By that, do you mean that you're not running any GPU tasks at the moment?
Quote:
I haven't changed any settings and running 4 E@H WU's (nothing else no POGS)
You say "4 E@H WU's" and that seems to imply 4 CPU tasks since you have said "GPU stuff aside" but perhaps you mean 2 CPU tasks and 2 GPU tasks. This is what should be running if the settings you previously mentioned still apply.
At some point recently you seem to have been running 4 CPU tasks with a GPU task because your task in this workunit took an order of magnitude longer to complete than other similar tasks in your work cache. That sort of thing happens when there are no free cores to support the GPU task. If GPU tasks normally take about an hour, they don't suddenly take 10 hours for no reason at all.
Quote:
BOINC is showing 40hrs and 70 hrs to complete 2 of the WU's. Surely that can't be right. They are S6 WU's.
You should disregard estimates, particularly if running multiple concurrent GPU tasks. Einstein still uses Duration Correction Factor (DCF) to adjust task estimates. If you run 2 concurrent GPU tasks, BOINC sees the extended completion time 'per task' without taking into account that the 'real' per task time is actually half of that (because of 2 concurrent). BOINC will effectively double the DCF and cause the estimates for CPU tasks to double as well. I see this all the time with my FGRP4 CPU tasks being estimated at 20+hours but actually taking maybe 6-8 hours. Each time CPU tasks complete, BOINC will use the quick completion time to adjust the DCF downwards (and all estimates will reduce) and each time GPU tasks complete, BOINC will think your machine has slowed and will bump the DCF (and the estimates) back up again.
So, you don't need to worry about fluctuating estimates on CPU tasks. They will take the correct amount of time irrespective of what the estimate says. Your most recently completed S6 task took around 5 hours. To get to estimates of 40 or 70 hours, that extremely long running GPU task probably caused that. If a GPU task takes 10 times longer than it really should, it will cause the DCF to increases by the same x10 factor, which when applied to the estimates of CPU tasks, gives the types of figures you quote. It's very easy to fix this. Don't run GPU tasks without the appropriate number of free cores to give support. As you complete tasks with the proper elapsed times, BOINC will progressively refine the estimates back down to more normal values.
Quote:
I'm not sure if it's a good idea for me to run different projects at the same time. Especially if I have to change settings to allow them to run simultaneously.
In the past, I have run POGS and Einstein simultaneously without issue. If you specify exactly what searches you would like to run, it should be possible to nominate settings to give you a trouble free experience. You should be pretty much able to set and forget without having further problems. Maybe you'd like to consider POGS on your CPUs with BRP6 on your GPU. POGS does have outages so you could consider E@H as a CPU backup.
Ok, well basically I have now set my E@H preferences for GPU utilization factor of BRP at 0.5.
I have also set my BOINC preference to "use at most 75% of the processors.
With that set up, currently running is
3 POGS WU's.
1 Einstein @ Home WU (BRP6)
GPU load is at 73%.
I did have my PC o/c'd but have seen some real instability in the last 24hrs which has been rather unusual so I'm now just running stock standard. (Power surges)
Now, not to confuse myself but I thought that by having my BRP6 setting at 0.5, that would allow a second BRP6 WU to begin. I'm guessing that's not happening because I don't have a spare CPU core?
Now, not to confuse myself but I thought that by having my BRP6 setting at 0.5, that would allow a second BRP6 WU to begin.
It is a "feature" of the otherwise rather nice Einstein web-page specified GPU task multiplicity function that changes don't actually take effect on a particular host until a new WU has been assigned and downloaded from the project after one has made the change at the web page.
Perhaps this had not yet happened on your machine at the time you made the post? You may be able to nudge the process along sooner by manually request of a project update.
Ok, well basically I have now set my E@H preferences for GPU utilization factor of BRP at 0.5.
I have also set my BOINC preference to "use at most 75% of the processors.
With that set up, currently running is
3 POGS WU's.
1 Einstein @ Home WU (BRP6)
That's fine and as archae86 points out, you will start running a second BRP6 task as soon as new BRP6 work is downloaded. You can quickly trigger this by adding a small amount to your preference setting for maximum work buffer. Just add 0.2 days to whatever value you currently have set. You can change preferences on the website (if that is what you prefer) by going to your account page and selecting 'computing preferences' or locally through a menu item in BOINC Manager - Advanced view.
If you set the value locally, it will over-ride (but not change) whatever is set on the website. You can cancel local prefs and return to website prefs at any time by clicking the appropriate button in the local view. It's documented on the top of the window when you select Advanced view -> Tools -> Computing preferences. In either case, website or local preferences, if you don't see immediate action, select the Einstein project (Projects tab of BOINC Manager) and click 'update' (which stops being greyed out when you 'select' a project) in order to communicate the pref change between client and server immediately. This should result in new work being downloaded and so the 2nd GPU task should start running.
You should notice that as soon as the 2nd GPU task starts, one of the three running CPU tasks will change to 'waiting to run'. You will have two free CPU cores, one caused by the 2 concurrent GPU tasks and the other because you have set BOINC to use 75% of the CPU cores. Your tasks list on the website shows two BRP6 tasks returned relatively recently taking around 4,900 secs. I assume that these two were crunched concurrently because you have earlier tasks around the 3,000 sec mark which were probably crunched singly. If this is true, you have a nice speedup (4,900 for the two instead of well over 6,000).
You're not doing too bad yourself :-). You have 6.5M total and 18K RAC. I'm pretty sure I saw 14K yesterday or the day before so you're really climbing quickly, now that you are starting to make better use of your GPU.
Your GPU is capable of a RAC of at least 150K. Take a look at this host owned by mountkidd which has probably the same GPU as yours. The RAC is over 180K and not very much of that would be coming from the somewhat better processor than your Sandy Bridge i5. Your CPU is quite suitable for the task. My most productive hosts have HD7850 GPUs running BRP6 tasks 4x (2 free cores) and have RACs between about 85-95K depending on the CPU. Many of mine run on 'old' Q8400 quads or newer i3s with HT. My 'best' CPU is an i5 3570K.
Since BRP6 tasks come from the Parkes (NSW) Radio Telescope data, both you and I should have a particular interest in crunching them :-). In the next year or two, when data from advanced LIGO becomes available, Gravity Wave tasks will be the hot stuff to crunch. In the meantime let's go find a previously undiscovered pulsar - a radio pulsar through BRP6 or a gamma ray pulsar through FGRP4. The current GW run, S6 follow-up #2, is doing closer inspection of some of the 'best' candidates from previous runs. Gravity waves are proving elusive so I decided to save my efforts for when advanced LIGO data is available. I guess it's a bit like the Higgs. It was proving elusive until the upgrade of the LHC finally delivered enough power to unlock it. I imagine there will be a similar frenzy when GWs are finally directly detected (or when Einstein's theories have to change to account for their non-discovery :-). Either way, it's a fascinating time to be involved in crunching the data.
RE: GPU stuff aside
)
By that, do you mean that you're not running any GPU tasks at the moment?
You say "4 E@H WU's" and that seems to imply 4 CPU tasks since you have said "GPU stuff aside" but perhaps you mean 2 CPU tasks and 2 GPU tasks. This is what should be running if the settings you previously mentioned still apply.
At some point recently you seem to have been running 4 CPU tasks with a GPU task because your task in this workunit took an order of magnitude longer to complete than other similar tasks in your work cache. That sort of thing happens when there are no free cores to support the GPU task. If GPU tasks normally take about an hour, they don't suddenly take 10 hours for no reason at all.
You should disregard estimates, particularly if running multiple concurrent GPU tasks. Einstein still uses Duration Correction Factor (DCF) to adjust task estimates. If you run 2 concurrent GPU tasks, BOINC sees the extended completion time 'per task' without taking into account that the 'real' per task time is actually half of that (because of 2 concurrent). BOINC will effectively double the DCF and cause the estimates for CPU tasks to double as well. I see this all the time with my FGRP4 CPU tasks being estimated at 20+hours but actually taking maybe 6-8 hours. Each time CPU tasks complete, BOINC will use the quick completion time to adjust the DCF downwards (and all estimates will reduce) and each time GPU tasks complete, BOINC will think your machine has slowed and will bump the DCF (and the estimates) back up again.
So, you don't need to worry about fluctuating estimates on CPU tasks. They will take the correct amount of time irrespective of what the estimate says. Your most recently completed S6 task took around 5 hours. To get to estimates of 40 or 70 hours, that extremely long running GPU task probably caused that. If a GPU task takes 10 times longer than it really should, it will cause the DCF to increases by the same x10 factor, which when applied to the estimates of CPU tasks, gives the types of figures you quote. It's very easy to fix this. Don't run GPU tasks without the appropriate number of free cores to give support. As you complete tasks with the proper elapsed times, BOINC will progressively refine the estimates back down to more normal values.
In the past, I have run POGS and Einstein simultaneously without issue. If you specify exactly what searches you would like to run, it should be possible to nominate settings to give you a trouble free experience. You should be pretty much able to set and forget without having further problems. Maybe you'd like to consider POGS on your CPUs with BRP6 on your GPU. POGS does have outages so you could consider E@H as a CPU backup.
Cheers,
Gary.
Ok, well basically I have now
)
Ok, well basically I have now set my E@H preferences for GPU utilization factor of BRP at 0.5.
I have also set my BOINC preference to "use at most 75% of the processors.
With that set up, currently running is
3 POGS WU's.
1 Einstein @ Home WU (BRP6)
GPU load is at 73%.
I did have my PC o/c'd but have seen some real instability in the last 24hrs which has been rather unusual so I'm now just running stock standard. (Power surges)
Now, not to confuse myself but I thought that by having my BRP6 setting at 0.5, that would allow a second BRP6 WU to begin. I'm guessing that's not happening because I don't have a spare CPU core?
Thx guys :)
RE: Now, not to confuse
)
It is a "feature" of the otherwise rather nice Einstein web-page specified GPU task multiplicity function that changes don't actually take effect on a particular host until a new WU has been assigned and downloaded from the project after one has made the change at the web page.
Perhaps this had not yet happened on your machine at the time you made the post? You may be able to nudge the process along sooner by manually request of a project update.
Slightly off topic sorry Gary
)
Slightly off topic sorry Gary but ...how the HECK do you achieve so many credit points for Einstein @ Home??
That's a phenomenal number!!
Ah I see. Thank you archae86.
)
Ah I see.
Thank you archae86.
RE: Ok, well basically I
)
That's fine and as archae86 points out, you will start running a second BRP6 task as soon as new BRP6 work is downloaded. You can quickly trigger this by adding a small amount to your preference setting for maximum work buffer. Just add 0.2 days to whatever value you currently have set. You can change preferences on the website (if that is what you prefer) by going to your account page and selecting 'computing preferences' or locally through a menu item in BOINC Manager - Advanced view.
If you set the value locally, it will over-ride (but not change) whatever is set on the website. You can cancel local prefs and return to website prefs at any time by clicking the appropriate button in the local view. It's documented on the top of the window when you select Advanced view -> Tools -> Computing preferences. In either case, website or local preferences, if you don't see immediate action, select the Einstein project (Projects tab of BOINC Manager) and click 'update' (which stops being greyed out when you 'select' a project) in order to communicate the pref change between client and server immediately. This should result in new work being downloaded and so the 2nd GPU task should start running.
You should notice that as soon as the 2nd GPU task starts, one of the three running CPU tasks will change to 'waiting to run'. You will have two free CPU cores, one caused by the 2 concurrent GPU tasks and the other because you have set BOINC to use 75% of the CPU cores. Your tasks list on the website shows two BRP6 tasks returned relatively recently taking around 4,900 secs. I assume that these two were crunched concurrently because you have earlier tasks around the 3,000 sec mark which were probably crunched singly. If this is true, you have a nice speedup (4,900 for the two instead of well over 6,000).
Cheers,
Gary.
RE: That's a phenomenal
)
You're not doing too bad yourself :-). You have 6.5M total and 18K RAC. I'm pretty sure I saw 14K yesterday or the day before so you're really climbing quickly, now that you are starting to make better use of your GPU.
Your GPU is capable of a RAC of at least 150K. Take a look at this host owned by mountkidd which has probably the same GPU as yours. The RAC is over 180K and not very much of that would be coming from the somewhat better processor than your Sandy Bridge i5. Your CPU is quite suitable for the task. My most productive hosts have HD7850 GPUs running BRP6 tasks 4x (2 free cores) and have RACs between about 85-95K depending on the CPU. Many of mine run on 'old' Q8400 quads or newer i3s with HT. My 'best' CPU is an i5 3570K.
Since BRP6 tasks come from the Parkes (NSW) Radio Telescope data, both you and I should have a particular interest in crunching them :-). In the next year or two, when data from advanced LIGO becomes available, Gravity Wave tasks will be the hot stuff to crunch. In the meantime let's go find a previously undiscovered pulsar - a radio pulsar through BRP6 or a gamma ray pulsar through FGRP4. The current GW run, S6 follow-up #2, is doing closer inspection of some of the 'best' candidates from previous runs. Gravity waves are proving elusive so I decided to save my efforts for when advanced LIGO data is available. I guess it's a bit like the Higgs. It was proving elusive until the upgrade of the LHC finally delivered enough power to unlock it. I imagine there will be a similar frenzy when GWs are finally directly detected (or when Einstein's theories have to change to account for their non-discovery :-). Either way, it's a fascinating time to be involved in crunching the data.
Cheers,
Gary.