Is there any chance that you can lower again to 450MB?
If you ask whether our app's memory requirements might be reduced to 450 MB at some point, then probably since we always strive to support as many cards as possible by keeping our requirements as low as possible. Right now we can't due to other technical constraints.
Is there a chance, that CPU-load for GPU-WUs could be lowered in the future?
Yes, we have some ideas for that in the pipeline, it's just a matter of time to implement and test them. However, keep in mind that not all algorithms can be ported 100% to the GPU. This really depends on the problem and our current implementation is already not that bad after all.
Right now we still need to focus on getting a first stable and numerically correct release out of the door -- and apart from our own code we largely depend on BOINC's readiness WRT OpenCL and proper AMD drivers, both are not easy to come by.
Flagging a weird one.
I caught this one running at zero GPU load for more than an hour, before I aborted it. http://albertathome.org/task/107592
It had been running for well over 4 hours and had an estimated time of another 4 hours, was at 41% done. Its progress was increasing. But still, I didn't trust it.
OK, I ended up aborting all my work in cache, after I found a couple of things.
a) For some unknown reason running an Albert task on my ATI GPU would make my internet connection stop working, leaving me with loads upon loads of page loading faults, even between same forum pages. Disable the Albert task and the internet would fly.
b) Checking in GPU-Z I found that all these tasks would start with some GPU load before it tapering off to no load at all, or only a small bump every 5 minutes or so. I've seen this happen before with Albert GPU tasks. A reboot of the system didn't help in this matter.
I'll test again after I found some I/O recording and debugging programs, to see if I can show you what I see.
Hi,
yesterday I installed my new system with two ATI-GPU's, latest BM 7.0.11 , latest AMD drivers and an outdated nVidia GPU (9500).
I've seen that the GPU usage is very very low, so it makes sense to run more than one wu on every GPU. Is there an app_info available for that purpose?
Hi,
yesterday I installed my new system with two ATI-GPU's, latest BM 7.0.11 , latest AMD drivers and an outdated nVidia GPU (9500).
I've seen that the GPU usage is very very low, so it makes sense to run more than one wu on every GPU. Is there an app_info available for that purpose?
We don't provide an app_info.xml but I'm sure people will help you prepare one yourself. Also, BOINC 7.0.12 is the latest and the GPU usage is usually between 60-80%. However, this is only the case when there is enough CPU power left to feed the GPU (see previous postings).
I checked the gpu usage over a longer period now; there are longer gaps with a usage between 0 and 3% (which caused me to say 'very very low'), shorter periods with ~35% and long times with a usage between 65 and 75%.
This is the on-chip GPU of my FM1 Cpu.
I got the whole system (Cpu, Mainboard and Ram) for ~250.- Euro.
A good GPU costs about the same price and I don't want to run that one at ~30% as on my HD5830, this is a wast of resources and its expensive.
Even for my on-chip gpu I calculate:
1 wu in ~ 7 hrs @ 65% average
2 wu's in ~ 11 hrs @ 95% average --> better performance. :-))
So if anyone can compile an app_info , PLEASE HELP!
Alexander
PS: intresting, the next wu is after 1:30:06 at 56%, quite faster than the first ones!
BTW, the first two completed and validated, thats much better than the CPU wu's ...
RE: OK, I found a hint in
)
FWIW, you'll find available BOINC packages here: http://boinc.berkeley.edu/dl
Oliver
PS: 7.0.12 is out
RE: Is there any chance
)
If you ask whether our app's memory requirements might be reduced to 450 MB at some point, then probably since we always strive to support as many cards as possible by keeping our requirements as low as possible. Right now we can't due to other technical constraints.
Oliver
RE: Is there a chance,
)
Yes, we have some ideas for that in the pipeline, it's just a matter of time to implement and test them. However, keep in mind that not all algorithms can be ported 100% to the GPU. This really depends on the problem and our current implementation is already not that bad after all.
Right now we still need to focus on getting a first stable and numerically correct release out of the door -- and apart from our own code we largely depend on BOINC's readiness WRT OpenCL and proper AMD drivers, both are not easy to come by.
Best
Oliver
Flagging a weird one. I
)
Flagging a weird one.
I caught this one running at zero GPU load for more than an hour, before I aborted it. http://albertathome.org/task/107592
It had been running for well over 4 hours and had an estimated time of another 4 hours, was at 41% done. Its progress was increasing. But still, I didn't trust it.
OK, I ended up aborting all
)
OK, I ended up aborting all my work in cache, after I found a couple of things.
a) For some unknown reason running an Albert task on my ATI GPU would make my internet connection stop working, leaving me with loads upon loads of page loading faults, even between same forum pages. Disable the Albert task and the internet would fly.
b) Checking in GPU-Z I found that all these tasks would start with some GPU load before it tapering off to no load at all, or only a small bump every 5 minutes or so. I've seen this happen before with Albert GPU tasks. A reboot of the system didn't help in this matter.
I'll test again after I found some I/O recording and debugging programs, to see if I can show you what I see.
Hi, yesterday I installed my
)
Hi,
yesterday I installed my new system with two ATI-GPU's, latest BM 7.0.11 , latest AMD drivers and an outdated nVidia GPU (9500).
I've seen that the GPU usage is very very low, so it makes sense to run more than one wu on every GPU. Is there an app_info available for that purpose?
Alexander
Hi Alex, RE: Hi, yesterday
)
Hi Alex,
We don't provide an app_info.xml but I'm sure people will help you prepare one yourself. Also, BOINC 7.0.12 is the latest and the GPU usage is usually between 60-80%. However, this is only the case when there is enough CPU power left to feed the GPU (see previous postings).
Best,
Oliver
Hi Oliver, Thx for your
)
Hi Oliver,
Thx for your reply.
I checked the gpu usage over a longer period now; there are longer gaps with a usage between 0 and 3% (which caused me to say 'very very low'), shorter periods with ~35% and long times with a usage between 65 and 75%.
This is the on-chip GPU of my FM1 Cpu.
I got the whole system (Cpu, Mainboard and Ram) for ~250.- Euro.
A good GPU costs about the same price and I don't want to run that one at ~30% as on my HD5830, this is a wast of resources and its expensive.
Even for my on-chip gpu I calculate:
1 wu in ~ 7 hrs @ 65% average
2 wu's in ~ 11 hrs @ 95% average --> better performance. :-))
So if anyone can compile an app_info , PLEASE HELP!
Alexander
PS: intresting, the next wu is after 1:30:06 at 56%, quite faster than the first ones!
BTW, the first two completed and validated, thats much better than the CPU wu's ...
RE: This is the on-chip
)
Now that's interesting. Can you post a link to that box? I'd like to take a look at the specs! Is this using an AMD APU or similar?
Good news :-)
Oliver
PS: we identified and fixed an issue that was causing a number of validation issues. v1.20 will be released short and should improve things.
RE: RE: This is the
)
pls check your PM