At this rate, I MAY NO LONGER be able to support 15 Hours of crunching a day! Electric Bills are hitting $500 and $600 a month here, AND I'M NOT PAYING THE BILL; DAD IS!!! I have to justify my activities with the Bill Payer!
I stopped crunching SETI due to the CreditNew issues there, and moved here for the Higher RAC to substantiate the work that is done on my three cards. I LOVE contributing; but, MUST be able to justify the work.
Something NEEDS MUST CHANGE! Or, am I alone in my feelings?????
Don't get your Dad hung up on the RAC. Point out the scientific importance, which is much greater than SETI in my opinion. This is the Golden Age of gravity waves. They have just been discovered (or proven to exist might be a better term), and whatever we find now is going into the science history books. It is all unexplored territory. Ask you Dad if he wants to help fund the Magellan expedition, or help Christopher Columbus discover the New World.
And put the PC where you need the heat anyway. Dads are very practical.
Christian will probably revisit and re-adjust credit when he's back at his desk tomorrow.
NVidia issues should get better with a CUDA version of the app, which we will work on.
This application (version) is as new to us as it is to you. Thank you very much for your contribution so far, and for your patience.
Happy new year to all!
Thank you Bernd for listening. Believe me, I'm NOT complaining about RAC for RAC's sake... I'm NOT, (and NEVER have been), a credit monger. I just need to justify my activities to my dad; who pays the bills.
Something that people here may not know about me and my living situation. I'm on SSDI for Bipolar Disorder. I CANNOT hold a regular full time job. The stresses associated with a full time job where working 40+, (typically 50+ Hours, now), in the IT Field where my talents lie, exacerbates the Bipolar Disorder and causes Manic episodes. (Usually fits of anger and anxiety.) Prior to diagnosis, I went from job to job, being "dismissed" without explanation. This, too, aggravated the Bipolar Disorder; because I didn't know WHY I was being dismissed and unable to hold a job for more than a few months.
I was diagnosed in late 2007, went immediately on medication. The medication seems to keep moods fairly well in check; BUT, I still CANNOT hold a regular job. SSDI ONLY yields $1059 per month, (effective Jan 1, 2017), and almost half of that income goes to medical insurances and dental and vision insurances. The rest goes to bills. This means I CANNOT afford rent outside of my parents' household. I DEFINITELY cannot contribute to the electric and other utility bills of the house. This makes crunching a "LUXURY" in my dad's eyes.
Crunching started as a hobby for me in 2006. I immediately joined the ranks of crunchers working on SETI and here at Einstein, and a few smaller backup projects. As time went on, I dropped the backup projects. SETI became my primary project, (until recently), and Einstein was secondary, (until recently), and I thankfully have been able to crunch 15 Hours a Day until seeing these HIGH and EXORBITANT electric bills from PG&E. (Pacific Gas and Electric.)
Dad refuses to go solar; so, there's no offset to these bills - EXCEPT the inevitability that I'm going to have to reduce crunching times due to the drop in RAC. So, I'm happy to hold on and will continue crunching at the level I'm contributing to here; now that Bernd states that credit levels will be re-examined.
Again, I LOVE this project. I want to contribute the MAXIMUM that I can. Right now, two EVGA GTX-750TI SC cards, and one EVGA GTX-760 card, are all that I can afford to utilize at this time. The MAC/Hackintosh system hardware is circa 2008/2009; EXCEPT for the 750TI SC cards which one is almost three years old, and the other is six months old. The Win XP Pro x64 system with GTX-760 is three and a half years old.
In close, I don't expect the galaxy when it comes to granted credit for work done. I do expect fairer credit for work done. If I can hover in the 90K to 103K range as I was doing a month ago, this is the minimum level I can pass off to my dad as "acceptable" for 15 Hours a day of crunching. With BRP6 dead and gone, I know I'll never get back to the 131K RAC that I had at that time... I don't expect that to happen again. If RAC doesn't get back to 90-103K RAC range, I, (unfortunately), will have to cut back on crunching times. I hope you all understand where I'm coming from.
Sorry to read your difficulties and I wish you a better health for this new year.
@BERND,
Credits are a recurrent discussion on all forums. I didn't read all of them, but I simply don't understand why a working unit is not paid a fixed amount of credit depending on the estimated gflops. This should avoid frauds and reflect hardware improvements.
Sorry to read your difficulties and I wish you a better health for this new year.
@BERND,
Credits are a recurrent discussion on all forums. I didn't read all of them, but I simply don't understand why a working unit is not paid a fixed amount of credit depending on the estimated gflops. This should avoid frauds and reflect hardware improvements.
Thanks for the kind words. btw: The Drupal system code here "CAP LOCKS" my Username... Actually, as originally entered by me on the old site, my Username is "TimeLord04"; but, can be abbreviated to "TL".
Also, (for my posts), I'm having to Double Space my paragraphs... Every time I try to Single Space my paragraphs, they end up running altogether WITHOUT any spaces between paragraphs... (Frustrating!!! DAMNED DRUPAL SYSTEM!!!)
[Update:] Day isn't even finished; and I'm down ANOTHER 3K RAC!!! Total combined RAC from both systems now at 70K...
I hope Christian can find a happier medium for everyone.
On the MAC; "Invalid" Reports are one 1.14 Unit. "Pending" Reports are one 1.14, three 1.17 Beta, and nine 1.17 standard Units. All the rest of my work is Valid and getting the 1365 Credits per Unit... So, I just can't understand this MASSIVE drop in RAC since 1.15 Units came out... ESPECIALLY when crunching LONGER on the 1.17 Units than was EVER done on BRP6 and BRP4G.
Still crunching two Units at a time per card. Six Total Units across three GPUs at a time. MAC is taking 2+ Hours to crunch Each Unit on the two 750TI SC cards, Win XP Pro x64 is taking almost twice that on a HIGHER LEVEL CARD, on the GTX-760. So, something is definitely OFF somewhere.
Here is an after the holidays update on the Credit, DCF and Memory issues:
Credit: The current policy is to bring the FGRPB1G search to the same level as the BRP4G search before. This is under the assumption that both searches behave roughly the same. As it turns out this is not the case. The BRP4G app was barely using the CPU at all while the FGRPB1G app is using a full core the whole time. As I wrote earlier I did measure runtimes for BRP4G and FGRPB1G on a GTX750 Ti (under Linux) and granted credit based on that benchmark (1000 Credits per 1h actual runtime). The effect TimeLord04 is seeing stems from the fact that the two searches perform very differently on his systems. On his Windows host the BRP4G search (CUDA implementation!) took 20 minutes per task and now FGRPB1G tasks (OpenCL implementation!) take just under 4 hours. A very naive solution would be to increase the amount of Credit per task which would increase TL's RAC to where it was before but it would still be the same amount of science done. As others pointed out RAC is a good way to measure performance if the apps and tasks are stable. Right now they are not. Everyones RAC is adjusting to the new app and will stabilize again. A better solution would be if the Windows CUDA app has the same runtime as the Linux OpenCL app.
I'll start discussing with the other Admins how to go forward here. Maybe we need to calculate Credit per Task in some other way that can be applied across searches and applications.
DCF (duration correction factor): I have been running FGRPB1 (CPU and GPU) on my own Desktop computer for some time now and monitored the estimated times per task more closely over the holidays. The estimation for CPU tasks is very accurate from the beginning. GPU tasks are generally overestimated by a factor of 2. This is consistent with my own calculation regarding the speedup factor and postings of other volunteers. So I just increased the speedup factor of the FGRPB1G GPU by a factor of 2. This should stabilize the DCF on all systems running only FGRPB1. If you also run O1MD1 your DCF may still fluctuate a bit as the performance of the O1MD1 app is not as stable as the FGRPB1 CPU app.
Memory: I'm going to deploy the beta versions with the lower memory requirement later today. But as I've already seen messages in the forum that show memory allocation problems on some GPUs I'm not too optimistic about this.
Here is an after the holidays update on the Credit, DCF and Memory issues:
Credit: The current policy is to bring the FGRPB1G search to the same level as the BRP4G search before. This is under the assumption that both searches behave roughly the same. As it turns out this is not the case. The BRP4G app was barely using the CPU at all while the FGRPB1G app is using a full core the whole time.
Thanks for the update Christian the new baby app is bringing new challenges!
Running this app on RX-480 / Ubuntu 16.04 certainly does not use a full core - ever, typically it would run say less than 20% of a core - i would say it is extremely well behaved here. edit: this host
The 1 core per GPU works fine and best throughput works at x3 or x4.
At x1 the run time is ~1200s (CPU time ~300s)
At x3 the elapsed time is ~2000s (CPU time 350s)
Quote:
Memory: I'm going to deploy the beta versions with the lower memory requirement later today. But as I've already seen messages in the forum that show memory allocation problems on some GPUs I'm not too optimistic about this.
I deployed Beta applications with lower memory restrictions for all platforms. Those are copies of the normal app and use the same amount of memory when running.
The normal apps are restricted to GPUs with at least 1022 MB of Video RAM.
The beta apps are restricted to GPUs with at least 766 MB of Video RAM.
As always: Use at your own risk. Tasks will fail if there is not enough video memory available during execution.
Edit: I edited the above wording a bit to make it clear that both app types use the same amount of memory when running.
Here is an after the holidays update on the Credit, DCF and Memory issues:
... The BRP4G app was barely using the CPU at all while the FGRPB1G app is using a full core the whole time. As I wrote earlier I did measure runtimes for BRP4G and FGRPB1G on a GTX750 Ti (under Linux) and granted credit based on that benchmark (1000 Credits per 1h actual runtime). The effect TimeLord04 is seeing stems from the fact that the two searches perform very differently on his systems. On his Windows host the BRP4G search (CUDA implementation!) took 20 minutes per task and now FGRPB1G tasks (OpenCL implementation!) take just under 4 hours.
Another "trick" is, that it was nice and easy to run 2 or 3 BRP4G tasks at once. It of course resulted in a quite nice RAC, because GPUs could run quite more effectively. With the current OpenCL app, it is almost impossible due to huge GPU memory load. Moreover, even when GPUs now show they are 99% loaded, their temperatures are significantly lower when compared with BRP4G. It is obvious that chips are not now as loaded as before with BRP4G.
Edit: experienced on linux 64bit machines with GT240, GT320, GTX460, GTX570, GTX580, GTX590, GTX760 and Quadro M1000M.
TimeLord04 wrote: At this
)
Don't get your Dad hung up on the RAC. Point out the scientific importance, which is much greater than SETI in my opinion. This is the Golden Age of gravity waves. They have just been discovered (or proven to exist might be a better term), and whatever we find now is going into the science history books. It is all unexplored territory. Ask you Dad if he wants to help fund the Magellan expedition, or help Christopher Columbus discover the New World.
And put the PC where you need the heat anyway. Dads are very practical.
Christian will probably
)
Christian will probably revisit and re-adjust credit when he's back at his desk tomorrow.
NVidia issues should get better with a CUDA version of the app, which we will work on.
This application (version) is as new to us as it is to you. Thank you very much for your contribution so far, and for your patience.
Happy new year to all!
BM
Bernd, that is the best news
)
Bernd, that is the best news I've heard this year.
Bernd Machenschalk
)
Thank you Bernd for listening. Believe me, I'm NOT complaining about RAC for RAC's sake... I'm NOT, (and NEVER have been), a credit monger. I just need to justify my activities to my dad; who pays the bills.
Something that people here may not know about me and my living situation. I'm on SSDI for Bipolar Disorder. I CANNOT hold a regular full time job. The stresses associated with a full time job where working 40+, (typically 50+ Hours, now), in the IT Field where my talents lie, exacerbates the Bipolar Disorder and causes Manic episodes. (Usually fits of anger and anxiety.) Prior to diagnosis, I went from job to job, being "dismissed" without explanation. This, too, aggravated the Bipolar Disorder; because I didn't know WHY I was being dismissed and unable to hold a job for more than a few months.
I was diagnosed in late 2007, went immediately on medication. The medication seems to keep moods fairly well in check; BUT, I still CANNOT hold a regular job. SSDI ONLY yields $1059 per month, (effective Jan 1, 2017), and almost half of that income goes to medical insurances and dental and vision insurances. The rest goes to bills. This means I CANNOT afford rent outside of my parents' household. I DEFINITELY cannot contribute to the electric and other utility bills of the house. This makes crunching a "LUXURY" in my dad's eyes.
Crunching started as a hobby for me in 2006. I immediately joined the ranks of crunchers working on SETI and here at Einstein, and a few smaller backup projects. As time went on, I dropped the backup projects. SETI became my primary project, (until recently), and Einstein was secondary, (until recently), and I thankfully have been able to crunch 15 Hours a Day until seeing these HIGH and EXORBITANT electric bills from PG&E. (Pacific Gas and Electric.)
Dad refuses to go solar; so, there's no offset to these bills - EXCEPT the inevitability that I'm going to have to reduce crunching times due to the drop in RAC. So, I'm happy to hold on and will continue crunching at the level I'm contributing to here; now that Bernd states that credit levels will be re-examined.
Again, I LOVE this project. I want to contribute the MAXIMUM that I can. Right now, two EVGA GTX-750TI SC cards, and one EVGA GTX-760 card, are all that I can afford to utilize at this time. The MAC/Hackintosh system hardware is circa 2008/2009; EXCEPT for the 750TI SC cards which one is almost three years old, and the other is six months old. The Win XP Pro x64 system with GTX-760 is three and a half years old.
In close, I don't expect the galaxy when it comes to granted credit for work done. I do expect fairer credit for work done. If I can hover in the 90K to 103K range as I was doing a month ago, this is the minimum level I can pass off to my dad as "acceptable" for 15 Hours a day of crunching. With BRP6 dead and gone, I know I'll never get back to the 131K RAC that I had at that time... I don't expect that to happen again. If RAC doesn't get back to 90-103K RAC range, I, (unfortunately), will have to cut back on crunching times. I hope you all understand where I'm coming from.
Thanks for reading this, and for understanding.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
@TIMELORD04 Sorry to read
)
@TIMELORD04
Sorry to read your difficulties and I wish you a better health for this new year.
@BERND,
Credits are a recurrent discussion on all forums. I didn't read all of them, but I simply don't understand why a working unit is not paid a fixed amount of credit depending on the estimated gflops. This should avoid frauds and reflect hardware improvements.
_AF_EDLS_GuL
)
Thanks for the kind words. btw: The Drupal system code here "CAP LOCKS" my Username... Actually, as originally entered by me on the old site, my Username is "TimeLord04"; but, can be abbreviated to "TL".
Also, (for my posts), I'm having to Double Space my paragraphs... Every time I try to Single Space my paragraphs, they end up running altogether WITHOUT any spaces between paragraphs... (Frustrating!!! DAMNED DRUPAL SYSTEM!!!)
[Update:] Day isn't even finished; and I'm down ANOTHER 3K RAC!!! Total combined RAC from both systems now at 70K...
I hope Christian can find a happier medium for everyone.
On the MAC; "Invalid" Reports are one 1.14 Unit. "Pending" Reports are one 1.14, three 1.17 Beta, and nine 1.17 standard Units. All the rest of my work is Valid and getting the 1365 Credits per Unit... So, I just can't understand this MASSIVE drop in RAC since 1.15 Units came out... ESPECIALLY when crunching LONGER on the 1.17 Units than was EVER done on BRP6 and BRP4G.
Still crunching two Units at a time per card. Six Total Units across three GPUs at a time. MAC is taking 2+ Hours to crunch Each Unit on the two 750TI SC cards, Win XP Pro x64 is taking almost twice that on a HIGHER LEVEL CARD, on the GTX-760. So, something is definitely OFF somewhere.
I will keep monitoring and reporting.
TL
TimeLord04
Have TARDIS, will travel...
Come along K-9!
Join SETI Refugees
Here is an after the holidays
)
Here is an after the holidays update on the Credit, DCF and Memory issues:
Credit: The current policy is to bring the FGRPB1G search to the same level as the BRP4G search before. This is under the assumption that both searches behave roughly the same. As it turns out this is not the case. The BRP4G app was barely using the CPU at all while the FGRPB1G app is using a full core the whole time. As I wrote earlier I did measure runtimes for BRP4G and FGRPB1G on a GTX750 Ti (under Linux) and granted credit based on that benchmark (1000 Credits per 1h actual runtime). The effect TimeLord04 is seeing stems from the fact that the two searches perform very differently on his systems. On his Windows host the BRP4G search (CUDA implementation!) took 20 minutes per task and now FGRPB1G tasks (OpenCL implementation!) take just under 4 hours. A very naive solution would be to increase the amount of Credit per task which would increase TL's RAC to where it was before but it would still be the same amount of science done. As others pointed out RAC is a good way to measure performance if the apps and tasks are stable. Right now they are not. Everyones RAC is adjusting to the new app and will stabilize again. A better solution would be if the Windows CUDA app has the same runtime as the Linux OpenCL app. I'll start discussing with the other Admins how to go forward here. Maybe we need to calculate Credit per Task in some other way that can be applied across searches and applications.
DCF (duration correction factor): I have been running FGRPB1 (CPU and GPU) on my own Desktop computer for some time now and monitored the estimated times per task more closely over the holidays. The estimation for CPU tasks is very accurate from the beginning. GPU tasks are generally overestimated by a factor of 2. This is consistent with my own calculation regarding the speedup factor and postings of other volunteers. So I just increased the speedup factor of the FGRPB1G GPU by a factor of 2. This should stabilize the DCF on all systems running only FGRPB1. If you also run O1MD1 your DCF may still fluctuate a bit as the performance of the O1MD1 app is not as stable as the FGRPB1 CPU app.
Memory: I'm going to deploy the beta versions with the lower memory requirement later today. But as I've already seen messages in the forum that show memory allocation problems on some GPUs I'm not too optimistic about this.
Christian Beer wrote:Here is
)
Thanks for the update Christian the new baby app is bringing new challenges!
Running this app on RX-480 / Ubuntu 16.04 certainly does not use a full core - ever, typically it would run say less than 20% of a core - i would say it is extremely well behaved here. edit: this host
The 1 core per GPU works fine and best throughput works at x3 or x4.
At x1 the run time is ~1200s (CPU time ~300s)
At x3 the elapsed time is ~2000s (CPU time 350s)
Will keep fingers crossed on that.
I deployed Beta applications
)
I deployed Beta applications with lower memory restrictions for all platforms. Those are copies of the normal app and use the same amount of memory when running.
The normal apps are restricted to GPUs with at least 1022 MB of Video RAM.
The beta apps are restricted to GPUs with at least 766 MB of Video RAM.
As always: Use at your own risk. Tasks will fail if there is not enough video memory available during execution.
Edit: I edited the above wording a bit to make it clear that both app types use the same amount of memory when running.
Christian Beer wrote:Here is
)
Another "trick" is, that it was nice and easy to run 2 or 3 BRP4G tasks at once. It of course resulted in a quite nice RAC, because GPUs could run quite more effectively. With the current OpenCL app, it is almost impossible due to huge GPU memory load. Moreover, even when GPUs now show they are 99% loaded, their temperatures are significantly lower when compared with BRP4G. It is obvious that chips are not now as loaded as before with BRP4G.
Edit: experienced on linux 64bit machines with GT240, GT320, GTX460, GTX570, GTX580, GTX590, GTX760 and Quadro M1000M.