David says that he is working on the ability for BOINC to run both CUDA and CPU apps simultaneously. It'll be a few weeks before it's available.

No offense, but I think this statement is *pure* rubbish. If CUDA app had been released as independently as AP app was, but still sharing result and other tables with CPU app, it could be done yesterdey not in few weeks time.

BR,

---

Unless they fixed something, I have 2 6.03 MB wu's and 1 6.05cuda MB running at the same time. System temps have not changed and the video card temp has not moved either (45°C). System acts as it did before but with extra processing. Some wu's that normally would not be running are in high priority but whatever. That is something that can be addressed.

Are you by chance running all stock application software? NOT the optimized AK_V8 versions requiring an app_info file? Please be specific about your setup.

Oops sorry. Yes standard 6.4.5 and the nVidia Cuda driver. Everything was quiet, nothing Cuda. After some questions and running the Cuda-Z app, we decided memory may be the trouble (reported 255.688mb). Shortly after it auto downloaded the Cuda apps (6.05, etc) and started running 1 6.03cpu and 1 6.05gpu. I tried changing my settings from 2 cpu's to 3. Sometime after that 2 6.03's and 1 6.05 were running.

Setup
Asus P5K
C2D E6750
2g ram
Asus 8600gt
XP Home

Anything else you want to know just ask. I haven't received any work for a while and the Cuda work that I had is complete. 2 6.03 wu's are running now. I hope things open up some soon as I will be out of work by this evening.

---

I gave it a try last night, a few pages back .

That machine has a Opteron 165 and 2 8800GTS's 320mg flavor in SLi.
I tried for a long time to get an AP and the CUDA going, no dice,
but the 8800 ate threw the WU's pretty fast.
I shut off SLi and tried using them both, which worked for a while,
then nVidia drivers crashed, big surprise there.

Then got random driver crash while it was crunching and lost 6 WU's.

I shouldnt be surprised, been waiting for nVidia to make a decent driver for a loooong time, oh, but what was I thinking, it must be Vista.......heh

---

After some questions and running the Cuda-Z app, we decided memory may be the trouble (reported 255.688mb). Shortly after it auto downloaded the Cuda apps (6.05, etc) and started running 1 6.03cpu and 1 6.05gpu. I tried changing my settings from 2 cpu's to 3. Sometime after that 2 6.03's and 1 6.05 were running.

Setup
Asus P5K
C2D E6750
2g ram
Asus 8600gt
XP Home

Ok, I'm not sure how this is possible, since I would have thought it was supposed to delete the 6.03 stock app when it downloaded the 6.05

I'm a bit confused now since Boinc isn't supposed to be able to do this. Have they fixed something? Do I need to remove the modified 6.05 app and let the stock 6.05 app download? Or is this because the person was still running 6.03 tasks? If so, why did the 6.05 download?

---

After some questions and running the Cuda-Z app, we decided memory may be the trouble (reported 255.688mb). Shortly after it auto downloaded the Cuda apps (6.05, etc) and started running 1 6.03cpu and 1 6.05gpu. I tried changing my settings from 2 cpu's to 3. Sometime after that 2 6.03's and 1 6.05 were running.

Ok, I'm not sure how this is possible, since I would have thought it was supposed to delete the 6.03 stock app when it downloaded the 6.05

I'm a bit confused now since Boinc isn't supposed to be able to do this. Have they fixed something? Do I need to remove the modified 6.05 app and let the stock 6.05 app download? Or is this because the person was still running 6.03 tasks? If so, why did the 6.05 download?

I suspect the sequence was:

1) User has machine running stock v6.03, and has a cache of work "branded" 603
2) User installs BOINC v6.4.5, drivers etc., and makes the install 'CUDA capable'
3) Because of the 255 < 256MB bug, BOINC thinks the machine is not CUDA capable, and continues to service the 603 app.
4) Last night, a server-side fix was applied, making the minimum requirement 254MB
5) BOINC now agrees that the host is CUDA capable, and supplies the v6.05 CUDA app. All new tasks will be branded 605, and specified for running with this app.
6) However, the user still has some left-over 603-branded tasks in their cache, and these continue to run in parallel.
7) Once the last 603-branded task has been completed and reported, the v6.03 app will be declared surplus to requirements, and will be automatically deleted by BOINC.
8) The host will then be a pure CUDA machine, and will become incapable of running more than one MB task at a time. AP or another project will be needed to keep the remaining 1.95 CPU cores warm.

Man I don't know. It just got some new tasks. I have several now including 8 labeled cuda. It is running 1 6.03 and 1 6.05. Screen shot of Boinc manager and Task manager. Last night when they were running in high priority, 2 6.03's were running with 1 6.05. Sorry didn't think about a screenshot then. Hope this works.

http://i258.photobucket.com/albums/hh243/Slow_Target/Image2.jpg

---

Man I don't know. It just got some new tasks. I have 20 now including 5 labeled cuda. It is running 1 6.03 and 1 6.05. Screen shot of Boinc manager and Task manager. Last night when they were running in high priority, 2 6.03's were running with 1 6.05. Sorry didn't think about a screenshot then. Hope this works.

http://i258.photobucket.com/albums/hh243/Slow_Target/Image2.jpg

As expected: older tasks (top of screen) are left-over 603s, and new tasks (bottom of screen) are all 605 - no interleaving. Once all these 603s are finished with, you won't be seeing any new ones.

Thats possible as I don't know if I have received any labeled 6.03 since getting 6.05. I wasn't keeping track of that. However, last night wu's due before the end of the year were in panic mode and 3 wu's were processing at the same time. I wouldn't mind if it ran like normally that but will leave it alone for now.

---

Ok, no more 6.03 wu's but lots of 6.05cuda's and now some AP's. Now is only processing one of each. Hopefully that will be worked out in the future. I have noticed that my video card temp has risen to 57c/134f.Thanks for the help.

Edit: I noticed that Smartdoctor was set to auto fan control(30%). Changed it to Smartcooling and the fan increased speed (85%) and the temp is coming back down.

---

David says that he is working on the ability for BOINC to run both CUDA and CPU apps simultaneously. It'll be a few weeks before it's available.

No offense, but I think this statement is *pure* rubbish. If CUDA app had been released as independently as AP app was, but still sharing result and other tables with CPU app, it could be done yesterdey not in few weeks time.

BR,

That's attractive at first glance, but impractical. If it had been released as a different application, there would have to be splitters producing work for that app as well as the current mb_splitter processes. Much worse, the CUDA results would not be validated against CPU results since only hosts with CUDA capable GPUs would be using that separate application. That's all due to BOINC server limitations, of course, but those limitations will probably never be removed.

                                                                Joe