Advanced search

Message boards : Graphics cards (GPUs) : How to receive only one WU at a time ?

Author Message
zioriga
Send message
Joined: 30 Oct 08
Posts: 46
Credit: 297,857,425
RAC: 4,578
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13724 - Posted: 29 Nov 2009 | 22:17:52 UTC

I'm asking if it's possible to receive only one WU at a time. Now I reduced, in many steps to 0.3 days, the "Maintain enough work for an additional" parameter.
But the result was quite the same: 2 WU each time on a GTX 260.
Is there another way to obtain this goal ??

Siegfried Niklas
Avatar
Send message
Joined: 23 Feb 09
Posts: 39
Credit: 144,654,294
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 13725 - Posted: 29 Nov 2009 | 22:52:01 UTC - in response to Message 13724.

Try to set "Network activity always available" in BOINC Manager.

(Open BOINC Manager window - advanced view, open the menu "Activity")

zioriga
Send message
Joined: 30 Oct 08
Posts: 46
Credit: 297,857,425
RAC: 4,578
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13726 - Posted: 30 Nov 2009 | 5:43:29 UTC

Niklas wrote:

Try to set "Network activity always available" in BOINC Manager.

(Open BOINC Manager window - advanced view, open the menu "Activity")

????????

First of all my BM is always in "Network activity always available" state

Second I don't know why the network activity can manage a situation where in my computer there is also 1 WU and only after this WU terminates, the GPUGrid server can send me only another 1 WU (and not 2 WU at the same time)
The only way I know to reach this goal, now is :
set in NoMore WU state -> terminate all the pending WUs -> Allow More work -> Updte the project -> as soon as a new WU arrives, with a lightning speed, set the project in NoMore WU state. (but not obviously is always working correctly)

in the meantime I'll reduce agan the "Maintain enough work for an additional" parameter

MarkJ
Volunteer moderator
Volunteer tester
Send message
Joined: 24 Dec 08
Posts: 738
Credit: 200,909,904
RAC: 0
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13727 - Posted: 30 Nov 2009 | 7:38:36 UTC - in response to Message 13726.

Niklas wrote:

Try to set "Network activity always available" in BOINC Manager.

(Open BOINC Manager window - advanced view, open the menu "Activity")

????????

First of all my BM is always in "Network activity always available" state

Second I don't know why the network activity can manage a situation where in my computer there is also 1 WU and only after this WU terminates, the GPUGrid server can send me only another 1 WU (and not 2 WU at the same time)
The only way I know to reach this goal, now is :
set in NoMore WU state -> terminate all the pending WUs -> Allow More work -> Updte the project -> as soon as a new WU arrives, with a lightning speed, set the project in NoMore WU state. (but not obviously is always working correctly)

in the meantime I'll reduce agan the "Maintain enough work for an additional" parameter


The quota is 2 wu per gpu. The idea is you have one ready to start as soon as you finish the current one. It doesn't take into account your cache settings as far as I know.
____________
BOINC blog

zioriga
Send message
Joined: 30 Oct 08
Posts: 46
Credit: 297,857,425
RAC: 4,578
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13729 - Posted: 30 Nov 2009 | 9:42:16 UTC - in response to Message 13727.

@Markj
So you are telling me that we cannot modify this daily quota; in other words this is managed only at the project manager level.
I think that there should be the possibility to choose how many WU people wants to crunch, may be picking a value between limits


Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1576
Credit: 5,603,736,851
RAC: 8,791,219
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13730 - Posted: 30 Nov 2009 | 9:50:23 UTC - in response to Message 13727.

The quota is 2 wu per gpu. The idea is you have one ready to start as soon as you finish the current one. It doesn't take into account your cache settings as far as I know.

?????

It's not a very good BOINC project if it doesn't respect that pretty fundamental BOINC setting - but I'm sure it does. Zioriga, remember that your cache is the sum of the 'connect about every ...' setting and the 'maintain work for an additional ...' setting - you may have to reduce both.

Quota is a maximum limit, not an obligation. My 9800GT cards only get one at a time, so maybe the answer is to get a slower card..... <grin>. Or did you install a 295 by mistake, thinking it was a 260?

Siegfried Niklas
Avatar
Send message
Joined: 23 Feb 09
Posts: 39
Credit: 144,654,294
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 13731 - Posted: 30 Nov 2009 | 9:59:56 UTC - in response to Message 13727.
Last modified: 30 Nov 2009 | 10:08:06 UTC


...It doesn't take into account your cache settings as far as I know.


A Server-side setting ?

The combination of the account/BM settings:

1. "Network activity always available" (Computer is connected to the Internet about every 0 days)
2. "Maintain enough work for an additional 0 days

tells the BM that "workcache = 0"

Should work like "[B^S] zioriga" described.

(I successfully testet this settings on other projects)




EDIT: Thanks Richard - I noticed your post to late...

zioriga
Send message
Joined: 30 Oct 08
Posts: 46
Credit: 297,857,425
RAC: 4,578
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13733 - Posted: 30 Nov 2009 | 12:52:46 UTC

@Richard

My 9800GT cards only get one at a time, so maybe the answer is to get a slower card..... <grin>. Or did you install a 295 by mistake, thinking it was a 260?


On the other computer where I have an 9800 GT in effect I receive only one WU at a time.
About the GTX295 by mistake, ha ha ha ha, amusing !! But I should have paid it .... much more !!!

By the way now I'll try the Niklas suggestions (0 -> "Maintain enough work for an additional" parameter)

Greetings from another BOINC alpha tester (you can see me as carlo.rigamonti@....)

zioriga
Send message
Joined: 30 Oct 08
Posts: 46
Credit: 297,857,425
RAC: 4,578
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13734 - Posted: 30 Nov 2009 | 13:47:59 UTC

this is the result of the Niklas suggestions:

KO !!!
I received again 2 WU

So, may be Markj is wright !!!

people cannot manage the quota

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13752 - Posted: 1 Dec 2009 | 20:14:57 UTC - in response to Message 13734.

I see you are using a Beta driver.
What version of Boinc are you using with it?

I normally keep my cache to 0.2 days.
My quad core system has a GTX260:
Running Boinc 6.10.18, 19038 driver and thats Cuda 2030 (2.3).

I normally get 1 work unit at a time. sometimes I have one waiting and sometimes not - depending on how far into the task I am. If Boinc thinks I am going to finish the task within 0.2days then it will download another task the next time it connects.

At the minute I only have one task running and none waiting. This is because it thinks it will take 17hours to complete! Its 11% in and has not run for an hour yet, so its not the best at calculating the time remaining early on. By the end it is usually fairly accurate. Somebody else would call this a Time Feature.

zioriga
Send message
Joined: 30 Oct 08
Posts: 46
Credit: 297,857,425
RAC: 4,578
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13757 - Posted: 2 Dec 2009 | 6:17:37 UTC

SKGiven wrote

I see you are using a Beta driver.
What version of Boinc are you using with it?

I normally keep my cache to 0.2 days.
My quad core system has a GTX260:
Running Boinc 6.10.18, 19038 driver and thats Cuda 2030 (2.3).

I normally get 1 work unit at a time. sometimes I have one waiting and sometimes not - depending on how far into the task I am. If Boinc thinks I am going to finish the task within 0.2days then it will download another task the next time it connects.

At the minute I only have one task running and none waiting. This is because it thinks it will take 17hours to complete! Its 11% in and has not run for an hour yet, so its not the best at calculating the time remaining early on. By the end it is usually fairly accurate. Somebody else would call this a Time Feature.



I use BM 6.10.19 with 195.62

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13765 - Posted: 2 Dec 2009 | 20:05:34 UTC - in response to Message 13757.

They are Both Beta's but I dont think that is the cause of the problem.

You would nearly think you had another card in the system!

Is your network usage set to, Connect about every 0.00 days?
And your Additional work buffer set to 0.2 days (or less)?

zioriga
Send message
Joined: 30 Oct 08
Posts: 46
Credit: 297,857,425
RAC: 4,578
Level
Asn
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13771 - Posted: 3 Dec 2009 | 13:58:52 UTC - in response to Message 13765.

SKGiven wrote

Is your network usage set to, Connect about every 0.00 days?

Yes it is.


And your Additional work buffer set to 0.2 days (or less)?

I setted to 0.0 days and quite few hous ago I received only one WU.
At Last !!!!!

Post to thread

Message boards : Graphics cards (GPUs) : How to receive only one WU at a time ?

//