A few days ago, I joind GPUGRID because I got a new graphics card: a CUDA enabled GeForce 9400GT (512 MB). As my BOINC also runs 11 other projects (SETI@home since 1999) I have some experience with the BOINC manager. So I wondered why my computer got so sluggish when the GPUGRID WU runs. After reading some threads in this forum, I know now that the 16 shaders in my 9400GT are not that many ... even though that the card runs current 3D games perfectly! But I also read that you need at least 64 shaders or more for the WUs to complete in time. And that's not true: my 9400GT did 10% in 2,5 hours and the WU still has 4 times left until deadline. Even though that it's not much fun using the desktop while GPUGRID is running, it's just not true, that such a card can't complete WUs in time.

I just wish that the CUDA functionality in BOINC would be more load balance friendly as the CPU projects. Using the desktop doesn't make any differency in speed when I have BOINC running (with 100% CPU) or not. But the GPUGRID WUs just hog the videocard so badly that the whole desktop is just not responsive enough anymore.

---

GPUGRID on 64 Bit Linux with GeForce 9400GT (512MB), AMD X2 4850e (2 * 2,5 GHz), 8 GB RAM

... it's just not true, that such a card can't complete WUs in time.

... that the whole desktop is just not responsive enough anymore.

The latter is the reason why it's advised to use a card with at least 50 shaders. The project team knows about the slow down and currently there's not much they could do about this, apart from recommending users not to do it. (see some of the other recent threads on this topic)

And if you read carefully you'll find that most people are not claiming you coudln't finish a WU in time, it's just that you will likely get into trouble with the deadlines. With 128 shaders you need about 12h, with 64 one day and with 32 shaders you'll be around 2 days. With 16 shaders it'll be around 4 days, depending on clock speed. So it may well be possible to finish...

... but imagine your disappointment if e.g. you game for a few hours and miss the deadline due to this and don't get any credit for 4 full days of crunching. We really don't want users to experience anything like this.

MrS

---

Scanning for our furry friends since Jan 2002

BUt this "problem" is easy to fix: why just 5 days deadline? Why not 2, 3 or 4 weeks? Climateprediction has a deadline over a year! ... It doesn't make sense to set a deadline with such a short timeframe. This is science, no commercial just in time product ...

By the way: I can't follow your calculation: my GPUGRID Wu is 9% finished after almost 3 hours. This would be about 35 hours for a WU .... and not 4 days ....

---

GPUGRID on 64 Bit Linux with GeForce 9400GT (512MB), AMD X2 4850e (2 * 2,5 GHz), 8 GB RAM

The deadline is short because the WUs depend on each other. They have to wait for results to get back before they can issue new ones. It's a serial process. When they extend the deadline as you suggest they could have started millions of models in parallel but they may not be able to finish a single one completely within a year. [to put some extreme numbers]
And science also means you work with the results.. you want to analyse them and improve your code etc. The goal is to get work done, not just started.

We'll see about the time when it's finished. By the way, some of the newer WUs have shorter runtimes. May well be you got such one.

MrS

---

Scanning for our furry friends since Jan 2002

Indeed, I would add that the variety of workunits (not including the new shortest ones) already varied alot on times. My very highly clocked 32 shader card had times ranging from 35-36 hours up to the 56 hour range. For workunits that exceeded 48 hours on that card (about 1/2 of all work), the translation would be over 4 days for a 16 shader card. Thus, there are many instances where, even if no other activity is done (e.g., gaming, etc.) a 16 shader card could not complete the work within the deadline. You will need to monitor this carefully and probably abort those longer workunits.

By the way: I can't follow your calculation: my GPUGRID Wu is 9% finished after almost 3 hours. This would be about 35 hours for a WU

On what clock speed are you running that?
I have a 9600GT (64 shaders) at 1600 GHz (not full speed at all) and finish a normal WU in about 25-26 hours (smaller ones less). I hardly can believe that it should only took 35 hours. Wait a day and you can better estimate the runtime.
Maybe you just got one of the smaller WUs...

---

Member of BOINC@Heidelberg and ATA!

Even though that it's not much fun using the desktop while GPUGRID is running, it's just not true, that such a card can't complete WUs in time.

A card with 240 shaders might complete a work unit in less than 5 hours, so it certainly seems possible that an equivalently clocked card with 16 shaders could complete one similar work unit in less than 96 hours.

Though, that doesn't leave much room for slack or longer work units.

But the GPUGRID WUs just hog the videocard so badly that the whole desktop is just not responsive enough anymore.

Ironically, if GPUGRID hogged that video card less, there would be a better chance of work units missing the deadline.

Even a card with 240 shaders, shows a slight hit in display performance. Smooth video playback especially suffers. Suspend GPUGRID on the display card, and everything runs smoothly again.

Don't fear to suspend the offender when you've work or play to do. :-)

19% after almost 6 hours ...thats estimated still less than 35 hours for the WU with a standard 9400GT (400 MHZ, 512 MB).

On the SETI@home global prefs website there is a new setting "Suspend GPU work while computer is in use? Enforced by version 6.7+". That's exactly what I was looking for. But even after upgrading to the latest BOINC 6.6.4 (Linux 64 Bit), this doesn't work. :-(

---

GPUGRID on 64 Bit Linux with GeForce 9400GT (512MB), AMD X2 4850e (2 * 2,5 GHz), 8 GB RAM

On the SETI@home global prefs website there is a new setting "Suspend GPU work while computer is in use? Enforced by version 6.7+". That's exactly what I was looking for. But even after upgrading to the latest BOINC 6.6.4 (Linux 64 Bit), this doesn't work. :-(

6.6.4 is less than 6.7.x ... not implemented yet .. :(

On the SETI@home global prefs website there is a new setting "Suspend GPU work while computer is in use? Enforced by version 6.7+". That's exactly what I was looking for. But even after upgrading to the latest BOINC 6.6.4 (Linux 64 Bit), this doesn't work. :-(

6.6.4 is less than 6.7.x ... not implemented yet .. :(

"Enforced" = enabled by default and not "available since ...".

---

GPUGRID on 64 Bit Linux with GeForce 9400GT (512MB), AMD X2 4850e (2 * 2,5 GHz), 8 GB RAM

"Enforced" = enabled by default and not "available since ...".

That's what the formulation implies, but I think this doesn't make sense.

If a feature is enabled by default or not is decided by the preferences and not by the client version. That's why we create an option in the preferences, after all.

Maybe they mean the BOINC server software 6.7 would have it enabled? But then there'd be no need to tell the users about this.

MrS

---

Scanning for our furry friends since Jan 2002

On the SETI@home global prefs website there is a new setting "Suspend GPU work while computer is in use? Enforced by version 6.7+". That's exactly what I was looking for. But even after upgrading to the latest BOINC 6.6.4 (Linux 64 Bit), this doesn't work. :-(

6.6.4 is less than 6.7.x ... not implemented yet .. :(

"Enforced" = enabled by default and not "available since ...".

In the context of BOINC Manager it means that this setting has no meaning for versions f the BOINC Manager below 6.7.0 ...

Setting it to one value or the other will not change the operation of any BOINC Manager that is not at least at that revision level. So, if I set it, and am running BOINC Manager 5.10.45 the setting will have no effect on the way that the BOINC Manager operates on my systems.

Though the development series may contain this setting in the near future, it is not promised until they actually have 6.7 on the street. History says that the change to the server side takes place a few dot releases of the BOINC Manager before the BOINC Manager gets the code to take advantage of the setting.

Just what I wanted to say :)

MrS

---

Scanning for our furry friends since Jan 2002

Just what I wanted to say :)

All those years tech-writing did not go to waste ... :)

So, I upgraded to 6.10.17 but it still doesn't work: no matter if I use mouse or keyboard - BOINC is continuing crunching ...

---

GPUGRID on 64 Bit Linux with GeForce 9400GT (512MB), AMD X2 4850e (2 * 2,5 GHz), 8 GB RAM

This thread is a bit old - what do you mean? You unchecked "use GPU while computer is in use" but it still doesn't stop?

MrS

---

Scanning for our furry friends since Jan 2002

dentaku, Zero credit in 5 weeks!
Your GPU is not being use here, and you have no tasks in your tasks list (present or past).

If you stop your GPU, CPU tasks keep running - this is to be expected.
To pause CPU tasks as well, use the Snooze feature:
(right click on the Boinc icon in the taskbar and click Snooze).