New build has improved performance and supports new command line switches:
-hp - sets high priority class
-no_cpu_lock - disables affinity setting
-instances_per_device N - will allow running N copies per each supported GPU device (don't forget to set <count> field in app_info to 1/N fraction to instruct BOINC to launch N tasks per GPU).

App processing correctness and work of these new switches should be tested during this beta run.

To participate in this beta run PM me link to host with ATI GPU where you plan to test this build.

@All who will participate:
Don't forget to finish current AP task before upgrade. Or you will need to manually update CL file not only in SETI project directory but in corresponding slot directory too. BOINC doesn't do this, design flaw IMHO.

To participate in this beta run PM me link to host with ATI GPU where you plan to test this build.

Для того что бы участвовать в тестировании тебе в личку линк кидать или что бы, те кто тестируют, кинули тебе в личку линк на хост?
ЗЫ: об машинный перевод себе можно мозг сломать.

To participate in this beta run PM me link to host with ATI GPU where you plan to test this build.

Для того что бы участвовать в тестировании тебе в личку линк кидать или что бы, те кто тестируют, кинули тебе в личку линк на хост?
ЗЫ: об машинный перевод себе можно мозг сломать.

"Утром деньги - вечером стулья" ;) Наученный горьким опытом, когда число скачавших несколько десятков, а число хоть что-то потом написавших - штуки, я бинарники выдаю только когда линк пришлют. Чтобы хотя бы самому потом видеть как дело идет.

Looks like r503 contains not disabled debug dumps that will lead to disk space shortage and -177 error from BOINC (max disk space exceeded).

Please, suspend current AP task in progress and await updated build.

That was close last night.
I just started the new app before i readed your coment.
Switched back quickly.

---

With each crime and every kindness we birth our future.

Actually testing can be continued but carefully. One need time to time stop processing and delete stderr.txt from corresponding slot directory (it grows too big otherwise).

Actually testing can be continued but carefully. One need time to time stop processing and delete stderr.txt from corresponding slot directory (it grows too big otherwise).

пока такого не наблюдаю.
не понял что означает -no_cpu_lock
при <cmdline>-hp -no_cpu_lock instances_per_device 1</cmdline> ВУ считаются процентов на 20 быстрее (при сопоставимом percent blanked)
какие по умолчанию ffa_block и ffa_block_fetch?

насколько помню 8192 и 4096 (лучше посмотреть релиз предыдущей версии, много воды с тех пор утекло).
надо бы попробовать только -hp - будет тот же прирост производительности или нет.
вторая опция выключает привязку к конкретному процессору, аффинити не будет задаваться.
Опция спорная. Есть случаи когда аффинити сильно помогает, а надо ли при обычном использовании - надо проверять.
Поэтому добавил возможность ее отключать если надобности нет.

One need time to time stop processing and delete stderr.txt from corresponding slot directory (it grows too big otherwise).

Is there an other limit than the limit set under user settings? I have set this to 100 GB maximum (or 90% of the total disk space).
The first task using the r503 build is now at 91% and the stderr.txt has grown to 616 kB.

---

Seti@Netherlands website

One need time to time stop processing and delete stderr.txt from corresponding slot directory (it grows too big otherwise).

Is there an other limit than the limit set under user settings? I have set this to 100 GB maximum (or 90% of the total disk space).
The first task using the r503 build is now at 91% and the stderr.txt has grown to 616 kB.

There should be no problems on silent tasks. But if task contains many reportable signals stderr will be overflowed with whole arrays dump. I'll disable verbose debug code then give links to new version.

The first task ended just fine, and it seems a lot faster than the r456 release. The average of 0% blanking tasks was 5,150 seconds GPU and 1,700 seconds CPU. This first task was 3,719 seconds GPU and 1,333 seconds CPU.

I sent PMs with link to r504. If I missed someone, please, let me know.

A few observations:

1. The load on the GPU increased from 40% (r456) to 60-70% (r503)
2. I tried running 2 tasks at once by setting "-instances_per_device 2" and <count> to 0.5. Although it worked, I estimated the eventual runtime at 20 hrs for both tasks (= 10 hrs per task). Both tasks had 0% blanking. The GPU load was 95% when running 2 tasks simultanously. After 30 minutes I switched back to 1 task at a time.

This was tested with an HD5870.

Hello Raistmer,

First message sent in forums (beleive it or not)
Habster here

has participated in the SETI@home project since 26 September 1999, and has contributed 3,566,059 Cobblestones of computation (3.08 quintillion floating-point operations) to SETI@home's search for extraterrestrial life.


I bought a new rig recently with 1090T , 5850, asus m4a89gtpro 890gx chipset.
Experimenting with r503..Would appreciate link to r504 and reporting my findings to the group..
Noticing you are doing great work with Ati GPU.

---

Habster 1998

Application and app_info updated ... but I'm waiting for AP WUs to start testing :(

Very preliminary observations on my low-end (80 stream processor, no fan) ATI card (GIGABYTE GV-R455D3-512I Radeon HD 4550 512MB) on a fairly high-end host E5620 Westmere overclocked at 3.42 GHz.

1. hot changeover to r504 from r456 seemed to work, all I did was to copy the two new files from the distribution rar to the SETI project directory and update the two executable references in app_info. I did not copy to any slot (I am set to suspend GPU work while computer in use, which seems to have the side effect not leaving the executable or .cl file behind in any slot when I am interacting with the host)
2. so far it appears that for my host with its settings there is not an obvious saving in GPU execution time. It seems more likely that an increase in GPU time will be needed, at least on the first WU.
3. so far it appears that there is a very big savings in CPU. With almost two hours gone, the CPU% displayed by BOINCTasks is 1.59%, I think in this condition with r456 this number would have been something on the order of 7%.

I emphasize that these are very preliminary observations--partly meant to suggest that others check for both GPU and CPU changes, and also that we consider sharing other configuration parameters (#stream processors available is obvious, others may be important) which might drive differences in result.

I've not yet tweaked any of the old configuration parameters, let alone any new options, seems best to get a little same-to-same comparison first.

---

Hello Raistmer,

First message sent in forums (beleive it or not)
Habster here

has participated in the SETI@home project since 26 September 1999, and has contributed 3,566,059 Cobblestones of computation (3.08 quintillion floating-point operations) to SETI@home's search for extraterrestrial life.


I bought a new rig recently with 1090T , 5850, asus m4a89gtpro 890gx chipset.
Experimenting with r503..Would appreciate link to r504 and reporting my findings to the group..
Noticing you are doing great work with Ati GPU.

Send me lonk to your host with ATI GPU via PM.

1. hot changeover to r504 from r456 seemed to work, all I did was to copy the two new files from the distribution rar to the SETI project directory and update the two executable references in app_info. I did not copy to any slot (I am set to suspend GPU work while computer in use, which seems to have the side effect not leaving the executable or .cl file behind in any slot when I am interacting with the host)

This may result in task goint to trash. CL file definitely changed and its update is required.

ONe thing I've noticed with all the AP ATI apps is the credits seem to shrink. my latest WU only granted 589 credit which is a far cry from the 1200 it's supposed to be granted

---

In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope