http://files.mail.ru/AZCHOK

Please, update your hosts with this builds. ATi and NV builds got speed increase, HD5-version added, NV version added.

CPU builds got some updates too.

Known issues:
1) As with OpenCL AstroPulse, last driver version from both vendors show increased CPU usage. AMD already aknowledged this issue and promised to fix in new Catalyst releases, NV still keeping silence about this issue.

2) OpenCL NV app can silently (i.e., w/o errors in stderr) produce incorrect results (overflows). Again, situation resembles NV AstroPulse rev521 case and usually means too long kernel call. Why NV OpenCL runtime doesn't report error code for kernel enqueue runtime call - no idea. But low-end NV GPUs could be not capable to use this app. This testing should determine GPU requirements for NV app too.

---

News about SETI opt app releases: https://twitter.com/Raistmer

Thank you
I installed the first machines allready and seems it runs ok
Now to fine tweak the different programs :D

Updated with the new build and now get an error saying..

"Your app_info.xml file doesn't have a usable version of SETI@home v7."


my app_info has this...

<app_info>
<app>
<name>setiathome_v7</name>
<user_friendly_name>SETI@home v7</user_friendly_name>
</app>
<file_info>
<name>MB7_win_x86_SSE3_OpenCL_ATi_r365.exe</name>
<executable/>
</file_info>
<file_info>
<name>MultiBeam_Kernels_r365.cl</name>
<executable/>
</file_info>
<app_version>
<app_name>setiathome_v7</app_name>
<version_num>691</version_num>
.....

just dropped in the new file names to my existing...

running with a 1090T cpu and an HD6970 gpu

@ElGatoDelNord

Most of your GPU tasks end in Maximum time exceeded errors (-177 errors)

The only way for you to fix this is to edit your client_state.xml file (using Notepad only),
first set your GPU activity menu to Never, and make sure you get some Seti Beta tasks to edit,
Now Shut down Boinc, (you can now make a Backup of the whole Boinc Data directory, or just the client_state.xml file)
find your client_state.xml file, and right click on it and choose edit,
you'll have to find the Seti Beta section, it'll start like this:

<project>
<master_url>http://setiweb.ssl.berkeley.edu/beta/</master_url>
<project_name>SETI@home Beta Test</project_name>
<symstore></symstore>
<user_name>Claggy</user_name>
<team_name>BOINC Synergy</team_name>
<host_venue>work</host_venue>

Now you need to find the <workunit> area, just scroll down until you find an area that starts like this:

<workunit>
<name>21dc10aa.5028.7192.13.14.75.vlar</name>
<app_name>setiathome_v7</app_name>
<version_num>691</version_num>
<rsc_fpops_est>14417324582489.793000</rsc_fpops_est>
<rsc_fpops_bound>144173245824897.940000</rsc_fpops_bound>
<rsc_memory_bound>33554432.000000</rsc_memory_bound>
<rsc_disk_bound>33554432.000000</rsc_disk_bound>
<file_ref>
<file_name>21dc10aa.5028.7192.13.14.75.vlar</file_name>
<open_name>work_unit.sah</open_name>
</file_ref>
</workunit>

your will need to add a Zero onto each <rsc_fpops_bound> line of each Workunit, so

<rsc_fpops_bound>144173245824897.940000</rsc_fpops_bound>

becomes

<rsc_fpops_bound>1441732458248970.940000</rsc_fpops_bound>

(Hopefully that will be enough), once your have done all your Beta Wu's close Notepad and over-write the original client.state.xml (make sure you use ANSI encoding)
(If you make a mistake at any time, just exit out of the client_state.xml without saving, then edit it again)

Now restart Boinc, and re-enable GPU activity, if you have done it correctly the Wu's should run to completion, if they don't, stop Boinc and add another Zero to all the workunit sections,

eventually you'll see tha APR of the SETI@home v7 (anonymous platform, ATI GPU) app go down from it's present 5598 value to a more reasonable value (probably under 500),
you'll have to do this quite a few times until your APR value stablises, Good Luck.

Claggy

Looks like 2 zeros is the magic number to get the errors to stop. Any idea what bonehead thing I did to get it to this point in the first place? So I can perhaps avoid it in the future?

Thanks a bunch!

Looks like 2 zeros is the magic number to get the errors to stop. Any idea what bonehead thing I did to get it to this point in the first place? So I can perhaps avoid it in the future?

Thanks a bunch!

You completed a number of -9 Wu's that took ~25 seconds, so the the Server scaled the <rsc_fpops_est> and completion time to that, nothing you could do about that apart from not doing those task, lol
Anyway, last week i sent an email asking for MB -9's, and the two Astropulse early exit conditions to be excluded from APR calculations,
I've had a reply from DA:

I'll do 2 things:

- short term: put a limit on the impact of ET estimates
in runtime estimation, to avoid -177 errors.

- long term: as you suggest, allow the project's validation function
to say "Ignore this job in timing statistics"
(i.e., because it exited early, like SETI at home's overflow jobs).

It may take a few days to deploy these on S at h.

Claggy

Sounds good. Can I assume the fixes will only avoid future occurences? Looks like it will be taking a long while to get my APR down to normal.... at least several hundred results... guess I'll be writing a script to add zeros, lol.

One other quick question... any advantage to add more CPU to the processing, say .5 instead of .05 ? Or will that mess something else up?

Thanks again...

Sounds good. Can I assume the fixes will only avoid future occurences? Looks like it will be taking a long while to get my APR down to normal.... at least several hundred results... guess I'll be writing a script to add zeros, lol.

One other quick question... any advantage to add more CPU to the processing, say .5 instead of .05 ? Or will that mess something else up?

Thanks again...

It could mess up your CPU tasks, atleast slow them down. Cause you reserve 50%
of your CPU-time, loading/unloading your GPU(s)! Maybe try 0.1 (10%)

And I just installed rev365, on my I7-2600, box with some help with the app_info.xml file. Thanx!
And turned HT ON, everything looking OK.
Apologies to my wingmen/women, if WUs were timed-out.

...
One other quick question... any advantage to add more CPU to the processing, say .5 instead of .05 ? Or will that mess something else up?

Thanks again...

No effect at all, since that parameter cannot control fractional CPU usage. It's merely an estimate.

The BOINC core client uses that estimate to decide whether it should reserve a CPU for feeding GPUs. That is, if there's more than 1 GPU or more than 1 task being run per GPU the estimates are added and if the sum shows the GPU processing needs 1.x CPUs then BOINC will run one less pure CPU task.

                                                                  Joe

Sounds good. Can I assume the fixes will only avoid future occurences? Looks like it will be taking a long while to get my APR down to normal.... at least several hundred results... guess I'll be writing a script to add zeros, lol.

One other quick question... any advantage to add more CPU to the processing, say .5 instead of .05 ? Or will that mess something else up?

Thanks again...

Changing the <max_ncpus> and <avg_ncpus> values to 0.5 won't have any direct affect on the amount of CPU used by the OpenCL app, these values are for Boinc's internal use, the app will use what CPU it needs,
There may be an indirect affect, on your AMD Phenom II X6 1090T processor, three cores (50% of 6) will get reserved for feeding the GPU, there should be a small speedup, (setting it to 0.166 will reserve a core)
using the -hp (high priority) switch will do the same thing (at the expense of other running apps)

Claggy

And I just installed rev365, on my I7-2600, box with some help with the app_info.xml file. Thanx!
And turned HT ON, everything looking OK.
Apologies to my wingmen/women, if WUs were timed-out.

That's very good Fred, But what about the Q6600 host that is still running r229 and is still throwing excessive inconclusives?

Pending tasks for computer 45894

Claggy

Hi Claggy, you're right, will check that host now, also have a weird problem
with rev.365.
After switching to Collatz C. it stopped doing SETI Bêta MB, with a message:

5-9-2011 17:10:08 [error] No URL for file transfer of
MB7_win_x86_SSE3_OpenCL_ATi_r365.exe!

But the executable is in place and it's name is in the app_info.xml
Any thoughts? For a start, I can disable Collatz C., but it is strange.

Hi Claggy, you're right, will check that host now, also have a weird problem
with rev.365.
After switching to Collatz C. it stopped doing SETI Bêta MB, with a message:

5-9-2011 17:10:08 [error] No URL for file transfer of
MB7_win_x86_SSE3_OpenCL_ATi_r365.exe!

But the executable is in place and it's name is in the app_info.xml
Any thoughts? For a start, I can disable Collatz C., but it is strange.

Make sure you change all the locations in the app_info to MB7_win_x86_SSE3_OpenCL_ATi_r365.exe Or MB7_win_x86_SSE3_OpenCL_ATi_HD5_r365.exe (the HD5 version is faster)
and make sure you haven't over-written any > or < entries.

Claggy

Make sure you change all the locations in the app_info to MB7_win_x86_SSE3_OpenCL_ATi_r365.exe Or MB7_win_x86_SSE3_OpenCL_ATi_HD5_r365.exe (the HD5 version is faster)
and make sure you haven't over-written any > or < entries.

Claggy

Found, (the stupid) mistake...............forgot, didn't notice, the HD5 :( , entry in my app_info.xml file!

I'll increase DPI to >150%, maybe then I'll notice it, sigh......

Anyway thanks for your swift and accurate advice. (As usual ;-) )

MB7_win_x86_SSE3_OpenCL_ATi_HD5_r365 seems to work well on my ATI HD 6970.

I'm running 4 instances per GPU, and so far they validate.

---

Morten Ross

I'm now running 2 MB WUs per 5870 and 8 on the I7-2600, with rev.365 for ATI HD5.
Even with HT turned on, using BOINC 6.10.60, now displays 8 'CPUs' or 'cores
and the "benchmark" shows 2300MIPS (Whetstone) and 9600MIPS (Drystone)(ops/sec).
Thats 12 concurrent WUs, GPU is better loaded, compaired to previous versions
of OpenCL. (Power-draw is a bit less, compaired with Milkyway or Collatz C.,
although you can't really compaire these different (Collatz C.) projects.
I think you did a good job, Raistmer :), and all who are/were involved in design, upgrade librairies, etc. and the compilation of this build!

Well, I could try 4, with 1024MByte GDDR5, this should fit?
But I'll have to install
OpenCL app on my Q6600+GTX470, now running a NVIDIA CUDA app. But also
is OpenCL capable. And has (just) enough memory (1280MByte) to run 4 per GPU.

Thats weird my card does not want to run any units with the hd5 version.
With the normal version i am getting constant lagg even with iterations set to 4
I wonder if some of these problems are related to the driver version i am running
Somehow it also does not run smooth with 2 units anymore, while with the 331 version 2 units only got it loaded about 84%, with the 365 it is with 1 unit 94% loaded
Also am wondering if the problem is a switch issue i had before with collatz on the 4770 which totally hung a couple of times and not even started the unit proper.

@ Fred J. Verster

And I just installed rev365, on my I7-2600, box with some help with the app_info.xml file. Thanx!
And turned HT ON, everything looking OK.
Apologies to my wingmen/women, if WUs were timed-out.

That's very good Fred, But what about the Q6600 host that is still running r229 and is still throwing excessive inconclusives?

Pending tasks for computer 45894

Claggy

Fred, we really need no more proof that r229 produces too many inconclusives. Please either upgrade that host to r365 or do some testing of the stock 6.96 build.

                                                                    Joe

Seems after server got up again the error disappeared the units got downloaded for the hd5 version for the ati
Also been preparing to run the 4770 to run with V7 seti lets see how well it does
I try with basic settings first

I'll change it a.s.a.p. , sorry if I caused (again) any inconvieneance.
I've set it to N.N.T., also stopped SETI Bêta processing, so I can change this first thing tomorrow, gettin a bit late and don't want to make another (typo) error.
I should take more time to check on my rigs, no excuse for this.

[ADDED]
Since the host with a Q6600+gtx470, should I try the non HD5 version?
Or just try, what works best?
It is OpenCL ready, according to GPUz (0.54)
Fred J. Verster.