Thank you Mike, that sounds probable, given that they can guesstimate based on the specs what a machine can handle (or not). I will dig a little deeper on the topic and leave my machine digesting whatever tasks it gets.
Another thought. There is a good reason why machine abilities and RAC may not correlate : user preferences for time usage fractions ie. usage limits. You could have a red hot machine but only allowed a small percentage of time to BOINC, thus have a low RAC. Now if RAC and machine capability do not correlate then you can't substitute one with the other in the logic about scheduling.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
I was under the impression that Binary Radio Pulsar Search (Arecibo, large) WUs had eight tasks within, where each of those is awarded 500 credits. So the total is 4000. But I could be wrong.
Indeed you are! :-) Just check Bernd's announcement in Tech News.
Before that announcement, BRP4 tasks were awarded 62.5 credits. Arecibo large ( a bundle of 8 tasks - now called BRP4G - confusingly) is therefore being awarded 8 x 62.5 = 500. These are for CPUs and Intel GPUs.
For more powerful discrete GPUs, there will be something called BRP7 (or maybe BRP7G if they want to highlight GPU use) which is probably going to be larger again, although Bernd didn't give any further details about that.
Way back in the day when there was something called FGRPB1 ( I think, but that was long, long ago so don't hold me to it) for CPUs, the credit award was 693 - the same as the current FGRP5 for CPUs. When a GPU app came along, the search using GPUs was called FGRPB1G and this was again a bundle - of 5 FGRPB1 tasks - so the credit award was 5 x 693 = 3465.
The FGRPB1G app turned out to be perfect for the parallel capabilities of GPUs and there was little CPU involvement so the speed improvement was quite dramatic - much more so, than the CPU -> GPU app transition for GW searches where there is no bundling of CPU tasks to create GPU tasks and a lot more CPU involvement. Both GW task types get The same credit award (I think). It's a long time since I've run CPU tasks so I could be wrong there.
Thanks for the extra info. And just to show you why I was a bit surprised, this is what my credit has been doing in recent days:
That trend until June 29 is what it has been since I restarted in Nov last year. Then, since the 29th I have suddenly been getting a lot of GPU-based tasks and then that explosion happened. In the last 2 weeks I earned almost 50% of the credits I have earned to date. Quite the acceleration.
Or as Maverick would say : we've gone ballistic. This will only improve yet again, as I had a bunch of errors on 6th July ( hardware problem, probably sorted now ) so the RAC/slope will rise further.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Thanks for the extra info. And just to show you why I was a bit surprised, this is what my credit has been doing in recent days:
That trend until June 29 is what it has been since I restarted in Nov last year. Then, since the 29th I have suddenly been getting a lot of GPU-based tasks and then that explosion happened. In the last 2 weeks I earned almost 50% of the credits I have earned to date. Quite the acceleration.
I'm going to guess that you are running your system with default preferences, which basically accepts any kind of GPU work. when this is the case, and both task types (gravitational wave and gamma ray) are available, you will primarily get GW work due to the idiosyncrasies of defined FLOPS for each task type and how both BOINC client and how the project scheduler assign priority to task types. so I will further suppose that you were previously primarily running GW tasks.
GW work ran out of work around the same sime that you see the dramatic increase in credit. this is because with the shift from GW to GR, you are now getting about 10x more credit per unit time with the FGRPB1G tasks. so the rate in which you are accumulating credit is much faster than before.
"I'm going to guess that you are running your system with default preferences"
I have been indeed. Mayhaps I misunderstood the premise, but I never wanted to tinker as my idea is: I do this for science and the tasks I am getting are the ones that need to be done. As to the tasks I got, I am not 100% sure, but GW sounds right, mixed with the Pulsar ones from Arecibo. Unfortunately my history doesnt go back to before June 29th anymore. And now I am indeed getting boatloads os the GR ones.
Thanks!
"I'm going to guess that you are running your system with default preferences"
I have been indeed. Mayhaps I misunderstood the premise, but I never wanted to tinker as my idea is: I do this for science and the tasks I am getting are the ones that need to be done. As to the tasks I got, I am not 100% sure, but GW sounds right, mixed with the Pulsar ones from Arecibo. Unfortunately my history doesnt go back to before June 29th anymore. And now I am indeed getting boatloads os the GR ones.
Thanks!
Yes you are currently running:
Gamma-ray pulsar search #5 v1.08 () windows_intelx86--gives 693 credits for each task
Binary Radio Pulsar Search (Arecibo, large) v1.33 () windows_x86_64--gives 500 credits per task
Gamma-ray pulsar binary search #1 on GPUs v1.28 () windows_x86_64-gives 3465 credits per task
The GR ones are very fast so yes you should get "boatloads" of them, if you aren't already you could probably run more than one of the GR tasks at one time so your Avg Credits would go up even faster as that Nvidia 3070 has alot of onboard memory and power.
Thank you Mike, that sounds
)
Thank you Mike, that sounds probable, given that they can guesstimate based on the specs what a machine can handle (or not). I will dig a little deeper on the topic and leave my machine digesting whatever tasks it gets.
E pluribus unum
Another thought. There is a
)
Another thought. There is a good reason why machine abilities and RAC may not correlate : user preferences for time usage fractions ie. usage limits. You could have a red hot machine but only allowed a small percentage of time to BOINC, thus have a low RAC. Now if RAC and machine capability do not correlate then you can't substitute one with the other in the logic about scheduling.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Mike Hewson wrote:I was under
)
Indeed you are! :-) Just check Bernd's announcement in Tech News.
Before that announcement, BRP4 tasks were awarded 62.5 credits. Arecibo large ( a bundle of 8 tasks - now called BRP4G - confusingly) is therefore being awarded 8 x 62.5 = 500. These are for CPUs and Intel GPUs.
For more powerful discrete GPUs, there will be something called BRP7 (or maybe BRP7G if they want to highlight GPU use) which is probably going to be larger again, although Bernd didn't give any further details about that.
Way back in the day when there was something called FGRPB1 ( I think, but that was long, long ago so don't hold me to it) for CPUs, the credit award was 693 - the same as the current FGRP5 for CPUs. When a GPU app came along, the search using GPUs was called FGRPB1G and this was again a bundle - of 5 FGRPB1 tasks - so the credit award was 5 x 693 = 3465.
The FGRPB1G app turned out to be perfect for the parallel capabilities of GPUs and there was little CPU involvement so the speed improvement was quite dramatic - much more so, than the CPU -> GPU app transition for GW searches where there is no bundling of CPU tasks to create GPU tasks and a lot more CPU involvement. Both GW task types get The same credit award (I think). It's a long time since I've run CPU tasks so I could be wrong there.
Cheers,
Gary.
Thanks for the extra info.
)
Thanks for the extra info. And just to show you why I was a bit surprised, this is what my credit has been doing in recent days:
That trend until June 29 is what it has been since I restarted in Nov last year. Then, since the 29th I have suddenly been getting a lot of GPU-based tasks and then that explosion happened. In the last 2 weeks I earned almost 50% of the credits I have earned to date. Quite the acceleration.
E pluribus unum
Me too :Or as
)
Thus also :
Or as Maverick would say : we've gone ballistic. This will only improve yet again, as I had a bunch of errors on 6th July ( hardware problem, probably sorted now ) so the RAC/slope will rise further.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Ah, a similar jump, nice
)
Ah, a similar jump, nice going! Looking forward to see what it will do once the fixed hardware issue becomes visible in the trend.
E pluribus unum
Jinkei wrote:Thanks for the
)
I'm going to guess that you are running your system with default preferences, which basically accepts any kind of GPU work. when this is the case, and both task types (gravitational wave and gamma ray) are available, you will primarily get GW work due to the idiosyncrasies of defined FLOPS for each task type and how both BOINC client and how the project scheduler assign priority to task types. so I will further suppose that you were previously primarily running GW tasks.
GW work ran out of work around the same sime that you see the dramatic increase in credit. this is because with the shift from GW to GR, you are now getting about 10x more credit per unit time with the FGRPB1G tasks. so the rate in which you are accumulating credit is much faster than before.
_________________________________________________________________________
"I'm going to guess that you
)
"I'm going to guess that you are running your system with default preferences"
I have been indeed. Mayhaps I misunderstood the premise, but I never wanted to tinker as my idea is: I do this for science and the tasks I am getting are the ones that need to be done. As to the tasks I got, I am not 100% sure, but GW sounds right, mixed with the Pulsar ones from Arecibo. Unfortunately my history doesnt go back to before June 29th anymore. And now I am indeed getting boatloads os the GR ones.
Thanks!
E pluribus unum
Jinkei wrote: "I'm going to
)
Yes you are currently running:
Gamma-ray pulsar search #5 v1.08 () windows_intelx86--gives 693 credits for each task
Binary Radio Pulsar Search (Arecibo, large) v1.33 () windows_x86_64--gives 500 credits per task
Gamma-ray pulsar binary search #1 on GPUs v1.28 () windows_x86_64-gives 3465 credits per task
The GR ones are very fast so yes you should get "boatloads" of them, if you aren't already you could probably run more than one of the GR tasks at one time so your Avg Credits would go up even faster as that Nvidia 3070 has alot of onboard memory and power.