Well, I put 10 days for connection and 10 days for work cache... and still Einstein is only asking work for the GPU and waiting 4 hours thereafter. This is... a little frustrating.
Edit: Never mind, BOINC finally asked for Einstein CPU WUs... but only after it managed to scrounge one SETI GPU WU. For the record, this is BOINC 6.6.36, the latest stable release.
I've given up. To get Einstein CPU WUs, I have to leave the GPU information in the app_info.xml and just put up with the repeated errors on the GPU WUs. As long as there are no GPU WUs across all projects, the scheduler will always ask for GPU WUs first. ):
I've given up. To get Einstein CPU WUs, I have to leave the GPU information in the app_info.xml and just put up with the repeated errors on the GPU WUs. As long as there are no GPU WUs across all projects, the scheduler will always ask for GPU WUs first. ):
If your CUDA card doesn't have enough memory to run this app, wouldn't it be better to devote your testing efforts to the parallel CPU-only v3.09 test, and relieve the servers of all that wasted download pressure?
You misunderstand. That's what I've been trying to do. 3.09 works great and seems to be a substantial improvement over 3.08, so that's why I've been trying to run it instead of just removing the app_info.xml altogether as the Einstein server keeps yelling at me to do. But no matter what I try, BOINC insists on giving priority to requesting GPU WUs before CPU ones. I strongly recommend releasing 3.09, it seems to be thoroughly tested now.
You misunderstand. That's what I've been trying to do ....
Maybe I'm missing something but if you really are trying to get CPU tasks rather than GPU tasks, why don't you ditch the GPU app_info.xml and set up the CPU one instead? You said earlier that you "have to leave the GPU information in the app_info.xml". I'm sorry but this is quite wrong - you certainly don't have to do that. You should re-read Richard's advice because he doesn't misunderstand and he's suggesting exactly what you need to do.
Here are some important points for you to consider:-
* Lots of people on many projects are complaining about scheduling problems in 6.6.36.
* If you don't want GPU capability, something like 6.4.7 or 6.5.0 should be much better for you.
* Setting 10/10 for your cache sizes is likely to worsen the already bad behaviour of 6.6.36. If you are 'always on' use something like 0/1 to 0/4.
* The EAH project has some sort of server bug that causes the scheduler to say randomly that it has no tasks available even if it has just given you some. With S5R5 this causes a 1 min backoff before a retry is done. With ABP1, the backoff is 4 hours. You don't have to endure the 4 hour backoff. Just wait 1 min and then force a retry with the 'update' button. Once you have sufficient tasks on board, the backoff shouldn't bother you any further.
If you need help with transitioning to the CPU only app_info.xml just ask. If you are not careful you can trash your current cache. If you have no EAH GPU tasks on board, you should be able to stop BOINC, uninstall BOINC, copy the CPU app_info.xml and the CPU ABP1 binaries over the top of the existing ones, overwriting existing files if the names are the same and then install a more suitable version of BOINC, as suggested above, to replace 6.6.36. You can get any particular version of BOINC you fancy from here.
First of all, I meant no offence. I know this is all volunteer-run.
Quote:
Maybe I'm missing something but if you really are trying to get CPU tasks rather than GPU tasks, why don't you ditch the GPU app_info.xml and set up the CPU one instead?
Because I had already set up a CPU-only app_info.xml previously. It appears that BOINC detects the presence of a GPU on this particular host and will request GPU units regardless of the fact that there is no GPU application included in the app_info.xml. And it is because of the fact that there are no GPU applications in the app_info.xml that the Einstein server responds with the "remove app_info.xml" message and will then set a four-hour delay for the next communication with it.
Quote:
You said earlier that you "have to leave the GPU information in the app_info.xml". I'm sorry but this is quite wrong - you certainly don't have to do that.
I agree! I shouldn't! But the fact is that - on this particular host, at least - until there is some GPU work on the system already, BOINC refuses to request CPU work from the Einstein server.
Quote:
Lots of people on many projects are complaining about scheduling problems in 6.6.36.
Well... I wasn't aware of this, but after my recent experience, I can see why.
Quote:
If you don't want GPU capability, something like 6.4.7 or 6.5.0 should be much better for you.
The problem is that I do. SETI GPU units work fine on this host. The unfortunate thing is that recent WU shortages from SETI means that there is no GPU work available. Hence BOINC asking Einstein for some, regardless of the fact that there is no corresponding GPU application in the app_info.xml.
Quote:
Setting 10/10 for your cache sizes is likely to worsen the already bad behaviour of 6.6.36. If you are 'always on' use something like 0/1 to 0/4.
Again, I agree. I was simply following Stranger7777's suggestion from earlier. My default cache is 0.5/0 and I already expressed my reluctance to use such an unreasonably large cache size previously.
Quote:
The EAH project has some sort of server bug that causes the scheduler to say randomly that it has no tasks available even if it has just given you some. With S5R5 this causes a 1 min backoff before a retry is done. With ABP1, the backoff is 4 hours. You don't have to endure the 4 hour backoff. Just wait 1 min and then force a retry with the 'update' button. Once you have sufficient tasks on board, the backoff shouldn't bother you any further.
I am aware that one can force a server request with the update button. Unfortunately, this is not helpful when BOINC keeps requesting GPU units.
In any case, hopefully my woes will all become moot shortly. I'm waiting on a replacement video card to come in which will surely be sufficient for processing the Einstein GPU WUs.
* If you don't want GPU capability, something like 6.4.7 or 6.5.0 should be much better for you.
Hi Gary,
Yust for information. I run GPU WU's (CUDA) with 6.4.7 and it works fine. Jobs take about 4 hours with a "cold" graphics card and all are validated, some still pending.
First of all, I meant no offence. I know this is all volunteer-run.
No offense was taken - sorry if my reply implied offense.
Quote:
Quote:
Maybe I'm missing something ...
Because I had already set up a CPU-only app_info.xml previously....
Ahhh, OK. I had read your previous messages and had interpreted the statements about BOINC requesting GPU work in preference to CPU work as meaning that you were setup for both. I didn't notice any statement about having a CPU only app_info.xml so I thought you hadn't made the change back to CPU only - which was why I asked the question/made the comment.
If using the AP mechanism, my previous experience (on older BOINCs) says that BOINC wont ask for work for an app that is not in app_info.xml. If more recent BOINCs do ask, this is unfortunate as it removes a degree of control from the user. It would be interesting to know if you reverted to 6.4.7 or 6.5.0, would you be able to get E@H CPU work through AP (without BOINC asking for GPU work) whilst still getting Seti CUDA work for your low memory GPU?
Quote:
In any case, hopefully my woes will all become moot shortly. I'm waiting on a replacement video card to come in which will surely be sufficient for processing the Einstein GPU WUs.
Even with your new video card, you may still have issues with 6.6.36. You may find you can better control your mix of projects/apps by using 6.4.7 or 6.5.0. I've seen many people reporting good behaviour with those versions.
Well, I put 10 days for
)
Well, I put 10 days for connection and 10 days for work cache... and still Einstein is only asking work for the GPU and waiting 4 hours thereafter. This is... a little frustrating.
Edit: Never mind, BOINC finally asked for Einstein CPU WUs... but only after it managed to scrounge one SETI GPU WU. For the record, this is BOINC 6.6.36, the latest stable release.
Soli Deo Gloria
RE: ...For the record, this
)
Well, it is the recommended version; that doesn't mean it's stable or bug free ;-)
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
I've given up. To get
)
I've given up. To get Einstein CPU WUs, I have to leave the GPU information in the app_info.xml and just put up with the repeated errors on the GPU WUs. As long as there are no GPU WUs across all projects, the scheduler will always ask for GPU WUs first. ):
Soli Deo Gloria
RE: I've given up. To get
)
If your CUDA card doesn't have enough memory to run this app, wouldn't it be better to devote your testing efforts to the parallel CPU-only v3.09 test, and relieve the servers of all that wasted download pressure?
You misunderstand. That's
)
You misunderstand. That's what I've been trying to do. 3.09 works great and seems to be a substantial improvement over 3.08, so that's why I've been trying to run it instead of just removing the app_info.xml altogether as the Einstein server keeps yelling at me to do. But no matter what I try, BOINC insists on giving priority to requesting GPU WUs before CPU ones. I strongly recommend releasing 3.09, it seems to be thoroughly tested now.
Soli Deo Gloria
RE: You misunderstand.
)
Maybe I'm missing something but if you really are trying to get CPU tasks rather than GPU tasks, why don't you ditch the GPU app_info.xml and set up the CPU one instead? You said earlier that you "have to leave the GPU information in the app_info.xml". I'm sorry but this is quite wrong - you certainly don't have to do that. You should re-read Richard's advice because he doesn't misunderstand and he's suggesting exactly what you need to do.
Here are some important points for you to consider:-
* If you don't want GPU capability, something like 6.4.7 or 6.5.0 should be much better for you.
* Setting 10/10 for your cache sizes is likely to worsen the already bad behaviour of 6.6.36. If you are 'always on' use something like 0/1 to 0/4.
* The EAH project has some sort of server bug that causes the scheduler to say randomly that it has no tasks available even if it has just given you some. With S5R5 this causes a 1 min backoff before a retry is done. With ABP1, the backoff is 4 hours. You don't have to endure the 4 hour backoff. Just wait 1 min and then force a retry with the 'update' button. Once you have sufficient tasks on board, the backoff shouldn't bother you any further.
If you need help with transitioning to the CPU only app_info.xml just ask. If you are not careful you can trash your current cache. If you have no EAH GPU tasks on board, you should be able to stop BOINC, uninstall BOINC, copy the CPU app_info.xml and the CPU ABP1 binaries over the top of the existing ones, overwriting existing files if the names are the same and then install a more suitable version of BOINC, as suggested above, to replace 6.6.36. You can get any particular version of BOINC you fancy from here.
Cheers,
Gary.
First of all, I meant no
)
First of all, I meant no offence. I know this is all volunteer-run.
Because I had already set up a CPU-only app_info.xml previously. It appears that BOINC detects the presence of a GPU on this particular host and will request GPU units regardless of the fact that there is no GPU application included in the app_info.xml. And it is because of the fact that there are no GPU applications in the app_info.xml that the Einstein server responds with the "remove app_info.xml" message and will then set a four-hour delay for the next communication with it.
I agree! I shouldn't! But the fact is that - on this particular host, at least - until there is some GPU work on the system already, BOINC refuses to request CPU work from the Einstein server.
Well... I wasn't aware of this, but after my recent experience, I can see why.
The problem is that I do. SETI GPU units work fine on this host. The unfortunate thing is that recent WU shortages from SETI means that there is no GPU work available. Hence BOINC asking Einstein for some, regardless of the fact that there is no corresponding GPU application in the app_info.xml.
Again, I agree. I was simply following Stranger7777's suggestion from earlier. My default cache is 0.5/0 and I already expressed my reluctance to use such an unreasonably large cache size previously.
I am aware that one can force a server request with the update button. Unfortunately, this is not helpful when BOINC keeps requesting GPU units.
In any case, hopefully my woes will all become moot shortly. I'm waiting on a replacement video card to come in which will surely be sufficient for processing the Einstein GPU WUs.
Soli Deo Gloria
RE: RE: * If you don't
)
Greetings from
TJ
And how is the the 3.10
)
And how is the the 3.10 doing.
Any better then the 80%. in 3.09
As in 100% with GPU the CPU would be @ 80%.
RE: First of all, I meant
)
No offense was taken - sorry if my reply implied offense.
Ahhh, OK. I had read your previous messages and had interpreted the statements about BOINC requesting GPU work in preference to CPU work as meaning that you were setup for both. I didn't notice any statement about having a CPU only app_info.xml so I thought you hadn't made the change back to CPU only - which was why I asked the question/made the comment.
If using the AP mechanism, my previous experience (on older BOINCs) says that BOINC wont ask for work for an app that is not in app_info.xml. If more recent BOINCs do ask, this is unfortunate as it removes a degree of control from the user. It would be interesting to know if you reverted to 6.4.7 or 6.5.0, would you be able to get E@H CPU work through AP (without BOINC asking for GPU work) whilst still getting Seti CUDA work for your low memory GPU?
Even with your new video card, you may still have issues with 6.6.36. You may find you can better control your mix of projects/apps by using 6.4.7 or 6.5.0. I've seen many people reporting good behaviour with those versions.
Cheers,
Gary.