hello,
my pc began to freeze last night when started after a few hours the graphics cards fans would go wild and the pc would freeze again
On testing today i find the work units are running 99% activity around 10% higher than usual and are complete in 15 seconds less time
all three cards are doing this and all three are overheating for some reason

so are the work units more compressed or something needing more work done and taking less time to do them ?

best regards
Ian

---

....Please Join team Scotland HERE

No change at my end. In fact my PC has gone for 36 hrs without freezing for once.

Not as far as I can see here, either

---

Go away, I was asleep

It is possible that your CPU is under less load than previously. This will increase the GPU load %. It is especially noticeable when running 3 or more GPUs in one box. The increased CPU resources available to support GPU processing speeds up task completion and increases GPU core temp which may also increase GPU fan speed.

If this is the case the more recent tasks that are finishing quicker will show an increased CPU time in Task Details. For example a task that completes in 159.825 seconds GPU time shows a CPU time of 8.78286 seconds, whereas a similar task that completes in 175.595 seconds GPU time shows a CPU time of 2.88602 seconds.

To reduce GPU temp and improve stability you can reduce the GPU load % by downloading and installing the optimised application and increasing the w parameter to a value higher than the default of 1.

Hi,

spot on, because the machine kept hanging i stopped all the climate models and it seems only running milkyway runs the cards alot harder and therefore alot hotter
i had read that there was a small performance drain running both cpu work and gpu work but never thought it would have such a big effect

activity on the cards went from 99% to between 93% and 95% on each card wich is quite a drop just hope the gpu work doesnt effect the cpu climate models in the same way although they take so long to show any differance it would take several days for any changes to show up

thank you for your explanation it was a real eye opener

best regards
Ian

---

....Please Join team Scotland HERE

Yes the difference is not as noticeable when only using one GPU. I adjust the ncpus values in an app_info.xml file to use 3 cores of the 8 virtual cores of my CPU to support speedy MilkyWay processing on 3 GPUs, which means only 5 cores are available for running CPU projects.

Due to contention issues on CPUs with hyperthreading, often there is not a large difference in output when running CPU projects on 6 or 7 cores rather than the full 8 cores and only a small drop when using 5 cores. On some projects such as PrimeGrid LLR SOB there is a only a 3-6% overall gain in output when running with HT on (8 cores) compared with HT off (4 cores) on an i7 950.

Running CPU projects on fewer cores reduces CPU temp significantly on "hot-running" projects and allows the use of a higher clockspeed or a more stable computer during summer heatwaves. A lower CPU temp also implies a lower power consumption which is useful for me. It is only a small saving compared to the power the GPUs use but it helps a little now that electricity bills are becoming gigantic here.

Many have no trouble but in the past I found my older 4 core computer became unstable when I tried to run Climate models on all 4 cores. It would crash or freeze many times a day.

Hi,

Thanks again for all that interesting info
i run an i7 920 using all 8 virtual cores at a standard overclock for this chip of 4 gig on air cooling
the funny thing was i could run the climate models ok then switch them off and run all three gpu's on here slightly overclocked but when i try to run both after about 30 mins things pear shaped
what seems to happen along with the pc freezing is one of the cards fans goes into overdrive so it may be a problem with one of the cards knocking out the rest of the system
will have to do some serious hardware tinkering to test the cards seperately to see if thats whats going on

thanks again

Ian

---

....Please Join team Scotland HERE

I also have encountered instability on my W3520 (=i7 920) at default clock when running the longer AQUA multi threaded work units and 3 GPUs. Other projects I have tried are OK but I have not done any Climate recently. I run the CPU at a reduced voltage, so that may contribute to AQUA instability. The same may be true for your level of overclock with Climate, I remember reading on their forum that Climate models can put more strain on overclocked CPUs than many other projects.

Perhaps you could try running Climate on less than the full 8 cores and run another project on the other core(s). I know it is almost blasphemy but you could also consider reducing your overclock slightly when running Climate models. Just suggestions, not nice waking up and discovering all your projects have been idle all night while you were asleep due to a crash or freeze.

I actually prefer higher wu sizes especially for 5xxx series because that gpu load drop occurs at each wu shift. Like what DNETC@Home did on wus for 5xxx series card (6 times bigger wus), it could be done on Milkyway@Home to reduce the server load and to provide a more efficient gpu utilization.