Hello, think something wend wrong with my GTX 460 gpu settings.

NATHAN WU , almost about 40 minutes crunching is at 1.62 percent completion.

Will it go faster or have i to change nvidia driver settings ?

I dont know what wend wrong, can u help ?

Greetings from germany :)

Those Nathan tasks do seem to fail more than others. That one of yours has failed on a previous host so best to abort it.

Regards from England

I1R73-NATHAN_FAX3-0-62-RND5587_1 3250204 9 Mar 2012 | 18:20:45 UTC 9 Mar 2012 | 20:05:00 UTC Aborted by user 4,704.47 1,033.44 --- Long runs (8-12 hours on fastest card) v6.16 (cuda31)

The recent tasks from Nate cause the GPU to run slightly hotter. Only a few degrees but probably enough to cause failures on some cards. The GTX590 that it failed on previously would probably be more prone to such failures. This is also likely to cause more downclocking. Anyone overclocking should take this into consideration; if you are experiencing failures or downclocking reduce your clocks and up the fan speed a bit.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

I agree with skgiven, I had to low factory OC of GTX 560Ti CC2.1 a bit more (previously 900 -> 890, now 890 -> 880) and to change a little the cooling curve in MSI afterburner. Problem is with the NATHAN-CB1 series only, the NATHAN-FA series can be run on standard factory OC.

Nate is extended the running time of his workunits just like he promised.
The NATHAN_FAX3 tasks will take more than 12 hours to complete on my GTX480s.

I have a I2R35_NATHAN_FAX3 that I estimate it's going to take 31 h. Normally long units took me before around 7h. I see that the 3 means x3
I see this reasanoble. This way long units are really long.

---

HOW TO - Full installation Ubuntu 11.10

I have a I2R35_NATHAN_FAX3 that I estimate it's going to take 31 h. Normally long units took me before around 7h. I see that the 3 means x3
I see this reasanoble. This way long units are really long.

Which also means that it is impossible for it to be returnrd within 24 hrs so these tasks are not eligible for the bonus, a bit of a backwards step. A penalty, in fact for doing the "long task". There must be a happy medium somewhere.

Which also means that it is impossible for it to be returnrd within 24 hrs so these tasks are not eligible for the bonus, a bit of a backwards step. A penalty, in fact for doing the "long task". There must be a happy medium somewhere.

It's not impossible, if you have a CC2.0 card, and a low (0.1 days) cache.
The long tasks were originally intended for the top end cards (GTX 470, 480, 560 Ti 448, 570, 580, 590), but the long tasks wasn't really long in the past (there were only one 'accidental' exception, which took 12 hours to complete on my GTX 480s).

I have not received one of Nate's FAX3 tasks yet, but Long tasks are supposed to take between 8 and 12h on the top cards (now the GTX580, down to the GTX470 or possibly GTX465).

4h was just silly, and GPUGrid seemed to be slipping away from the 8-12h decision. The result was confusion between what was a long and what was a short task; if a long task only takes 10 or 20% longer it's not really 'long', and yet over-rewards for 20% longer. Originally the Long tasks were ~2.5 times as long as the short tasks (on GTX400 series cards).

The idea of 8 to 12h for top cards means tasks can return even with large cache for the full bonus. Mid-range cards might still finish within 24h, but if not within 48h for 25% bonus. Remember long tasks reward higher anyway. If your card is prone to errors or downclocking when running the long tasks then stick to the short tasks. Lesser cards should only run the short tasks (otherwise there is no point having 2 queues or the bonus system).

Nate has been trying to create tasks that take 8-12h on top GPU's. Hopefully this will help rectify the credit system.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Nate is extended the running time of his workunits just like he promised.
The NATHAN_FAX3 tasks will take more than 12 hours to complete on my GTX480s.

How much over the 12 hrs are they taking on your GTX 480's as my GTX 570 is looking close to 24 hrs plus the couple of hrs in the tray = no chance of returning within 24 hrs

http://www.gpugrid.net/forum_thread.php?id=2883&nowrap=true#23722

According to the above my GTX 570 should not be far behind your GTX 480

Just to inform. My workunit on GeForce GTX 560 Ti didn't finish yet. But it's 51% with 16h already. So it will end sometime 30-31h.

I really don't care about bonus nor credit. But they seem maybe too long with the criteria mentioned.

---

HOW TO - Full installation Ubuntu 11.10

Just to inform. My workunit on GeForce GTX 560 Ti didn't finish yet. But it's 51% with 16h already. So it will end sometime 30-31h.

I don't have SWAN_SYNC variable set. I'm reading the forums to understand what is this about, maybe this is oner reason why is running slow. A few weeks ago I freed 2 cores and that speeded up the task a lot.

---

HOW TO - Full installation Ubuntu 11.10

How much over the 12 hrs are they taking on your GTX 480's as my GTX 570 is looking close to 24 hrs

On my host having an older CPU (Core2 Quad 9650 @ 4GHz) it took 15 hours 29 minutes to finish on a GTX 480 @ 800MHz with SWAN_SYNC=0.
On my host having a Core i7 970 @ 4.2GHz (24x175MHz) it took 14 hours 15 minutes to finish on a GTX 480 @ 800MHz with SWAN_SYNC=0.

...plus the couple of hrs in the tray = no chance of returning within 24 hrs

To achieve optimum performance, you should set a low cache (say 0.1 days) in your preferences, set the environmental walue SWAN_SYNC to 0, and free up one CPU core (and use Windows XP or Linux, if it's not possible, than choose a desktop profile in Windows 7 which is not using the Aero features).

http://www.gpugrid.net/forum_thread.php?id=2883&nowrap=true#23722

According to the above my GTX 570 should not be far behind your GTX 480

In theory it's true, but your systems are not optimized for GPUGrid as mine.

If a GTX 570, i7 990x @ 3.88GHz system which is optimised to a large extent cannot achieve a return within 24hrs then I think too much is being asked, as I need to use 1 of my machines for other mundane things ... the unit is too long.

How does the project benefit by cutting out such a high % of hosts. Why not have a long task as posted by skgiven "Originally the Long tasks were ~2.5 times as long as the short tasks". That would allow for many more hosts to participate.
Why so long? it's gone from sublime to rediculous.

Even with your systems being as optimised as they are, you not achieving the 8-12 hrs for a long task. The unit is too long.

If a GTX 570, i7 990x @ 3.88GHz system which is optimised to a large extent cannot achieve a return within 24hrs then I think too much is being asked, as I need to use 1 of my machines for other mundane things ... the unit is too long.

I would say, it's annoying to loose the 24h bonus, but after all I'm contributing to this project to aid science, not for earning virtual credits which has no real value in life. The real value of the work we do here does not change if it's not bonused or not.
These NATHAN_FAXs will wore out soon, so the chance of receiving two in a row will be minimal.

How does the project benefit by cutting out such a high % of hosts.

This is a misinterpretation, but some of the hosts that were capable of processing the so-called long workunits will miss the 5 day deadline with NATHAN_FAX.

Why not have a long task as posted by skgiven "Originally the Long tasks were ~2.5 times as long as the short tasks". That would allow for many more hosts to participate.

I would prefer this length, too.

Why so long? it's gone from sublime to rediculous.

The Kepler will be released soon, so I think the time of much longer workunits will arrive with this new GPU architecture.

Even with your systems being as optimised as they are, you not achieving the 8-12 hrs for a long task. The unit is too long.

I agree with that. It is an excessive change to make a workunit 4 times longer than the original one. I think a 3 times longer workunit would be fair enough.

It's not about credit or bonus's granted, it's about being as useful to the project as my systems can be. It apparently helps the project to return wu's within 24 hrs (that is why the bonus system was introduced) so why make them so long that even a top end card without so much optimising cannot comply.

I am looking at 2 Kepler cards in the near future but is the 24 hrs going to be out of reach again. It's a bit like reaching a target and having the goalposts moved when you do, until the target is unobtainable.
There has got to be a happy medium.

Thats not counting the higher chance of failures on these wu's, to loose a task thats taken 18 hrs say compared to loosing one after 36 hrs.

I hope my comments are being viewed as constructive criticism, as this is how they are meant.

The Kepler will be released soon

Yes, but not yet and it would take time to be available for mass public.

It is an excessive change to make a workunit 4 times longer than the original one. I think a 3 times longer workunit would be fair enough.

Maybe the mistake was making the change too abruptly. Start with x2, and later go x3. I almost aborted the unit thinking it was corrupt.

Thats not counting the higher chance of failures on these wu's, to loose a task thats taken 18 hrs say compared to loosing one after 36 hrs.

I remember Climate Prediction. Their units were a few weeks longuer. It was really a miracle seeing one finish.

Anyway, I prefer units shorter (x3), but I guess Admin has balanced pros and cons, reducing banthwith I guess it's been in consideration too.
If they take too long I will go back to ACEMD. Let's see in a few days.

---

HOW TO - Full installation Ubuntu 11.10

It's not about credit or bonus's granted, it's about being as useful to the project as my systems can be. It apparently helps the project to return wu's within 24 hrs (that is why the bonus system was introduced) so why make them so long that even a top end card without so much optimising cannot comply.

I am looking at 2 Kepler cards in the near future but is the 24 hrs going to be out of reach again. It's a bit like reaching a target and having the goalposts moved when you do, until the target is unobtainable.
There has got to be a happy medium.

Thats not counting the higher chance of failures on these wu's, to loose a task thats taken 18 hrs say compared to loosing one after 36 hrs.

I hope my comments are being viewed as constructive criticism, as this is how they are meant.

So maybe the answer is to continue with the older units at double the credit and then ALSO come out with newer units, letting us users choose between them, and give three times the credit for finishing the ones in 24 hours. I have a crippled 560 Ti and was finishing the old units just fine in less than 24 hours, my first unit of the new batch is going to take almost 34 hours!!! Less credits for MORE work is not going to keep me here. Science is fine but if it is not at least somewhat efficient it is not worth my money.