Ever since I upgraded my CPU from an AMD dual core 4200 to a 8 core 8320 FX. I've been getting work units that take over 60 hours to compute.
You have no tasks listed on the website that have taken anything like 60 hours. The longest completion time (only 5 completed tasks showing) is about 9 hours. Are you talking about estimated crunch time rather than actual time? If you've just upgraded the CPU, the estimates may be quite wrong until BOINC fixes the DCF (duration correction factor).
You seem to have aborted lots of tasks. Why not set your cache size lower rather than excessively downloading and then aborting? With an 8 core host, you will be able to churn through lots of tasks anyway so you shouldn't really need to abort.
Quote:
Why does this happen?
Probably because of the CPU upgrade. You should not panic too soon. Wait until you have say 20 tasks completed and the estimates should be a lot better by then.
The Gamma-ray pulsar search #2 v1.10 tasks are very large. They take my older processor about 24 hours. Your completed task took more like 9 hours though.
Initial estimates are often off by a HUGE amount. Let a few run (rather than aborting them all...) and see how long they actually take.
Ever since I upgraded my CPU from an AMD dual core 4200 to a 8 core 8320 FX. I've been getting work units that take over 60 hours to compute.
You have no tasks listed on the website that have taken anything like 60 hours. The longest completion time (only 5 completed tasks showing) is about 9 hours. Are you talking about estimated crunch time rather than actual time? If you've just upgraded the CPU, the estimates may be quite wrong until BOINC fixes the DCF (duration correction factor).
You seem to have aborted lots of tasks. Why not set your cache size lower rather than excessively downloading and then aborting? With an 8 core host, you will be able to churn through lots of tasks anyway so you shouldn't really need to abort.
Quote:
Why does this happen?
Probably because of the CPU upgrade. You should not panic too soon. Wait until you have say 20 tasks completed and the estimates should be a lot better by then.
Yes, I dont know why. Before the upgrade, I only had a about 7-10 tasks pending. Now all of a sudden BONIC downloads 87 tasks, just after changing the CPU. My older processor was going through tasks every 5 hours, my new CPU took 10 hours for the last two task. Am I getting the same work units? or does E@h adjust size of the WU based on the CPU?
I think your BOINC installation filled its cache based on the initial Benchmark synthetic results multiplied by CPU cores and the duration correction factor (calibrates expected vs. actual previous runtimes, set by the previous CPU) still in the config.
If it's still like in BOINC V6, this can take upto 14 days to correct, as your new actual output is calibrated and cache finetuned to match its actual performance.
If you see fast/new CPUs chugging along apparently much slower than expected, check a few things :
I see you went from an old Socket AM2 CPU to a brand new AM3+ CPU, for which I assume you also installed a new Motherboard.
If you kept your existing installation with a new motherboard, old Chipset Drivers may interfere with your new setup's performance. Check the device manager for any missing Drivers or troubled devices. Where applicable, uninstall obsolete Drivers/Utilities from the Program Manager.
A Windows reinstall should not be mandatory with Win7 when changing motherboards to avoid problems, but there's still no guarantee for that, lots of issues can still arise.
- check that Cool&Quiet is working properly
(normally Windows 7 should do that, but manually looking at it should confirm that it's working properly - maybe your CPU isn't throttling up as it should. If you have Fusion/Catalyst Software installed, you can look it up there; otherwise CPU-Z is a small, useful Tool as well. As a temporary workaround or test, you could also set your energy saving in Win7 to max. performance, which should disable the CPU to throttle down)
- check CPU temperatures for CPU thermal throttling
(although unlikely if installed correctly, missing thermal paste between CPU and cooler is all it takes to run pretty high CPU temps that force it to clock down to prevent overheating. Temps can be seen in CPU-Z as well I believe - thermal throttling under load as well by monitoring clock rates, otherwise the BIOS Hardware Monitor should give sufficient temperature clues as well)
I think your BOINC installation filled its cache based on the initial Benchmark synthetic results multiplied by CPU cores and the duration correction factor (calibrates expected vs. actual previous runtimes, set by the previous CPU) still in the config.
If it's still like in BOINC V6, this can take upto 14 days to correct, as your new actual output is calibrated and cache finetuned to match its actual performance.
If you see fast/new CPUs chugging along apparently much slower than expected, check a few things :
I see you went from an old Socket AM2 CPU to a brand new AM3+ CPU, for which I assume you also installed a new Motherboard.
If you kept your existing installation with a new motherboard, old Chipset Drivers may interfere with your new setup's performance. Check the device manager for any missing Drivers or troubled devices. Where applicable, uninstall obsolete Drivers/Utilities from the Program Manager.
A Windows reinstall should not be mandatory with Win7 when changing motherboards to avoid problems, but there's still no guarantee for that, lots of issues can still arise.
- check that Cool&Quiet is working properly
(normally Windows 7 should do that, but manually looking at it should confirm that it's working properly - maybe your CPU isn't throttling up as it should. If you have Fusion/Catalyst Software installed, you can look it up there; otherwise CPU-Z is a small, useful Tool as well. As a temporary workaround or test, you could also set your energy saving in Win7 to max. performance, which should disable the CPU to throttle down)
- check CPU temperatures for CPU thermal throttling
(although unlikely if installed correctly, missing thermal paste between CPU and cooler is all it takes to run pretty high CPU temps that force it to clock down to prevent overheating. Temps can be seen in CPU-Z as well I believe - thermal throttling under load as well by monitoring clock rates, otherwise the BIOS Hardware Monitor should give sufficient temperature clues as well)
Yes, I have WIN 7. However I had lot of issues with drivers that are now resolved, I initially kept getting BSOD.
My CPU temps from Core temp max around 57c under max loading. My core speed is 3.7GHZ under load.
Installed all the chipset drivers, as far as I'm aware, there are no drivers for the processor.
Device manager indicates that all the 8 cores in the processor are ok. There aren't any drivers in device manager that indicate an error.
My boinc version is 7.0.64
Also there was a driver called EPU, I didn't instal this. I thought this was something to save power.
All of what you're seeing could be completely normal. As others have already said, you've got a time-based setting for how many WUs should cache. If BOINC thinks you're completing WUs much faster now, then it will download more so that your cache will last the same time as before. If you're getting too many WUs reduce cache size.
And your new "8 core CPU" is actually more like a quad core with hyperthreading, speaking in Intel terms. 2 cores in a "module" share some ressources, and most importantly the FPU, which is elemental for Einstein and most other BOINC projects. Hence the performance of one module per clock is less than what you're used from each core of your previous Athlon 64.
I looked as his CPU tasks and the long-running (pending/finished) as well as all aborted ones are the Gamma-Ray Pulsar Searches.
These take significantly longer on my Systems as well.
Got one Gamma-Ray Pulsar Search done in ~26hrs for example, compared to
- Binary Radio Pulsar Search (~2 hrs)
- Perseus Arm Survey (~2.8 hrs)
- Gravitational Wave S6 Directed Search (~9.3 hrs).
Dependent on how RAM bandwidth sensitive the Client crunching those WorkUnits is, it is also noteworthy that the 8 Cores of his system have to share the bandwidth of his DDR3 Dual Channel setup. That might turn out to be somewhat of a bottleneck, compared to i.e. 4 similar performing cores getting the same bandwidth like my FX-4100.
I would make sure that such a fast 8-Core CPU is using the fastest RAM I can/want to afford, but maybe that's already the case.
Add to that the Duration Correction factor that still has to work out for at least 10 days, those predicted runtimes could occur initially but his achieved actual completion times seem in order. Those predicted runtimes should correct down with every finished & reported workunit then.
My new setup (A10-6700 3.7GHz 4 Cores w/ its on-CPU GPU disabled) using DDR-1866 hasn't turned out sufficient CPU results to make a sound judgement of the impact of its faster RAM (the FX-4100 3.6GHz 4 Cores is running DDR3-1333) - would be a rough estimate at best anyway due to slightly different CPU architecture/caches etc...
But that's one for the patient benchmarking guys to discover. At least it's clear that different WorkUnits doing different searches massively vary in runtimes.
Just one question, is the amount of credit given constant per calculation? My older CPU will have to do the exact same amount of calculation to be given 1 credit unit as my new CPU?
I heard something a while back in 2010, that E@h was adjusting the amount of credit given than previously, to compensate for a newer generation of CPU's. Is there any truth to this?
Yes, basically it should be when comparing CPUs, but it'll always be just an approximation for a number of reasons (such as synthetic nature of benchmarks which only provide very coarse performance indicators).
If my memory serves me correctly, the accepted formula of Cobblestones / Credits granted for GFlops per time period is more or less given by Berkeley (BOINC developer/maintainer).
This used to be a rather rarely updated and more often intensively discussed guideline for all BOINC based projects to follow within some degree of flexibility.
I remember a few discussions about having more realistic/thorough benchmarks implemented instead of the very old Drystone/Whetstone benchmarks to achieve more realistic/precise results but I believe those suggestions were never adopted.
Just one question, is the amount of credit given constant per calculation?
That's a difficult question, actually. A much easier answer would be this: "The credits per WU are fixed. How fast you can return the WUs will determine your daily credits." But how many calculations your CPU needed to finish a WU actually depends on its architecture. The less efficient, the more it will need (e.g. Pentium 4, or memory bandwidth staved machines like an 8-core CPU with lowly clocked single channel memory).
Work units too large
)
You have no tasks listed on the website that have taken anything like 60 hours. The longest completion time (only 5 completed tasks showing) is about 9 hours. Are you talking about estimated crunch time rather than actual time? If you've just upgraded the CPU, the estimates may be quite wrong until BOINC fixes the DCF (duration correction factor).
You seem to have aborted lots of tasks. Why not set your cache size lower rather than excessively downloading and then aborting? With an 8 core host, you will be able to churn through lots of tasks anyway so you shouldn't really need to abort.
Probably because of the CPU upgrade. You should not panic too soon. Wait until you have say 20 tasks completed and the estimates should be a lot better by then.
Cheers,
Gary.
None of your completed tasks
)
None of your completed tasks ran for 60 hours.
The Gamma-ray pulsar search #2 v1.10 tasks are very large. They take my older processor about 24 hours. Your completed task took more like 9 hours though.
Initial estimates are often off by a HUGE amount. Let a few run (rather than aborting them all...) and see how long they actually take.
RE: RE: Ever since I
)
Yes, I dont know why. Before the upgrade, I only had a about 7-10 tasks pending. Now all of a sudden BONIC downloads 87 tasks, just after changing the CPU. My older processor was going through tasks every 5 hours, my new CPU took 10 hours for the last two task. Am I getting the same work units? or does E@h adjust size of the WU based on the CPU?
I think your BOINC
)
I think your BOINC installation filled its cache based on the initial Benchmark synthetic results multiplied by CPU cores and the duration correction factor (calibrates expected vs. actual previous runtimes, set by the previous CPU) still in the config.
If it's still like in BOINC V6, this can take upto 14 days to correct, as your new actual output is calibrated and cache finetuned to match its actual performance.
If you see fast/new CPUs chugging along apparently much slower than expected, check a few things :
I see you went from an old Socket AM2 CPU to a brand new AM3+ CPU, for which I assume you also installed a new Motherboard.
If you kept your existing installation with a new motherboard, old Chipset Drivers may interfere with your new setup's performance. Check the device manager for any missing Drivers or troubled devices. Where applicable, uninstall obsolete Drivers/Utilities from the Program Manager.
A Windows reinstall should not be mandatory with Win7 when changing motherboards to avoid problems, but there's still no guarantee for that, lots of issues can still arise.
- check that Cool&Quiet is working properly
(normally Windows 7 should do that, but manually looking at it should confirm that it's working properly - maybe your CPU isn't throttling up as it should. If you have Fusion/Catalyst Software installed, you can look it up there; otherwise CPU-Z is a small, useful Tool as well. As a temporary workaround or test, you could also set your energy saving in Win7 to max. performance, which should disable the CPU to throttle down)
- check CPU temperatures for CPU thermal throttling
(although unlikely if installed correctly, missing thermal paste between CPU and cooler is all it takes to run pretty high CPU temps that force it to clock down to prevent overheating. Temps can be seen in CPU-Z as well I believe - thermal throttling under load as well by monitoring clock rates, otherwise the BIOS Hardware Monitor should give sufficient temperature clues as well)
RE: I think your BOINC
)
Yes, I have WIN 7. However I had lot of issues with drivers that are now resolved, I initially kept getting BSOD.
My CPU temps from Core temp max around 57c under max loading. My core speed is 3.7GHZ under load.
Installed all the chipset drivers, as far as I'm aware, there are no drivers for the processor.
Device manager indicates that all the 8 cores in the processor are ok. There aren't any drivers in device manager that indicate an error.
My boinc version is 7.0.64
Also there was a driver called EPU, I didn't instal this. I thought this was something to save power.
All of what you're seeing
)
All of what you're seeing could be completely normal. As others have already said, you've got a time-based setting for how many WUs should cache. If BOINC thinks you're completing WUs much faster now, then it will download more so that your cache will last the same time as before. If you're getting too many WUs reduce cache size.
And your new "8 core CPU" is actually more like a quad core with hyperthreading, speaking in Intel terms. 2 cores in a "module" share some ressources, and most importantly the FPU, which is elemental for Einstein and most other BOINC projects. Hence the performance of one module per clock is less than what you're used from each core of your previous Athlon 64.
MrS
Scanning for our furry friends since Jan 2002
I looked as his CPU tasks and
)
I looked as his CPU tasks and the long-running (pending/finished) as well as all aborted ones are the Gamma-Ray Pulsar Searches.
These take significantly longer on my Systems as well.
Got one Gamma-Ray Pulsar Search done in ~26hrs for example, compared to
- Binary Radio Pulsar Search (~2 hrs)
- Perseus Arm Survey (~2.8 hrs)
- Gravitational Wave S6 Directed Search (~9.3 hrs).
Dependent on how RAM bandwidth sensitive the Client crunching those WorkUnits is, it is also noteworthy that the 8 Cores of his system have to share the bandwidth of his DDR3 Dual Channel setup. That might turn out to be somewhat of a bottleneck, compared to i.e. 4 similar performing cores getting the same bandwidth like my FX-4100.
I would make sure that such a fast 8-Core CPU is using the fastest RAM I can/want to afford, but maybe that's already the case.
Add to that the Duration Correction factor that still has to work out for at least 10 days, those predicted runtimes could occur initially but his achieved actual completion times seem in order. Those predicted runtimes should correct down with every finished & reported workunit then.
My new setup (A10-6700 3.7GHz 4 Cores w/ its on-CPU GPU disabled) using DDR-1866 hasn't turned out sufficient CPU results to make a sound judgement of the impact of its faster RAM (the FX-4100 3.6GHz 4 Cores is running DDR3-1333) - would be a rough estimate at best anyway due to slightly different CPU architecture/caches etc...
But that's one for the patient benchmarking guys to discover. At least it's clear that different WorkUnits doing different searches massively vary in runtimes.
Just one question, is the
)
Just one question, is the amount of credit given constant per calculation? My older CPU will have to do the exact same amount of calculation to be given 1 credit unit as my new CPU?
I heard something a while back in 2010, that E@h was adjusting the amount of credit given than previously, to compensate for a newer generation of CPU's. Is there any truth to this?
Yes, basically it should be
)
Yes, basically it should be when comparing CPUs, but it'll always be just an approximation for a number of reasons (such as synthetic nature of benchmarks which only provide very coarse performance indicators).
If my memory serves me correctly, the accepted formula of Cobblestones / Credits granted for GFlops per time period is more or less given by Berkeley (BOINC developer/maintainer).
This used to be a rather rarely updated and more often intensively discussed guideline for all BOINC based projects to follow within some degree of flexibility.
I remember a few discussions about having more realistic/thorough benchmarks implemented instead of the very old Drystone/Whetstone benchmarks to achieve more realistic/precise results but I believe those suggestions were never adopted.
RE: Just one question, is
)
That's a difficult question, actually. A much easier answer would be this: "The credits per WU are fixed. How fast you can return the WUs will determine your daily credits." But how many calculations your CPU needed to finish a WU actually depends on its architecture. The less efficient, the more it will need (e.g. Pentium 4, or memory bandwidth staved machines like an 8-core CPU with lowly clocked single channel memory).
MrS
Scanning for our furry friends since Jan 2002