Too much work
30 Dec 2012 17:37:08 UTC
Topic 196711
(moderation:
Dec 25 I brought up an old, retired host with a new video card. I attached to E@H and got sent a slew of tasks that Boinc thought would take less than 2 hrs ea.
The reality is they are taking a bit more than 6. Many will time out Jan 8. What is the best way for me to abort them. I have no intention of wasting electricity processing timed out data and I would like to minimize the impact on the project.
Too much work
)
It depends on your ambitions: even tho the Pentium 4 has a much slower CPU than your second PC (also using a GT 430), the runtime of BRP 4 tasks running mainly on the CPU is a bit disappointing, so you might want to investigate first why the GT 430 in the P4 is that slow. Things to check first would be CPU and GPU temperature (is ventilation working ok?) and whether the PCI bus speed is ok. Tools like CPU-Z and GPU-Z are great for basic checks like this. You also don't want graphics intensive screen savers to steal away performance from BOINC.
If you are ok with the performance, you can basically let BOINC manage the tasks. BOINC will have noticed by now that its initial estimation was too optimistic and will throttle down further work fetching. If you want to be nice to your fellow crunchers by aborting tasks that you know will not finish in time, you can use BOINC manger in "advanced mode" to do so.
- if in simple mode, switch BOINC manager to advanced mode (so that you see a tabbed view with tabs like "projects," "tasks" , statistics etc.
- you can sort tasks by dealine by clicking on the deadline column header in the tasks tab view
- select some of the tasks , and click on the "abort button" . A confirmation dialog will come up
If you want to make sure that no additional tasks are assigned for your PC from Einstein@Home for some time, got to the "projects" tab, select the Einstein@Home Project from the list of projects, and click on the "no more work" button. Click on the button again to allow more work whenever you want.
But again, this is not strictly necessary as BOINC will adjust itself in the long run.
Cheers
HB
RE: Dec 25 I brought up an
)
Whenever I attach or re-attach a host to the project, I always (temporarily) set the cache size (on that host - local prefs) to something like 0.2 days. That way there will be no surprises in how big the initial supply of tasks might be. I also tend to increase the cache setting in several small steps to see what happens. When everything is running as intended, I can revert to website prefs simply by clearing the local settings.
This has saved me a number of times over the years :-).
Cheers,
Gary.
Thanx for the tips I did
)
Thanx for the tips
I did expect this P4 to be slower because of the PCI buss vs PCIe buss.
These are simular to the results I have seen with this PCI card over at Seti, I was hoping for more, but it only cost $36 incl S@H.
The temps on the slow one according to CPUz sre in the low 60's C.
I have not yet found a temp monitor that is compatible for this old CPU.
I do appreciate the advice on aborting the tasks.
Oh well an inexpensive experiment and since I live in a part of the world with "cheap" electricity the old P4 will crunch on for a while.
I notice the old P4 is a
)
I notice the old P4 is a single core 2.4GHz. Are you crunching anything else on that single core? If you are, it would be worthwhile to temporarily suspend all CPU tasks and see if that makes any difference to the overall GPU crunch time. You might be able to get a significant improvement by having the free core.
Can you change the motherboard in that machine? If it's not a proprietary form factor you could consider replacing it with an el cheapo H61 board and install something like a cheap Celeron or Pentium dual core (G550 or G645) and 4GB of DDR3 RAM. You can get cheap H61 boards with a legacy PCI slot - ASUS P8H61-MLE for example. The total cost of those three (Mobo/CPU/RAM) would be around $100 in your country. I've just paid $113 for them here in Aus where prices are dearer than in the US.
If you lashed out a bit more and bought a GTX650 - should cost ~$100 - I pay $109 - you would have a very good performance and end up with a RAC north of 25K. Take a look at this host. It was an old Athlon XP of exactly the same vintage as your P4. It now has exactly what I just described with the Pentium G645 rather than the Celeron. Using the slightly dearer Pentium dual core gives a small improvement to CPU tasks but not the GPU ones. If you're on a very tight budget the G550 is fine. That machine has a current RAC of 26K and it is still rising.
The really big bonus of doing something like this is something that most people don't even consider. Sure, electricity in the US is still (for the moment) cheap, but old P4 style machines really use it hard. At full load (2xGPU 2xCPU) the above linked machine is drawing 140 watts from the wall. The Celeron variety is drawing 125 watts. I hate to think what your P4 is drawing. I imagine you could pay for the entire upgrade (including GPU) from electricity savings, well within the expected life of the upgraded machine.
EDIT: Just a small extra note on the importance of using a Kepler series GPU. I have an identical machine to the one drawing 125 watts. The only difference is that it has a GTX550TI - not Kepler. It is drawing 190 watts from the wall. Sure, it is doing GPU tasks around 20% faster but it's using around twice as much electricity to do so. Without a GPU, the basic machine consumes 55 watts on full load. A GTX650 adds 70 watts. A GTX550TI adds 135 watts.
Cheers,
Gary.
RE: EDIT: Just a small
)
Yes I run a couple of the 550Ti's (OC'd) here at home along with a SC'd 660Ti and the 610M on this laptop for GPU tasks and a couple others just running GRP's and I just try not to think of mt light bill until I go pay the bill and since it is every 2 months I can quickly forget about that part of the project.
I'm up north in NW Washington State and I get all my parts from tigerdirect.com/
Happy New Year