Running distributed computing projects lets me participate in science in a meaningful way. I hope as we advance the frontiers of science there will be a tangible benefit to society.
Joseph Weber published an article in Physical Review Letters in 1970 which had the title "Evidence for discovery of gravitational radiation". He was wrong, of course, his resonant mass detectors did not have the necessary sensitivity but Weber's idea, that of placing two or more detectors at great distance and considering only those events which appeared on all detectors, was correct and is still applied by LIGO and Virgo. I published an article by prof.Peter G. Bergmann, a former coworker of Einstein, in the 1970 Mondadori Yearbook of Science and Technology and placed an image of its detectors in the front page. The term "evidence" is difficult to translate and my managing editor titled the caption as "scoperte the onde gravitazionali". We were blasted by prof.Antonino Zichichi, an elementary particle physicist. We were only 45 years in advance So this is why I am running Einstein@home. I was right.
2) OpenCl projects are easy to intercept and therefore easy to improve. I do not have to set up the whole thing. I just grab all of those GPU kernels' source code and do the magic. You should see some 17%-60% improvements with the current beta app and later on with the official one in Einstein. (That's my hobby. And there is a rumor that the magic should bring some good vibrations to ATI cards too, even when I am running NVIDIA and can not test with AMD.)
3) It is a hobby of mine. I used to work in the IT business. The whole of IT is lot more fun as a hobby.
4) In the winter I can make use of the heat. In the summer I have to blow the energy out of my house. Maybe I need a new heatSink.
To help make a small dent in the progress. I only have one GPU but that's something. :)
That not something, that is everything the project ever needs.....it's about the number of people having a GPU and crunching the workunits. Every one of us is just a small light bulb compared to the aggregated computing power. Some brighter than others, but still small in comparison.
You should see some 17%-60% improvements with the current beta app and later on with the official one in Einstein. (That's my hobby.
Oh you are improving the code on the E@H apps? That's quite the talent. I did check that box to test that new version. Runs much better (approx. 45-50% faster on my RTX 2060). That's what I call an upgrade.
You should see some 17%-60% improvements with the current beta app and later on with the official one in Einstein. (That's my hobby.
Oh you are improving the code on the E@H apps? That's quite the talent. I did check that box to test that new version. Runs much better (approx. 45-50% faster on my RTX 2060). That's what I call an upgrade.
yes, petri wrote the code and ive been helping him test different cards and cases and situations. The lions share of the credit goes to petri tho :). He’s a wizard with this stuff.
He nor I are affiliated with the project. We just shared our findings with the project so that they could make the change in their apps for everyone. We have ways to inject our own code real-time, but it’s too problematic to try to instruct a large user base on how to do that. So for everyones benefit and ease of use it’s better for the project to make the update and distribute to everyone.
He nor I are affiliated with the project. We just shared our findings with the project so that they could make the change in their apps for everyone. We have ways to inject our own code real-time, but it’s too problematic to try to instruct a large user base on how to do that. So for everyones benefit and ease of use it’s better for the project to make the update and distribute to everyone.
Good to see that part of the community takes a more active approach. And good that the official staff is recognizing this effort. I mean ,GPUs are the workhorses on the project, and pushing NVidia cards by like 50% more, the impact on the project is measured in +PFlops, if GPU vendor distribution here is similar to non-E@H figures.
yes, petri wrote the code and ive been helping him test different cards and cases and situations. The lions share of the credit goes to petri tho :). He’s a wizard with this stuff.
This is great stuff. Congratulations to both of you. I have a couple of Windows computers with Nvidia GPU's attached to E@H and I updated the driver to the latest. The GTX1070 has picked up the new app but the GTX970 has not, it gets the older beta app. It must be too old to qualify for the new app. The stock app on 970 seems to be about a minute faster with the new driver though.
yes, petri wrote the code and ive been helping him test different cards and cases and situations. The lions share of the credit goes to petri tho :). He’s a wizard with this stuff.
This is great stuff. Congratulations to both of you. I have a couple of Windows computers with Nvidia GPU's attached to E@H and I updated the driver to the latest. The GTX1070 has picked up the new app but the GTX970 has not, it gets the older beta app. It must be too old to qualify for the new app. The stock app on 970 seems to be about a minute faster with the new driver though.
from our testing, the Maxwell and older cards didn’t really see a benefit in the new code. But the gains for Pascal/Turing/Ampere were massive. Ampere cards are over 2x (100%+) faster with the code change.
the project admins have set the new app delivery to be only for pascal and newer. So if it detects a Maxwell card as the primary card, it won’t send it.
Just keep in mind that with the increased performance comes more power draw and heat. But overall efficiency jumps A LOT since the speed increase greatly outweighs the power draw increase.
Running distributed computing
)
Running distributed computing projects lets me participate in science in a meaningful way. I hope as we advance the frontiers of science there will be a tangible benefit to society.
tullio wrote: Joseph Weber
)
A big thumb up! (iso peukku! -- in finnish)
Why do I ...?1) To
)
Why do I ...?
1) To advance the physics science. Any of it.
2) OpenCl projects are easy to intercept and therefore easy to improve. I do not have to set up the whole thing. I just grab all of those GPU kernels' source code and do the magic. You should see some 17%-60% improvements with the current beta app and later on with the official one in Einstein. (That's my hobby. And there is a rumor that the magic should bring some good vibrations to ATI cards too, even when I am running NVIDIA and can not test with AMD.)
3) It is a hobby of mine. I used to work in the IT business. The whole of IT is lot more fun as a hobby.
4) In the winter I can make use of the heat. In the summer I have to blow the energy out of my house. Maybe I need a new heatSink.
--
petri33 - nowadays 53 or something.
To help make a small dent in
)
To help make a small dent in the progress. I only have one GPU but that's something. :)
Guðni Már Gilbert wrote: To
)
That not something, that is everything the project ever needs.....it's about the number of people having a GPU and crunching the workunits. Every one of us is just a small light bulb compared to the aggregated computing power. Some brighter than others, but still small in comparison.
petri33 wrote: You should
)
Oh you are improving the code on the E@H apps? That's quite the talent. I did check that box to test that new version. Runs much better (approx. 45-50% faster on my RTX 2060). That's what I call an upgrade.
Exard3k wrote:petri33
)
yes, petri wrote the code and ive been helping him test different cards and cases and situations. The lions share of the credit goes to petri tho :). He’s a wizard with this stuff.
He nor I are affiliated with the project. We just shared our findings with the project so that they could make the change in their apps for everyone. We have ways to inject our own code real-time, but it’s too problematic to try to instruct a large user base on how to do that. So for everyones benefit and ease of use it’s better for the project to make the update and distribute to everyone.
_________________________________________________________________________
Ian&Steve C. wrote: He nor
)
Good to see that part of the community takes a more active approach. And good that the official staff is recognizing this effort. I mean ,GPUs are the workhorses on the project, and pushing NVidia cards by like 50% more, the impact on the project is measured in +PFlops, if GPU vendor distribution here is similar to non-E@H figures.
Great work!
Ian&Steve C. wrote: yes,
)
This is great stuff. Congratulations to both of you. I have a couple of Windows computers with Nvidia GPU's attached to E@H and I updated the driver to the latest. The GTX1070 has picked up the new app but the GTX970 has not, it gets the older beta app. It must be too old to qualify for the new app. The stock app on 970 seems to be about a minute faster with the new driver though.
Harri Liljeroos
)
from our testing, the Maxwell and older cards didn’t really see a benefit in the new code. But the gains for Pascal/Turing/Ampere were massive. Ampere cards are over 2x (100%+) faster with the code change.
the project admins have set the new app delivery to be only for pascal and newer. So if it detects a Maxwell card as the primary card, it won’t send it.
Just keep in mind that with the increased performance comes more power draw and heat. But overall efficiency jumps A LOT since the speed increase greatly outweighs the power draw increase.
_________________________________________________________________________