My computer has a bit of laggyness when running the project on my 5870 card..I just tried changing the "Resource share" to 80% to see if that might help as it was set on 100% but it hasn't take effect yet. What is Resource Share for exactly?

And what is causing this slowdown?

Edit: I just saw under my tasks that I see this

Completed, validation inconclusive

Is that normal/good?

First, the easy part: As far as your tasks go, that is a completely normal response. When you wingman returns their task you will get the credit for it.

Setting the resource share only affects the CPU so does not have the same effect on a GPU task.
There are other threads you can look at to see how to reduce the priority ranking for BOINC as far as the GPU goes. Otherwise the easiest is to right click on the BOINC icon and snooze the GPU part while you need the screen, then let things run when you are not abut.

---

sorry to bring this thread back to life, but I also noticed the visual sluggishness when crunching mw and collatz; however and the reason I write, I noticed no discernible visual impact crunching dnetc. interestingly, dnetc causes hottest gpu temps, mw 2nd hottest, and collatz coolest--at least for my setup. that also corresponds to the credit-given hierarchy. is there a correlation that at least partially explains the credit differences across the three ati projects?


would someone please explain how the projects work the gpu such that temps vary as much as 10c.


my guess, this has been answered again and again--have yet to find it, so please provide a link(s).


thanks.

Now i am a bit baffled by this post.
You say a 5870 and it responce is sluggish ?!?!
I am running also a 5870 and believe it or not while running any project still can do games/work/surf except when i set my games to ultra high settings.
For normal operation i have in no way any experience of a sluggish screen.
Hell even on my other old ati 4770 i can surf, wordprocess and do whatever and not even see this. Ofcourse when i start gaming on that machine i see a result.
Now to find the cause of the sluggish reaction there must be ofcourse a reason.
One reason could be that you run the card in a slower system.
For instance i was running the 5870 on a 2,4 Ghz Q6600 and when i ran 3dmark06 benchmark only got 14835 points, so i overclocked the cpu to 3,2 Ghz and the immediate result was that i got 17821 score in the same benchmark

But i think you must check for how many pci-e slots your system has if you have more then 1 slot, which slot is your video card in ?.
And second does your system has other addin cards in the normal pci slots.
If so some of these other card is maybe fighting for the same irq in your system this can be sometimes solved by changing the slot where the videocard is put in or the pci card ( mostly sound/network cards)
If you want to check if an irq is conflicting you can check for instance with hwinfo/everest/speccy or other system info programs.

---

Its new, its relative fast... my new bicycle

Now i am a bit baffled by this post.
You say a 5870 and it responce is sluggish ?!?!
I am running also a 5870 and believe it or not while running any project still can do games/work/surf except when i set my games to ultra high settings.
For normal operation i have in no way any experience of a sluggish screen.
Hell even on my other old ati 4770 i can surf, wordprocess and do whatever and not even see this. Ofcourse when i start gaming on that machine i see a result.
Now to find the cause of the sluggish reaction there must be ofcourse a reason.
One reason could be that you run the card in a slower system.
For instance i was running the 5870 on a 2,4 Ghz Q6600 and when i ran 3dmark06 benchmark only got 14835 points, so i overclocked the cpu to 3,2 Ghz and the immediate result was that i got 17821 score in the same benchmark

But i think you must check for how many pci-e slots your system has if you have more then 1 slot, which slot is your video card in ?.
And second does your system has other addin cards in the normal pci slots.
If so some of these other card is maybe fighting for the same irq in your system this can be sometimes solved by changing the slot where the videocard is put in or the pci card ( mostly sound/network cards)
If you want to check if an irq is conflicting you can check for instance with hwinfo/everest/speccy or other system info programs.

I am not sure if you are addressing the initial post, but to respond for my setup:

I have two 5870s (not crossfired while crunching, kept under 60c, vrms under 85c) sitting on an evga x58 3way class (760 w/up2date bios) with 6gb corsair ram and an i7. no addin cards here, but that matters not. I see no screen sluggishness with dnetc. I have a suspicion this is not about hardware nor the uptodate 64b os, which narrows it down a bit. as noted, I see distinct temp differences between the projects which suggests they tax different aspects of the gpu. the temp differences are known, but I have yet to discover why.

the oddity and the reason I added to the thread, dnetc runs the gpus hottest yet no lag. the lag is not show stopping, but discernible.

as for running apps like games, I think boinc backs off at some point of resource contention. doubtful boinc is looking to back off when moving a window around the screens and that is when it most evident on this end.

when I have WUs there is a screen lag on my machine with mw and collatz, not dnetc.

I had a similar problem with 2 5870's. My problem was that I run a single
monitor and I had disabled the monitor of the second card in CCC. I connected
a second monitor, enabled it after reinstalling CCC, allowed CCC to identify it and all the lag disappeared. And the time per wu decreased by about 40%. I guess this is where the dummy plug comes in handy. The second monitor I have switched off but keep it plugged in the card. My cards are also in slots 1 and 3 to keep them as far apart as possible for overheating problems.
I hope this helps :-)

I had a similar problem with 2 5870's. My problem was that I run a single
monitor and I had disabled the monitor of the second card in CCC. I connected
a second monitor, enabled it after reinstalling CCC, allowed CCC to identify it and all the lag disappeared. And the time per wu decreased by about 40%. I guess this is where the dummy plug comes in handy. The second monitor I have switched off but keep it plugged in the card. My cards are also in slots 1 and 3 to keep them as far apart as possible for overheating problems.
I hope this helps :-)

I cannot test at this time; no wus from all three projects. wth? when crunching, I see no crunching time changes running 2monitors. 156-159s for 4 MW WUs--crunched concurrently. I post the times just in case you report running 40% better. 40% is significant, highly, so you might have had another issue. I am not experiencing that kind of lag, but will get your idea a try.

as for overheating, yes, aggravating. 5870s are not, ots, ready for crossfire even in a haf box. to compensate they become unacceptably noisy for my environment. heck, even with the noise, they are hot. currently, I am running the accelero xtreme 5870s on both which are whisper quiet and significantly cool the cards. one noteworthy issue: the vrms, though with beefer heat sink, run hotter than with stock cooler. go figure. the result, you have to turn up the fans to keep the vrms cooler while the core temp at 48-50c for lower gpu, 58-60c for upper gpu. not a problem as the fans quiet, but aggravating and no answer from ac. with my case, I realize no significant heat increase in the box, haf932. e.g., cpu cooler stayed in the 1700rpm and per efleet the temps marginally increased across the board. also, strongly advise using gpu-z as there are many temps to keep in mind. the core temp could be great says ccc while the vrms flaming, like way over 100c. supposedly the vrms are built to handle up to something like an insane 130c, but that is too hot for my taste.

EDIT: the noted gpu (upper/lower) temps are crunching dnetc, the hottest. much cooler for both mw and collatz, nearly 10c for collatz.

I had a similar problem with 2 5870's. My problem was that I run a single
monitor and I had disabled the monitor of the second card in CCC. I connected
a second monitor, enabled it after reinstalling CCC, allowed CCC to identify it and all the lag disappeared. And the time per wu decreased by about 40%. I guess this is where the dummy plug comes in handy. The second monitor I have switched off but keep it plugged in the card. My cards are also in slots 1 and 3 to keep them as far apart as possible for overheating problems.
I hope this helps :-)

Ahhh yes yes i now understand your question.

Toppie hits the spot with that problem with some weird reaction of dual cards, my approach is different.

I use dummy plugs and also not run in crossfire.

I totally removed CCC and just installed the plain 10.X drivers then after they are installed i go into the config screen and under monitor select the other card and extend my desktop to it, which does the same trick as toppie posted.

Your right about not a hardware issue, but you compare 2 totally different projects, first of all collatz/dnetc are single precision projects

Milkyway however is a double precision project, now i am not totally sure but i am under the impression that double precision does stress the gpu much more.

That Dnetc heats up your cards more is kinda weird since for my cards its MW who heats them up much more, but maybe thats because i edit the app_info.xml to let the card work harder (about 99% load).

I have not looked into dnetc (about 96% load) but from what is saw is that on that project the cards overall run many times idle while some data is being processed by the cpu.

I am sure if cruncher see this he will react ;)

---

Its new, its relative fast... my new bicycle

it is implied, imo, that I understand these are different projects, what I don't understand is what they ask of the gpu such that the temps are as noted above.

you noted that mw runs hotter, but it sounds like you have not tried dnetc. this is not a percentage gpu usage thing, I run 100% for all gpu crunching. what brilliant soul would pose this matter and then slide in a note that he/she runs only 50% gpu when crunching mw and 25% for collatz? wow, slap me down.

I observed the temps when crunching across the three projects and dnetc is clearly the hottest making of my gpus. again, that is the primary reason I burped this thread, it makes no sense.

by temps, hottest to coolest: dnetc, mw, collatz
by credit, highest to lowest: dnetc, mw, collatz
(as observed for weeks for all projects)

NOTE: I think the credit hierarchy is understood, for whatever reason, dnetc is a much higher credit given project. the temps, that is what I have seen and might just be my little world though I doubt it.

is it a coincidence the temp to credit hierarchy? part of me says yes yet part says no. I hope someone can explain this matter who is familiar enough with this project (come on admin), and therefore likely to have very good understanding of the other two.


thanks.

DNETC also runs my GPUs hotter than MW with Collatz running cooler than MW.

Dnet does give more credit, but with MW, as many have already said, you can run your memory speed at minimum and not affect your performance (maybe 1 or 2 seconds from observation)! Running my 5870's memory at 300mhz in lieu of 1200mhz helps quite a bit to keep the card cooler!

Still would like an option to have bigger wu's, like Dnet. At +/- 85 seconds per WU, they just go by too fast.

Oh, and never experienced any Lag issues, probably another factor which causes it. Using 10.8 drivers right now, no issues.

L.

Assuming full GPU load percentage is achieved I think ATI cards reach about 85% of the maximum efficiency possible on MilkyWay which is about 70% of peak flops and Collatz it is about 58% of peak flops for HD 58xx cards. This is related to how much work can be assigned to the groups of 5 VLIW processors. It could be described as how full of work they are. This is why the GPU core runs hotter on MilkyWay than Collatz. It is doing more work.

"On MW a typical GPU reaches about 85% of efficiency (which is not exactly the ratio of peak flops, that's only ~70% ;)" For Collatz "the average utilization in the innermost loop is about 58% for HD5000 cards. As 100% would equal 1360 GigaInstructions/s (on a HD 5870), one arrives roughly at 785 GInstructions/second."

I don't know the figures for DNETC but judging from the heat of the GPU core, it is possibly attaining a similar efficiency to MilkyWay.

Collatz and MilkyWay require more CPU resources than DNETC for fastest processing and GPU load percentage as shown in GPU-Z will often be less than 100% on multiple GPU configurations when all CPU cores are running CPU projects. To enable the fastest GPU processing (and therefore the hottest GPU temperature) for my 3 GPU core configuration requires me to use the b-1 parameter and leave 2 CPU cores free for Collatz and 3 CPU cores free for MilkyWay.

DNETC uses very little CPU resources except for when tasks get stuck at the end. Also multiple GPU configurations only process one task at a time shared between all the GPU cores. Because of this DNETC processing speed is almost unaffected by having all the CPU cores running CPU projects. Also DNETC tasks run for over 5 minutes on all GPU cores whereas on MilkyWay a task completes every 80 seconds or so on each GPU core so there is a small time between each task which stops the temp building up as much. In other words DNETC runs at full speed and highest GPU temperature at all times, whereas a default configuration of MilkyWay or Collatz will often run at less than full speed and so a corresponding lower GPU temperature.

Therefore those people who run MilkyWay unoptimised with all CPU cores in use may run at a GPU load of less than 100%. This is particularly the case for those with a multiple GPU configuration. These people will experience that DNETC runs much hotter than MilkyWay. A fully optimised MilkyWay which has been allocated sufficent CPU resources running on a multiple 58xx GPU core setup will give similar temps on the GPU cores and VRMs to DNETC if a suitable low GPU memory speed is used on both.

Running a default configuration at slightly less than full GPU load percentage is not a bad thing and is the safest option for those who live in hot climates or have poor case ventilation and do not monitor GPU temperature closely. For MilkyWay I adjust the parameters in summmer to reduce GPU temps to a more reasonable level or run Collatz during the hottest part of the day. As there is no adjustment available with DNETC it would not be possible for me to run it on a hot Australian summer day without risking my GPUs. It would require air conditioning, water cooling or unacceptable GPU fan noise.

ok kashi, I appreciate the post, lots of good stuff. thanks.

you are not in summer, right? so you are running opti-MW at this time, is that correct? you are running 2 WUs at a time? right? I run four at a time. actually, sorry, currently not running as I just discovered 2 WUs hosed at ~60min each. :| when running they are usually done in ~157-160s, all four. I just noticed a bunch of new results, like doubled times so not sure if that interference on this end or changes server side. no matter, I am curious about your times to see how much cpu cores influence WU time to crunch. why?

I do not set aside cpu cores, run no cpu. could that influence the screen lag issue, that I do not configure cpu cores for collatz and mw? the screen lag appears the same for both collatz and mw, and again(bump) no lag for dnetc.


about that heat, lovely heat: I am not sure about non-reference 5870s, but for my 5870 reference cards the accelero xtreme 5870 cooler is very quiet and much cooler, mostly cooler anyway. I am still working (thoroughly mystified by) vrm temps which run hotter than with stock cooler. overall, a much better solution provided by the 3rd party cooler, but I currently have to run the upper card fans at 70% to keep the vrms under 90c. the core temp for both cards is <=60c. by the way, 70% fans is whisper quiet. I have two of these coolers running 50% lower gpu, 70% upper gpu and both are quieter than one stock xfx cooler at 35%. but then, the xfx at 40% was unacceptable for my environment. again, reference cards.

thanks again.

Yes that is correct I'm running 2 tasks concurrently per GPU core with <count>0.5</count> parameter. I assign 3 CPU cores to support GPU crunching by adjusting the parameters. So for 6 tasks processing concurrently on the 3 GPU cores it is
<avg_ncpus>0.6</avg_ncpus>
<max_ncpus>3</max_ncpus>

I have done this for a long time now. When I first started GPU processing on a HD 3850 I noticed a significant MilkyWay slowdown when I had Einstein tasks on all 8 cores. So I used to reduce the cores available to BOINC to 6 or 7. Later I did it with the ncpus values in the app_info.xml file. That is better for me because when MilkyWay runs out of work the computer will start using 8 cores again for CPU tasks. When I added a 5970 to my 5870 I increased the CPU cores allocated to MilkyWay from 2 to 3.

Not Summer but Spring here now, has been cold until yesterday when I needed to reduce the clockspeed a little. My 5870 is a Version 1 Vapor-X model, it cools both the core and VRMs extremely well even if I run it at 980MHz but is noisy when the fan exceeds 70%. 5970 is a reference model, fan stays at 30-38% and is reasonably quiet but it gets a bit hot. When the 5970 temp hits 83 °C I reduce the clockspeed a little because I know the VRMs will be at 95-97 °C then on MilkyWay or DNETC. I actually have an Accelero Xtreme 5970 I purchased about 4 months ago but I haven't fitted it due to some concerns I had about possible damage. Plus I need to flash the BIOS on the 5970 to change the default fan speed because fan speed control is global so I must use auto fan speed. If I adjust fan speed manually either the 5870 or the 5970 will be at the incorrect speed, which means either one will be too noisy or the other too hot.

Regarding your hot VRMs using the Accelero Xtreme 5870, I read a review somewhere of the Xtreme 5970 or Xtreme 5870 that said they needed to use 100% fan speed to keep VRMs at a reasonable temperature when doing intensive GPU benchmarking and stress testing.

I don't see any benefit from running more than 2 tasks concurrently. I only run 2 concurrently to minimise thermal cycling, not for the extra output. Perhaps if you tried running only 2 concurrently it may help with your screen lag issue as it would use less GPU memory.

I have heard of people getting not just screen lag but also everything going slowly due to memory mapping issues with multiple 1024MB GPUs. I needed to adjust a setting in the BIOS called Memory LowGap when I added a 5970.

kashi, thanks you for the time and the good information.

regarding the vrms, on this end. the stock cooler kept the vrms below core temp and the stock cooler is nothing more than a thermal pad on the vrms. the accelero has the pad + heat sink with more fans. my guess, and this is a guess, with reference card, again 5870, the stock fan is on the end that houses the vrms so a lateral cooler blast hits the vrms (but then the flow warmed for the rest of the card as the flow continues, i.e., warmer core.) vs. the accelero which has the larger thermal pads and heat sink that obstruct the air flow. ok, there is a lateral flow but with multiple fans there has to be flow contention and diminished lateral across the vrms. again, just guessing. either that or the thermal pads for the much beefier accelero are significantly inferior. NOTE: I do have contact, I checked and rechecked. the cooler overall runs the card much cooler, but as said, to keep the vrms down the fans must run more. very disappointing and no answer from artic cooling. one day I will experiment with removing the vrm heatsink and trimming the thermal pad obstruction. just so you have an idea how deficient the vrm cooling: if I leave the cooler on auto, the core is <70c for dnetc upper gpu (hottest), and the fans run at ~20%. meanwhile, the vrms run ~110c, steady. I continue to read vrms are built to handle it, but consider it way too much for 24/7.

regarding 4 at a time. I might have mislead you; actually, I did. two at a time per gpu. my apologies. and for the same reasons as you noted.

ok, mate, that is great help. I appreciate it.

Yes I don't like the VRMs to go above 100, even if they are supposed to be able to handle more than that. When they get that hot I either reduce the GPU core temp, increase the w parameter to reduce the GPU load percentage slightly or switch to Collatz until the weather cools. Hot VRMs only happen on my 5970, Vapor-X 5870 can run on the hottest days with no problem, only an an increase in noise.

I think you are correct about the different air path of the Accelero coolers being not as good as the reference models for VRM cooling. I believe a previous Turbo Pro model I had on my HD 4890 had the same disadvantage although it cooled the core well and was very quiet.

By the way, the wait factor (w parameter) is also the main parameter used to deal with a sluggish user interface, so you may be interested in experimenting with it.

"You could use this for throttling of the GPU in case the high GPU load leads to a very sluggish behaviour of the user interface or even VPU recover events. Setting w1.1 could improve the situation (see also the f, b, and p options)." - from the readme file.

I have a 5970. How do you measure the VRMs temperature?

Shows it in GPU-Z if you scroll down in the Sensor tab section. VDDC Phase #1, VDDC Phase #2, VDDC Phase #3 are the ones that get hot on MilkyWay and DNETC. VDDC1 Phase #1 and VDDC1 Phase #2 relate to the memory I think because they run cooler at a reduced memory speed on MilkyWay and DNETC and warmer on Collatz with maximum memory speed.

Shows it in GPU-Z if you scroll down in the Sensor tab section. VDDC Phase #1, VDDC Phase #2, VDDC Phase #3 are the ones that get hot on MilkyWay and DNETC. VDDC1 Phase #1 and VDDC1 Phase #2 relate to the memory I think because they run cooler at a reduced memory speed on MilkyWay and DNETC and warmer on Collatz with maximum memory speed.

Thanks....

it is implied, imo, that I understand these are different projects, what I don't understand is what they ask of the gpu such that the temps are as noted above.

you noted that mw runs hotter, but it sounds like you have not tried dnetc. this is not a percentage gpu usage thing, I run 100% for all gpu crunching. what brilliant soul would pose this matter and then slide in a note that he/she runs only 50% gpu when crunching mw and 25% for collatz? wow, slap me down.

I observed the temps when crunching across the three projects and dnetc is clearly the hottest making of my gpus. again, that is the primary reason I burped this thread, it makes no sense.

by temps, hottest to coolest: dnetc, mw, collatz
by credit, highest to lowest: dnetc, mw, collatz
(as observed for weeks for all projects)

NOTE: I think the credit hierarchy is understood, for whatever reason, dnetc is a much higher credit given project. the temps, that is what I have seen and might just be my little world though I doubt it.

is it a coincidence the temp to credit hierarchy? part of me says yes yet part says no. I hope someone can explain this matter who is familiar enough with this project (come on admin), and therefore likely to have very good understanding of the other two.


thanks.

Dude start messing with the optimized application and start tinkering with the settings to get you gpu loaded more then come back with a serious answer.
If you run plain mw from the website it is not running at his max, the main reason is that you start complaining that your desktop starts being slow as hell if it does.
Again the reason for dnetc to heat your gpu more is that it runs more then one thread, if you run MW like we all run in the paste with 3 to 5 units you'll see what i mean by that.
So its plain and simple most am too lazy to run it the old way and just let it run how it is set by boinc with only 1 unit > PERIOD <
If you don't use the pc for other purpose other then this projects you can set it that way because you won't do anything other then check it once in a while, but if not you won't like the slowness when running 3 or more threads

ps: when you run the way MW is made since start with mulitple threads you won't see your cards used on the site anymore it will only report your running anonymous platform, and your output will slightly grow also since it will stress the card more (1 unit stresses my card up to 96-98% with 3 it shows 99-100% )

The reason why i stopped at Dnetc is that my card was getting units which gave other persons 6500 points and i got 5000 i have posted about that on dnetc about that but never got an answer.

---

Its new, its relative fast... my new bicycle