I can subscribe that a video card with only 2GB vram is not enough for current WU's.
My GT 1030 with 2GB ram does NOT work.
Its true for NV GPUs.
But for AMD/OpenCL platform actual VRAM usage by current GW O3AS WUs < 1900 MB.
So 4 GB AMD GPUs can run x2 O3AS WUs fine. (usual VRAM usage about 3.7/4.0 GB )
As for the 2 GB AMD GPU, they also can run x1 O3AS.
But provided that it is a dedicated cruncher (there is no use of VRAM by other applications) and it require a small BOINC "hack" to fake the reported amount of VRAM, because otherwise the E@H server simply does not issue tasks to the 2 GB GPUs and requires about 2.5 GB of free VRAM.
"hack" is just edit VRAM values in coproc_info.xml file and disable write access for BOINC to it (to prevent reverting edited values at each BOINC restart).
I have a gt 710 2MB running under Windows 11. It is running 2x Brp7/meerKat. The only reason I am running 2x is because the other two gpus will run 2x faster.
A Proud member of the O.F.A. (Old Farts Association).
It seems like some users are able to run more than 1 GPU task at a time. When I try - it's not worth it, and I tired it with both BRP7 and O3AS apps. Processing times slow down too much. I have a NVIDIA Titan V on Windows 10. Are there some optimizations that can be done? I found that overclocking GPU RAM a bit significantly improves run times but that's the only thing I found that makes any improvement from default.
It seems like some users are able to run more than 1 GPU task at a time. When I try - it's not worth it, and I tired it with both BRP7 and O3AS apps. Processing times slow down too much. I have a NVIDIA Titan V on Windows 10. Are there some optimizations that can be done? I found that overclocking GPU RAM a bit significantly improves run times but that's the only thing I found that makes any improvement from default.
After you ran tasks at 2x, did you divide the runtime by 2 to figure out if it is actually faster? Running a Titan V with that cpu should be better at 2x, even in windows. None of our windows systems run 1x as their ideal state.
It seems like some users are able to run more than 1 GPU task at a time. When I try - it's not worth it, and I tired it with both BRP7 and O3AS apps. Processing times slow down too much. I have a NVIDIA Titan V on Windows 10. Are there some optimizations that can be done? I found that overclocking GPU RAM a bit significantly improves run times but that's the only thing I found that makes any improvement from default.
You may want to compare your system to this Windows and Titan V system. I believe he is running 2x.
I looked at your Tasks. It does look like you are running 1x. As Boca Raton pointed out. To get some idea of how much you are gaining by running 2x you need to divide your runtime at 2x by 2 to compare it to your run time at 1x. As long as the result after you divide by two is lower than the run time you were getting at 1x you are getting a net gain in production.
====edit===
deleted a lot of how to estimate your RAC calculations.
Assuming a 300 second Brp7/meerKat task is the result of 1x then if a 2x task is taking less than 600 seconds you are likely to gain total production.
You may be able to run 1.2Million RAC if you run 2x and let it run steadily for as much as 6 weeks. (No promises though).
Respectfully,
A Proud member of the O.F.A. (Old Farts Association).
It seems like some users are able to run more than 1 GPU task at a time. When I try - it's not worth it, and I tired it with both BRP7 and O3AS apps. Processing times slow down too much. I have a NVIDIA Titan V on Windows 10. Are there some optimizations that can be done? I found that overclocking GPU RAM a bit significantly improves run times but that's the only thing I found that makes any improvement from default.
Maybe I'm somewhat of beat, but I don't have troubles running 3 O3AS tasks on a single Titan V.
Same with Titan Xp.
Running 3 BRP7 tasks on a single GTX750 Ti.
I'm probably not getting the Problem ?
So please help me to understand !
We used to run 3x O3AS on the Titan V and were getting good results. 2x was about the same, negligible difference. For us, our system with the Titan V with the 14900KS is very interesting because this CPU only boosts two cores to 6.2GHz when under ~65C and the cores generate SO much instantaneous heat that any additional core was hindering overall performance (albeit only slightly). I know that others run 2x for different reasons, but that is our reason.
I think we are running 3x on a few systems though with RTX A4500 and A6000, but just wasn't ideal for our system with the Titan V.
Maybe I'm somewhat of beat, but I don't have troubles running 3 O3AS tasks on a single Titan V.
Same with Titan Xp.
Running 3 BRP7 tasks on a single GTX750 Ti.
I'm probably not getting the Problem ?
So please help me to understand !
sfv
There is no reason not to run 3x O3AS on a GPU. Unless 2x gives you the same production, and frees up a CPU thread you can use for CPU crunching. Ditto Brp7/meerKat.
I and Boca Raton think it was a matter of AndreyOR not equating the task times he was getting at 2x with their equivalent under 1x by dividing by two. He/She/They just looked at the raw 2x number and since it was higher than the raw 1x number thought they were losing production.
We don't think that was what was happening.
Tom M
A Proud member of the O.F.A. (Old Farts Association).
In the previous high frequency version of O3AS, I ran my P4000's at 3x, as the CPU portion of the app was a huge bottleneck on my system, so the more staggered the tasks became over time, the more the CPU portion was masked by other tasks, keeping the GPUs going constantly.
Now, the CPU portion is much faster, and I get basically the same performance going from 2x to 3x. So I do indeed use the extra CPU threads to run CPU tasks. It is just unfortunate that the CPU tasks get so much less credit that the extra crunching I get is barely anything extra.
In the previous high frequency version of O3AS, I ran my P4000's at 3x, as the CPU portion of the app was a huge bottleneck on my system, so the more staggered the tasks became over time, the more the CPU portion was masked by other tasks, keeping the GPUs going constantly.
Now, the CPU portion is much faster, and I get basically the same performance going from 2x to 3x. So I do indeed use the extra CPU threads to run CPU tasks. It is just unfortunate that the CPU tasks get so much less credit that the extra crunching I get is barely anything extra.
It seems most users are settling with the new O3AS tasks running at either 2x or 3x. It almost feels like a completely different application.
MPG wrote: I can subscribe
)
Its true for NV GPUs.
But for AMD/OpenCL platform actual VRAM usage by current GW O3AS WUs < 1900 MB.
So 4 GB AMD GPUs can run x2 O3AS WUs fine. (usual VRAM usage about 3.7/4.0 GB )
As for the 2 GB AMD GPU, they also can run x1 O3AS.
But provided that it is a dedicated cruncher (there is no use of VRAM by other applications) and it require a small BOINC "hack" to fake the reported amount of VRAM, because otherwise the E@H server simply does not issue tasks to the 2 GB GPUs and requires about 2.5 GB of free VRAM.
"hack" is just edit VRAM values in coproc_info.xml file and disable write access for BOINC to it (to prevent reverting edited values at each BOINC restart).
I have a gt 710 2MB running
)
I have a gt 710 2MB running under Windows 11. It is running 2x Brp7/meerKat. The only reason I am running 2x is because the other two gpus will run 2x faster.
A Proud member of the O.F.A. (Old Farts Association).
It seems like some users are
)
It seems like some users are able to run more than 1 GPU task at a time. When I try - it's not worth it, and I tired it with both BRP7 and O3AS apps. Processing times slow down too much. I have a NVIDIA Titan V on Windows 10. Are there some optimizations that can be done? I found that overclocking GPU RAM a bit significantly improves run times but that's the only thing I found that makes any improvement from default.
AndreyOR wrote: It seems
)
After you ran tasks at 2x, did you divide the runtime by 2 to figure out if it is actually faster? Running a Titan V with that cpu should be better at 2x, even in windows. None of our windows systems run 1x as their ideal state.
AndreyOR wrote:It seems
)
You may want to compare your system to this Windows and Titan V system. I believe he is running 2x.
I looked at your Tasks. It does look like you are running 1x. As Boca Raton pointed out. To get some idea of how much you are gaining by running 2x you need to divide your runtime at 2x by 2 to compare it to your run time at 1x. As long as the result after you divide by two is lower than the run time you were getting at 1x you are getting a net gain in production.
====edit===
deleted a lot of how to estimate your RAC calculations.
Assuming a 300 second Brp7/meerKat task is the result of 1x then if a 2x task is taking less than 600 seconds you are likely to gain total production.
You may be able to run 1.2Million RAC if you run 2x and let it run steadily for as much as 6 weeks. (No promises though).
Respectfully,
A Proud member of the O.F.A. (Old Farts Association).
AndreyOR wrote: It seems
)
Maybe I'm somewhat of beat, but I don't have troubles running 3 O3AS tasks on a single Titan V.
Same with Titan Xp.
Running 3 BRP7 tasks on a single GTX750 Ti.
I'm probably not getting the Problem ?
So please help me to understand !
sfv
We used to run 3x O3AS on the
)
We used to run 3x O3AS on the Titan V and were getting good results. 2x was about the same, negligible difference. For us, our system with the Titan V with the 14900KS is very interesting because this CPU only boosts two cores to 6.2GHz when under ~65C and the cores generate SO much instantaneous heat that any additional core was hindering overall performance (albeit only slightly). I know that others run 2x for different reasons, but that is our reason.
I think we are running 3x on a few systems though with RTX A4500 and A6000, but just wasn't ideal for our system with the Titan V.
San-Fernando-Valley
)
There is no reason not to run 3x O3AS on a GPU. Unless 2x gives you the same production, and frees up a CPU thread you can use for CPU crunching. Ditto Brp7/meerKat.
I and Boca Raton think it was a matter of AndreyOR not equating the task times he was getting at 2x with their equivalent under 1x by dividing by two. He/She/They just looked at the raw 2x number and since it was higher than the raw 1x number thought they were losing production.
We don't think that was what was happening.
Tom M
A Proud member of the O.F.A. (Old Farts Association).
In the previous high
)
In the previous high frequency version of O3AS, I ran my P4000's at 3x, as the CPU portion of the app was a huge bottleneck on my system, so the more staggered the tasks became over time, the more the CPU portion was masked by other tasks, keeping the GPUs going constantly.
Now, the CPU portion is much faster, and I get basically the same performance going from 2x to 3x. So I do indeed use the extra CPU threads to run CPU tasks. It is just unfortunate that the CPU tasks get so much less credit that the extra crunching I get is barely anything extra.
taketwicedailey wrote: In
)
It seems most users are settling with the new O3AS tasks running at either 2x or 3x. It almost feels like a completely different application.