Thanks in advance and please any help is welcomed.

Стоит проверить, была ли в логе опция TWIN_FFA в старой версии.
Новый билд использует TWIN_FFA? поэтому оптимальный набор параметров для FFA изменился.
Это касается опций -ffa_block и -ffa_block_fetch.
Возможно, стоит подобрать новые оптимальные значения. Как первая аппроксимация подойдет уменьшение обоих параметров в 2 раза.

In My Humble Opinion:

It seems that everybody is talking about Astropulse workunits.

More credits, I understand it, everybody want's to be top on statistics.

What about Lunatics SETI@home v7 optimized applications? Any "-use_blah_blah -unroll 10 -ffa_block 666999 -ffa_brick_fetch 6144 -Itune 1 64 4 1" tricks?

Sorry, but getting irritated of this Credit Screw AP feeding frenzy....

---

"Please keep Your signature under four lines so Internet traffic doesn't go up too much"

- In 1992 when I had my first e-mail address -

Yes we talhk about AP WU, why no MB, the MB builds are the same about a year and we allready explore its configuraions, so it´s dificult something new apears on that area.

About creditscrew i totaly hate it too, but that was old too, all knows that - LOL.

---

There should be absolutely no difference in how r2399 runs, whether it's delivered by the installer or downloaded and installed manually. The only difference might be in the recommended configuration settings - @ Mike, can you remember if you changed any advice when you reviewed the NVidia ReadMe file in preparation for the installer release?

I think Mark hasn't got the parameters right yet, it seems it's the first time he's used the -use_sleep parameter, when using it you need to load up the -unroll value too, -unroll 12 might not be enough,

another thing to note about r2399 GPU apps, they have the TWIN_FFA mod, meaning the ffa parameter sweet point will have changed because of that.

Claggy

Thanks in advance and please any help is welcomed.

Стоит проверить, была ли в логе опция TWIN_FFA в старой версии.
Новый билд использует TWIN_FFA? поэтому оптимальный набор параметров для FFA изменился.
Это касается опций -ffa_block и -ffa_block_fetch.
Возможно, стоит подобрать новые оптимальные значения. Как первая аппроксимация подойдет уменьшение обоих параметров в 2 раза.

Google Translate:

It is necessary to check if there was an option in the log TWIN_FFA in the old version.
New build uses TWIN_FFA? therefore the optimal set of parameters for FFA changed.
This applies to options and -ffa_block -ffa_block_fetch.
It may be worth to pick up the new optimal values​​. As a first approximation to fit the decrease of both parameters in 2 times.

Claggy

In My Humble Opinion:

It seems that everybody is talking about Astropulse workunits.

More credits, I understand it, everybody want's to be top on statistics.

What about Lunatics SETI@home v7 optimized applications? Any "-use_blah_blah -unroll 10 -ffa_block 666999 -ffa_brick_fetch 6144 -Itune 1 64 4 1" tricks?

Sorry, threads like this it do get a little stuck in a rut once in a while.

And it depends which SETI@home v7 application you're talking about. I see you have four hosts with NVidia cards, two hosts with ATI/AMD cards, and several hosts with no GPU at all. The answers will be different.....

In reverse order,

The CPU SAH_v7 apps don't need, or have, any fine-tuning capability.

The AMD/ATI applications do have command-line options, but my understanding is that they are less important than the AP variants (I run NVidia myself, so I don't have any personal experience of the ATI/AMD setup).

The NVidia app for SAH v7 cam be tuned via a file called 'mbcuda.cfg'. That file already has brief notes about what configuration changes are possible, and sample lines which can be activated by removing the ';' comment prefix. There is fuller documentation in the file ReadMe_x41zc.txt

If you have run the Lunatics installer, the working files will have been copied to the setiathome project folder in your BOINC data directory. The setiathome folder will also contain a 'docs' subfolder, which - apart from the various legal and licencing requirements - will contain ReadMe files for each of the applications which you have chosen to install on that computer. Have a read through those, and come back to us here if there are any further questions you'd like help with.

Thanks. It does help if you bring questions and problems here to the public forums, where everyone can mull over the examples you've provided.

I know that, but i don´t like to keep posting too much on the public forums since as you kindly explain we use a highly specialyzed configuraitons who could cause a lot of confusion to the major part of the normal volunteers, and could easly leave to wrong assumptions. That why i normaly rise those question on the team forum only.

Плохой подход, если только в вашей команде нет своего собственного разработчика.
У меня нет высокопроизводительного оборудования, чтобы тестировать новые версии. Основная отладка и тьюнинг проходят на относительно старом оборудовании. Поэтому без регулярных отчетов о проблемах с высокопроизводительным оьборудованием не стоит ожидать каких-либо улучшений.

What I think is that some things perhaps were not fully tested and perhaps should not have been released into the wild.

This release, no matter what some of it's purponents may say.
Sucks big time.
You take that for the comment it is.
The app sucks.

It can be corrected for, perhaps. But apps should be self correcting, not having to be a last ditch effort to save, kids.

Out of the box, they should be almost the best that they can be.
You got it wrong, and you should admit it.

Mewoclawsout.

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

Where people have posted details of specific incompatibilities with specific hardware, I have already done my best to make alternative builds available to work round the problems identified.

But in this case, Juan has made it clear that he has been using r2399 for some time - I don't know how long exactly - and didn't install it using the installer.

It is entirely possible that some combination of specific hardware and specific data from the recently split tapes has unearthed a previously unknown problem.

But in order for anything to work here, you need both hardware and software. Neither works without the other.

It would be better if the hardware specialists and the software specialists worked together, and between them came up with solutions which had been subjected to more complete testing than the few remaining volunteers at Lunatics can supply on their own.

But in this case, Juan has made it clear that he has been using r2399 for some time - I don't know how long exactly - and didn't install it using the installer.

For a long time since Mike´s ask me to test in 12 Jun 2014, 13:44:29 UTC so i must start to use maybe few days after since was in the middle of the Fifa World Cup and was a heavy drinking season here.

So it´s about 2 months ago and i never notice a drop like this in the production. I keep an eye on my farm production at least once a day when the Free-DC daily stats are released. So a drop like that i will sure see.

I´m sure you Richard allways try to do your best to solve all the problems, but before i try to present a problem i need to understand it first and how it happening, and if it realy exists. that´s why i ask yesterday if anyone else experimenting the same problem in our team forum before i go to other place.

One thing comes to my bugged mind few moments ago, could be related to something changed in the last week windows update? Most of my host updated in the weekend. Have no ideia, just guessing, i´m trying to find an un-updated host here to test but can´t find anyone maybe in the ofice, but for that i need to wait until monday.

@Raistmer

I ask allways first in my team forum where the ambient is more "friendly" than the open forums and with a lot less "noise", and we all speak the same language and i don´t need to use the translator to try to understand the posts. A lot of meaning is loosed when you use the automatic translator like the google translator i use. Even in english sometimes is hard for me to understand, but that is my fault only.

As i do this time, the post i copy was made 23 Aug 2014, 13:54:17 UTC and wait for answers. Normaly i find all my answers there.

If i could not fix the problem i try on other forum depending on the problem. But this time since i notice the atmosphere feels to heavy today i try to put some calm on that and simply explain way.

And as i say, i´m not using the 0.42 installer so the incompatibilitie can´t come from it. So put something about that on the installer thread is certainly of topic and sure not welcomed.

I only start to be sure about the problem yesterday when the Free-DC stats confirm the drop on my daily production, before that i was not even sure a problem exists. You know the daily production oscilate a lot from one day to the other.

If i can´t find my answer in some time i sure call for help on the apropriate forum, in this case the Lunatic forum.

All my host remains at your disposal to make any test necessary, if you have time & pacience to explain to me what i need to do of course, i´m realy just a "mad" user with no experience on this specific software development area.

---

Since I ran a few thousand AP Workunits with the latest AP Apps on both NVidia and AMD cards :

I've noted there recently were some tapes producing a very high number of extremely blanked AP Workunits (basically all of them, the whole tape), namely :

ap_06dc08ac
ap_08ja09ad

They got distributed about 2 days ago.

If someone downloaded a whole bunch of those - that would downright double computing time on the GPUs for quite a while... Maybe worth checking out.
I know I took my time fighting through them, and it's not that I'm lacking hardware right now ;)

---

Did you try the v7.02 app on Beta? It is based on build r2601.

The last I try was this build (AP7_win_x86_SSE2_OpenCL_ATI_r2567_oclFFT_11.12drv_issue_fix.7z):
http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=2185&postid=51771#51771

It works in offline test
.....

It is good the app version to fix that issue works on your system.
Your message:

Add also this issue:
AP6_win_x86_SSE2_OpenCL_ATI_r2058 ... r2567
Fail on Catalyst 11.12 on Windows XP

Seemed to say that all revisions from r2058 on did not work. Including AP7 r2567.
You can use app_info.xml on beta to specify whichever plan class you like.

"Including AP7 r2567":
Yes, I included this number because if you look again on the Beta thread - there is NOT ONLY ONE debug package posted with r2567
So I prefer to include r2567 in the range as only the last debug sub-version have the fix (the exe file have the same name for 'bad' r2567 and 'good' r2567)

The version included in the Lunatics Windows Installer v0.42 have this problem (AP6_win_x86_SSE2_OpenCL_ATI_r2399.exe)
Do not run it under Catalyst 11.12 as it will Fail

Tested only on Windows XP but seems it is driver issue and will probably be the same on other Windows:
"11.12 can't accept NULL for buffer.
I would say it's driver bug, cause nothing about it in OpenCL specs as far as I can recall."
http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=2185&postid=51770#51770

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

Since I ran a few thousand AP Workunits with the latest AP Apps on both NVidia and AMD cards :

I've noted there recently were some tapes producing a very high number of extremely blanked AP Workunits (basically all of them, the whole tape), namely :

ap_06dc08ac
ap_08ja09ad

They got distributed about 2 days ago.

If someone downloaded a whole bunch of those - that would downright double computing time on the GPUs for quite a while... Maybe worth checking out.
I know I took my time fighting through them, and it's not that I'm lacking hardware right now ;)

Thats exactly the point.

On my host AP`s with blanking <60% take approx 4-5 times longer.
Normal AP`s ~2000 seconds high blanked tasks 8000 - 9000 seconds.

I havent seen so many high blanked tasks for quite a while so it might have been plain luck for some with over committed hardware.

Considering some have 4 top GPU`s running with a slow CPU and running at least 2 instances on each card would require at least a 8 core CPU to cope with high blanked AP`s.

---

With each crime and every kindness we birth our future.

Considering some have 4 top GPU`s running with a slow CPU and running at least 2 instances on each card would require at least a 8 core CPU to cope with high blanked AP`s.

And that's why we wanna AP v7 release ASAP.

Considering some have 4 top GPU`s running with a slow CPU and running at least 2 instances on each card would require at least a 8 core CPU to cope with high blanked AP`s.

And that's why we wanna AP v7 release ASAP.

~Raistmer

It works on all My Hosts, http://setiweb.ssl.berkeley.edu/beta/hosts_user.php?userid=19465
I even loaded up XP with old NV 8800GT and ATI HD4 cards, it worked with them as well.

I say release it and sort out the problem cases later...

Considering some have 4 top GPU`s running with a slow CPU and running at least 2 instances on each card would require at least a 8 core CPU to cope with high blanked AP`s.

And that's why we wanna AP v7 release ASAP.

~Raistmer

It works on all My Hosts, http://setiweb.ssl.berkeley.edu/beta/hosts_user.php?userid=19465
I even loaded up XP with old NV 8800GT and ATI HD4 cards, it worked with them as well.

I say release it and sort out the problem cases later...

You'll have to discuss that with Eric. He makes the call, and we'll follow as quickly as we can afterwards.

Thanks Richard fo the quick workaround i will wait till the current block of units is almost complete then i will try the new version Boinc and your workaround and try it with the last few units i have on the machine with the GT 220 and let you know unless the new version AP comes out and then i'll try that ,

Sorry about waiting till i have completed most of the units i have but i've all ready stuffed enough on that machine and there aren't any more so if i lose them i'll have to wait till there are more next week and i've only just got my Rac up to a normal

anyway thanks again for the quick fix

---

I've found that current highly blanked AP tasks are taking as much time to run now on a 4770K w/ 2 x GTX750Ti @ 2ea w/.5 core for each task (2~2.5 hrs.) with these options -use_sleep -unroll 10 -ffa_block 12288 -ffa_block_fetch 6144 -tune 1 64 4 1 -hp as it did running normal tasks with these options -unroll 12 -ffa_block 8192 -ffa_block_fetch 4096 -hp, with 1 core for each task and 3 tasks each. CPU & total run times dropped significantly after inserting the sleep option, and the tune option run times did lower total runtime a little, but the most effect that I've seen from that is the loading of the kernel. I'm noticing one other effect in that Precision X is showing GPU usage is all over the map varying 59-98% when it was usually a steady 99%. I don't know if this if a side effect of heavy blanking or not.

---

I don't buy computers, I build them!!

I've found that current highly blanked AP tasks are taking as much time to run now on a 4770K w/ 2 x GTX750Ti @ 2ea w/.5 core for each task (2~2.5 hrs.) with these options -use_sleep -unroll 10 -ffa_block 12288 -ffa_block_fetch 6144 -tune 1 64 4 1 -hp as it did running normal tasks with these options -unroll 12 -ffa_block 8192 -ffa_block_fetch 4096 -hp, with 1 core for each task and 3 tasks each. CPU & total run times dropped significantly after inserting the sleep option, and the tune option run times did lower total runtime a little, but the most effect that I've seen from that is the loading of the kernel. I'm noticing one other effect in that Precision X is showing GPU usage is all over the map varying 59-98% when it was usually a steady 99%. I don't know if this if a side effect of heavy blanking or not.

Thats correct Cliff.
Since blanking is processed on CPU on AP6 the GPU usage will drop by the amount of blanking.
This means a task with 86% blanking will only use 24% of the GPU.
This will change for AP7.

---

With each crime and every kindness we birth our future.

Any ideia when the AP 7 will be avaiable?

---