Message boards :
Number crunching :
Gaia 5
Message board moderation
Author | Message |
---|---|
Send message Joined: 21 Aug 19 Posts: 115 Credit: 888,695 RAC: 2,016 |
Hi, I create Gaia 5, it is the same as Gaia 4 but with small correction of code. I hope fix some problem with terminate error. |
Send message Joined: 21 Aug 19 Posts: 115 Credit: 888,695 RAC: 2,016 |
Gaia 4 and Gaia 5 need libquadmath .... |
Send message Joined: 11 Jul 22 Posts: 3 Credit: 960,519 RAC: 22,086 |
Will they be able to checkpoint in the future? Since they are running for many hours now that would be nice... |
Send message Joined: 19 Oct 24 Posts: 8 Credit: 474,600 RAC: 2,267 |
I do want checkpointing as well as I am running stuff on a laptop, also is there any news regarding the research behind all of this? |
Send message Joined: 21 Aug 19 Posts: 5 Credit: 6,523,508 RAC: 102,541 |
Gaia 4 and Gaia 5 need libquadmath .... already present in ubuntu repositories... |
Send message Joined: 3 Sep 19 Posts: 21 Credit: 28,509,619 RAC: 103,896 |
I received one. It progressed rather quickly compared to Gaia 4 tasks of today but then the progress bar slowed down the closer it got to 100% and but its been at 100% for about 1 hr of its 2 hr run time. Can we get app selection in user preferences? I'd like to have more Gaia 5 tasks for WUProp hours. |
Send message Joined: 3 Sep 19 Posts: 21 Credit: 28,509,619 RAC: 103,896 |
|
Send message Joined: 21 Aug 19 Posts: 115 Credit: 888,695 RAC: 2,016 |
I can't find where these restrictions are :( There are tasks where they count for 4-5 hours without any problems ... |
Send message Joined: 21 Aug 19 Posts: 115 Credit: 888,695 RAC: 2,016 |
I changed one parameter ... Maybe will be better |
Send message Joined: 22 Oct 24 Posts: 27 Credit: 494,000 RAC: 15,469 |
Failed due to unnecessary time limits Your host details show an abnormally large floating point and integer benchmark rating. That skews your estimated times to completion way short of what the task rsc_fpops_estimate value for the task think the host should complete task in. Either you need to hamstring your benchmark values by running benchmarks again without SMT on or maybe having virtualization enabled is what is skewing the benchmarks so high. I don't know why the Boinc client can have such highly skewed benchmark results for certain hardware and OS. You can have hugely absurd values on Raspberry Pi's compared to the latest AMD and Intel cpu benchmarks for example. My benchmarks for my 9950X are a fraction of your 5950X and theoretically, the Zen 5 architecture should have much larger benchmarks over your Zen 3 cpu. The other solution is for @zupa to adjust the rsc_fpops_estimate values in the task generator templates higher. |
Send message Joined: 21 Aug 19 Posts: 115 Credit: 888,695 RAC: 2,016 |
Hi, I increased this number ... I hope this solves the problem |
Send message Joined: 22 Oct 24 Posts: 27 Credit: 494,000 RAC: 15,469 |
Hi, Yikes, you sure did. Estimated computation size But that causes more issues. Now every Gaia@home task will run in EDF or high-priority mode and shove aside your other tasks from other projects especially with such a short 2 day deadline. The jump from 3,600 GFLOPs to 1,200,000 GFLOPs is TOO MUCH |
Send message Joined: 3 Sep 19 Posts: 21 Credit: 28,509,619 RAC: 103,896 |
Yeah my ETA is now 46 days. |
Send message Joined: 22 Oct 24 Posts: 27 Credit: 494,000 RAC: 15,469 |
Why did you hide your computers now? There is nothing harmful exposed by leaving them visible? |
Send message Joined: 21 Aug 19 Posts: 115 Credit: 888,695 RAC: 2,016 |
I changed to 1 day |
Send message Joined: 22 Oct 24 Posts: 27 Credit: 494,000 RAC: 15,469 |
I changed to 1 day Not sure I understand what you changed. The deadline to only a single day? Or this from the task properties. Task size: 300000.0000 GFLOPs (rsc_fpops_est) App speed: 3000.0000 GFLOPs/sec (rsc_fpops_bound) That is much better. Only 100X the original 3600 GFLOPS for the tasks we were originally sent before you made the big jump to 1.2Million GFLOPS That should produce a much more accurate estimation of task completion times. And still prevent the exceeded time limit errors for those hosts with too large and inaccurate cpu benchmark values. Still has the task referenced above in EDF mode though. Should eventually settle out once enough get completed with the new value and not get forced into EDF. |
Send message Joined: 22 Oct 24 Posts: 27 Credit: 494,000 RAC: 15,469 |
On a different subject . . . do you know the cause for the errors that occur early in the task computation for 'output file missing? Maybe you can explain what each of the two applications that run for each task do. Same app_names EXCEPT for either a little 'g' or capitalized 'G' in the Gaia name part of the application. I see the capital G app mostly sits in sleeping mode before waking up for bit before going back to sleep. And you upload 4 output files for each task. Are two output files for each of the application name types. Is the semi-sleeping app sort of an accumulator or feeder for the main app that runs all the time? |
Send message Joined: 21 Aug 19 Posts: 115 Credit: 888,695 RAC: 2,016 |
Hi , 'g' is wrapper for 'G' app 'G' have 3 files for input and create 3 files with results. (one output file is the internal system communication) "errors that occur early in the task computation for 'output file missing" I hope the Gaia 5 solve this problem ... |
Send message Joined: 19 May 20 Posts: 38 Credit: 44,565 RAC: 900 |
Will they be able to checkpoint in the future? Since they are running for many hours now that would be nice... +1 |
Send message Joined: 1 Feb 21 Posts: 1 Credit: 378,205 RAC: 12,046 |
Will they be able to checkpoint in the future? Since they are running for many hours now that would be nice... +2 Also, it would be nice if linux dependencies were listed somewhere. I had tasks failing for a while until I found a forum post with the solution. That issue, together with finding there are no checkpoints, almost made me skip this project. Got it running fine now though. |
©2024 GAVIP-GC