Seemes like CUDA is getting to work pretty fine. I'm running CUDA / AP on 4 of my Quads at the moment:

Q1 Q9300@3,3GHz / 2x 8600GT (WXP64)
Q2 Q6600@2,8GHz / 1x 9500-GTOC (WXP32)
Q3 Q9400@3,5GHz / 2x 9500-GTOC (WXP32)
Q4 Q6600@3,3GHz / 2x 8400GS (WXP64)

Using GeForce v.181.20 driver and v.6.02 system tool with <ncpus>5.


It has happend a few things I would like to share because I was told it couldn't happen (and I have some questions too).

Jan14 I noticed on Q4 (computer):
At 20:00 I had only 2 AP and 22 AK_v8 left. Needed more AP so I tried to get it by Changing Additional Workbuffer from 1 up to 9 but I got no extra WU. When I sat 10 in AWB I Downloaded 21 AK_v8 but no AP(!) so now I was pretty soon out of AP.
My config was (is) set to accept S&H Enhanced, AP and GPU.

At the same time I noticed that I was downloading SETI6.0.3 and suddenly the GPU were using v.606 on CUDA-AK_v8 and CPU-core was using v.603 crunching AK_v8 (and the old AP-WU were "Waiting To Run")!
I had all cores crunching AK_v8 at the same time!
Had to check twice but all cores (at the moment 5(?) CPU and 2 GPU) were crunching AK_v8 (all in HighPriority)! Well done!! (This could not happen, I was told)
Well, half an hour later, everything was back to normal (4 CPU-cores crunching AP and 2 GPU crunching AK_v8).

Later on Q4 was using AP v.500 for some AP, and a while later I got AK_v8 running on CPU-core (HighPriority) instead of AP-WU... GREAT it was working fine one more time! (after that I havn't noticed it again). Now running Enhanced 6.03, 6.06 and 6.07

(Jan 15) I try to set preferencies for Q4 not to accept AP and see what happens but I still got 4 AP working and 5 waiting so I won't se any result until tomorrow.

Jan 15:
Q1 had no AK_v8 only 4 active AP and 8 waiting. (ad.workbuffer=4days)
Increased ad.work.buffer to 6 days: got 9 AK_v8 and 2 AP. But 4 of the AK_v8 immediatly finished with calc.error (great - it didn't freeze) and when I ran
findstr /M "<true_angle_range>0.[0-9]" "*.*" > myfile.txt
the 4 errored WU showed up in the list.

At noon all AK_v8 was finished and there was 4 AP working and 10 AP waiting.
1-6 days extra work buffer didn't give any new WU but buffer=7 gave me 8 AK_v8 but all finished in calc.error and all errored WU's was found with:
findstr /M "<true_angle_range>0.[0-9]" "*.*" > myfile.txt
but they already errored out so fast I couldn't run the check first.

I tried again. Buffer=7 gave nothing but buffer=8 requested 121.000sec of work wich gave me 3 AP(!) and 4 AK_v8.
(Now got 13AP and only 3 AK_v8 waiting)
Ran VLAR batch and 2 of the 3 AK_v8 was indicated, so now I don't know how to get enough AK_v8...

Tried up to 10 additional work buffer (in snooze) but that gave no new WU. Quit snooze and I received 19 AK_v8 and 4 AP.
None of the new was found by VLAR batchfile so now I have WU's until tomorrow... :)

That was a small information how hard it could be to get WU’s to feed all cores… (sorry for the long text)

I have not had any frozen/halted WU since I installed Raistmer's_opt_package_V5b. Thanks!


So to my questions:

1. When using the VLAR batchfile wich option should we use. Have seen different ones?
<true_angle_range>0.[0-1]
<true_angle_range>0.[0-9]
<true_angle_range>0.0[0-9]
<true_angle_range>0.00[0-9]

2. Couldn't it be possible to sort these VLAR at the server? It's hard to babysit everytime we download new WU's. And if we ran out of AK_v8 the new can error out immediatly.

3. I have been a little worried about my decreasing RAC (25.000 down to 14.000 since I started CUDA less than o month ago) but I noticed that my pending credits is increasing 15-20.000 credits a day. Yesterday 85.897 and today 102.946.
It's gonna be a LOT at the end of the month... When can I expect to se the result in RAC instead of PENDING?

4. When I checked some of the granted results I noticed that I don't receive my estimated credits for my CUDA-WU. Instead I only get approx 2/3 depending on my wingmen. Why is that?
See a couple:
http://setiathome.berkeley.edu/workunit.php?wuid=393982677
http://setiathome.berkeley.edu/workunit.php?wuid=382280090
http://setiathome.berkeley.edu/workunit.php?wuid=393689441

5. How can we make sure to get enough AP's and AK_v8 to feed our PC continously? On one I was lack of AP and on another I missed AK_v8... (I’m travelling a lot and sometimes I could be out for some days so I can not babysit 24/7)

6. Is there anything else I could check / run to get things smoother?

Thanks!

---

//Peter

Try this one
Its a modded version of the cuda stock app which terminates WU's by itself (no babysitting needed). It also includes a modded version of the AP-stock app. But be warned: Before installing this mod's empty your cache by setting seti to NNT. Is all work done/uploaded and reported, close BM, install, restart, allow new tasks.

Edit: At the end of the linked thread above is a brand new version wich should handle all AR's link !read instructions carefully!

---

Try this one

Thanks but I already installd it and as I wrote

I have not had any frozen/halted WU since I installed Raistmer's_opt_package_V5b. Thanks!

I ment that things are doing fine with that mod but if I'm out of AK_v8 and receives a bunch wich is VLAR and finish in error the server won't give me new ones due to I have consumed the daily quota. Please se my question:

2. Couldn't it be possible to sort these VLAR at the server? It's hard to babysit everytime we download new WU's. And if we ran out of AK_v8 the new can error out immediatly.

The other suggestion

At the end of the linked thread above is a brand new version wich should handle all AR's link !read instructions carefully!

I could give V6 a try but as Raistmer said

But be prepared to VLRT (very long run time) for VLAR tasks.
I'm not sure that it's worth to do VLAR on GPU at all with current algorithm. So, you always can reverse to V5b and leave VLARs to hosts with CPU-based opt apps.

Thats why I would like the server to inhibit distribution of VLAR to GPU-based users if possible.

Thanks!

---

//Peter

[quote]Try this one

Thanks but I already installd it and as I wrote

I have not had any frozen/halted WU since I installed Raistmer's_opt_package_V5b. Thanks!

Ops ;)
English isnt my first language and after this 'big story' it seems i lost some concentration in reading ^^
How ever, the only way i know to get 'more' cuda-wu's is to got to your prefs and mark 'no' ap-work. Einstein could be a alternative for your cpu-cores ;) (but not without loosing some rac-points)

---