 4GB of RAM installed, with the OS seeing 3.5G, so no hard disc thrashing.

It seems I occasionally hit a batch of 4+K second WUs and this needs a project reset and the ini file to 9 to cure. It then runs fine, usually for weeks, until the next batch.

Cause, at this time, unknown.

Been watching this one with interest. I have a 5850 running in a Phenom II quad 3.2Ghz, 8Gb memory, operating at 775/500 on Win7 Home Premium - I have found no difference whether its at 4 or 9. I usually keep it at the default 4 now.

No idea why I dont see a difference and you have, its a strange one. There must be other factors coming into play here somewhere ... no idea what they might be though.

Zydor

I've noticed my HD5970 is taking about 25% to 30% longer to crunch WUs than earlier today. Perhaps there is a bigger difference in WU crunch times than we realise (actual differenced between WU crunching needs/demands).

Do you notice this?

Mine has been consistent. The 2x5970s are doing a WU in 460 secs @775/300 with the 1090T running @3.7Ghz. The ini setting remains at 4.

Is there "overloading" of the cores? I keep two cores running CPU WUs (from PRPNet in two DOS boxes). The other four Cores I leave alone for the 5970s to play with. Essentially I am leaving one core free for each GPU.

I find if I try to use 3 or more Cores for CPU WUs, the times of the GPU WUs goes off the rails, varying a lot as they compete for time on the Cores. By the time I get to using 5 or 6 cores .... the world goes mad and likely falls over, hardly surprising in the latter case, GPUs do need some attention.

So I stay at 2 CPU WUs running, and leave one core free per GPU.

Mine has been consistent. The 2x5970s are doing a WU in 460 secs @775/300 with the 1090T running @3.7Ghz. The ini setting remains at 4.

Is there "overloading" of the cores? I keep two cores running CPU WUs (from PRPNet in two DOS boxes). The other four Cores I leave alone for the 5970s to play with. Essentially I am leaving one core free for each GPU.

I find if I try to use 3 or more Cores for CPU WUs, the times of the GPU WUs goes off the rails, varying a lot as they compete for time on the Cores. By the time I get to using 5 or 6 cores .... the world goes mad and likely falls over, hardly surprising in the latter case, GPUs do need some attention.

So I stay at 2 CPU WUs running, and leave one core free per GPU.

Mine don't seem to be so sensitive, Zydor. The quads are Intel QX6700 and QX6950s (65nm and 45nm fabs).

I run the 3 CPUs for PrimeGrid (no GPU),this leaves 1 whole core for both the HD5850 and on the other quad the HD5970.

I also run two CPU non-intensive projects (4 x FreeHAL and 1 x WUProp@Home). Speed Fan shows the maximum use of the CPUs runs between 89.6% and 96.5%, leaving some of it free.

My HD5850 is now running OK at 2,155 to 2,350 seconds (normal) and the HD5970 has crunched a bunch of WUs at 1,250 seconds (25% above normal), but has now done a whole bunch (>60) @ 1,110 seconds.

This seems to lie OK with Zydor's figures for2 x HD5970s, as the project client uses all GPUs for the single WU.

NOTE: Zydor's higher CPU per GPU dedication could account for a slightly lower WU crunch time. I calculate, on a straight line basis, my WU crunch time with Zydor's set up would be 560 seconds not 460 seconds.

When my RAC for the HD5970 rig has settled at it's max for a week or two, I will NNT PrimeGrid and see if there is an impact on the WU crunch time (it falls) as there will be more dedicated CPU per GPU. The other way would be to restrict the local machine to using 2 cores for PrimeGrid - same effect.