My observation, on running the NOELIA_RC3 beta units, is they running faster and are less CPU dependent the NOELIA_RL2 units, but they still need work. On Windows XP, the gpu usage is about 60% to 80%, and by freeing another cpu, this increased by a few points. On Windows 7, the gpu usage is about 30% to 50%, with no noticeable increase is gpu usage when I free up a cpu. Also on Windows 7, gpu usage decreases as the unit crunches, though on XP, I didn't notice the drop.

I just finish crunching a I47R1-NATHAN_tstdhfr6-0-1-RND5218_0 on a Windows XP machine. This had gpu usage at 97%. See link:


http://www.gpugrid.net/result.php?resultid=6463597

My observation, on running the NOELIA_RC3 beta units, is they running faster and are less CPU dependent the NOELIA_RL2 units, but they still need work. On Windows XP, the gpu usage is about 60% to 80%, and by freeing another cpu, this increased by a few points. On Windows 7, the gpu usage is about 30% to 50%, with no noticeable increase is gpu usage when I free up a cpu. Also on Windows 7, gpu usage decreases as the unit crunches, though on XP, I didn't notice the drop.

I just finish crunching a I47R1-NATHAN_tstdhfr6-0-1-RND5218_0 on a Windows XP machine. This had gpu usage at 97%. See link:


http://www.gpugrid.net/result.php?resultid=6463597

I have a couple more things to point out.

WU I45R1-NATHAN_tstdhfr6-0-1-RND9717_0 finished crunching on a Windows 7 machine. The gpu usage was 88%. There was no decrease in gpu usage from beginning to end.

http://www.gpugrid.net/result.php?resultid=6463595

The next thing is when the computers run these NOELIA_RC3 beta units, one after the other over several hours, the gpu usage drops for the later units compared to the previous ones, and the computers (both XP and 7) need to be rebooted, quite frequently (every few hours) to get the gpu usage level back up.

The next thing is when the computers run these NOELIA_RC3 beta units......

The same here (W7 64bit, GTX560Ti driver 310.90, core 6.12.34, i7-3770K one thread free for GPU, CPU process tamed to high). My system doesn't need to be rebooted, suspending and next enabling GPU computing via GUI is enough. I have noticed spontaneous/autonomic(not sure about the right expression) restarts of the acemd.2764.cuda42.exe process; suspend/enable GPU is necessary just after process restarting. I can see restarts by Balloon Message system of Process Tamer.

6 days after changing all my profiles to NOT accept beta WU, they are still being forced on me. This is disgusting behavior, causing my machines to lock up and requiring a reboot. I've had to place a "No new tasks" embargo on GPUGRID until you get your act together. Moderator, please notify us when it's SAFE to resume volenteering for GPUGRID. Dissappointed, Rick

---

Do you have set yes in the "run test apps" check-box? If you do, un-check it.

Basically I was seeing the same issues on my systems:
My W7x64 system's GTX660Ti dropped to 1032MHz as the GPU utilization was not high enough (it's normally between 1175 to 1201MHz). Tasks ran at around 40 to 45% GPU Utilization. Temperatures were cool, memory usage was low, and I also noticed the 'autonomic' driver restarts, that required me to suspend and resume tasks.

Presently things are slightly better:
The GPU is at 1124MHz, utilization is at 53% and I have not experienced a driver crash for a while. I also noticed a very small increase in GPU utilization by 'freeing-up' another CPU thread (~2%), but this is normal and to be expected.

Rick, I haven't been able to replicate your problem, and I can't check your settings. All I can do is suggest the settings to use. If you used a manager such as BAM! or GridRepublic (and I can't tell that either) then that might be the source of the issue, and you might get help from them. However, if this is a bug in the server code it's down to the Boinc developers to resolve. They might be interested in some Boinc log files.

---

FAQ's

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

I've had the "ACEMD Beta" UNCHECKED for 6 days. I'm NOT using ANY manager. The only project I'm having unwanted beta WU's is from GPUGRID. However, I would really appreciate you letting me (us) know when the beta testing has stopped, so I can resume accepting ANY GPUGRID tasks. Thanks in advance, Rick

---

I've had the "ACEMD Beta" UNCHECKED for 6 days.

If skgiven says he can't replicate your problem then that's a pretty good indication you've got the settings wrong. On the other hand, there is a chance he turned off beta tasks just when there were no beta tasks in the queue and mistakenly figured he received no betas because the settings work as intended.

Or he didn't even bother trying to replicate your problem.

Here's what I'm gonna do... I'll turn off betas and then watch some other hosts' task lists. If I do not get betas while they do then that means there is no bug in the server code in use here and you screwed up.

Just to make this interesting and to attempt to force you to double-check your settings.... if it turns out you've screwed up then you have to attach your fastest host to my account via my weak account key and crunch 500,000 credits for me. I'll PM you and skgiven my password, you can verify my settings for yourselves.

If it turns out there is a bug then skgiven's gonna attach his fastest host to your account via the weak account key and crunch 500,000 credits for you. (Actually this is just a proposal, he hasn't agreed to this so far.)

So who's willing to put their money... errmmm credits.... where their mouth is? Do we have a deal, gentlemen?

btw, you're right, I bear no risk and stand only to gain, because I'm the one proposing a way to break the deadlock and I should be duly rewarded for my magnanimous effort

---

BOINC <<--- credit whores, pedants, alien hunters

I've had the "ACEMD Beta" UNCHECKED for 6 days.

Do you have set yes in the "run test apps" check-box? If you do, un-check it.

These two are separate settings.
Unchecking "ACEMD Beta" won't stop the server sending you beta workunits.
You have to uncheck the "Run test apps" checkbox also.

Retvari, you are my hero! I also feel real dumb, because I did not see the run test apps box above the other application choices. Thanks all of you for helping me get it right. I'll begin accepting WU's again, and contribute to the cause. Regards,(and embarassed) Rick

---

Dagorath & Rick, I have nothing to lose or gain either, other than fun so it's fine by me and I'm quite prepared to 'up the ante', significantly!

---

FAQ's

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

Dagorath & Rick, I have nothing to lose or gain either, other than fun so it's fine by me and I'm quite prepared to 'up the ante', significantly!

Lol! I had a hunch you had the confidence to up the ante :)

I apologize for doubting everybody's word and now that Rick has solved the problem I hope y'all can understand why I am a doubter. In one of Rick's earlier posts (message 28338) in this thread he assures everybody he unchecked the "Run test apps?" setting. Now it turns out he did not.

Not saying I'm any better, I've done much the same or even dumber on many occasions. First the knees go, then the eyes, before you know it you have to carry a map with your home marked with a big red X so you can find your way home.

---

BOINC <<--- credit whores, pedants, alien hunters

Sometimes you need to double, triple, quadruple, 'and the next-one' check :)
We all make mistakes, and sometimes repeatedly - It's the nature of the Boinc.
Seen this one a few times, and I'm sure we will see it again...

BTW. Being wrong once in a while isn't a bad thing. I appreciate being corrected, not because I like being wrong, but because I like being right. If and when I'm wrong tell me so I can correct myself and so others can benefit. Ultimately this isn't about me or you, it's about the research, and for us the journey. If there is a problem, there is a solution, and when it gets fixed it's fixed for everyone.
Being a bit of a pious get

---

FAQ's

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

It does go both ways, I wanted to run some Beta Apps on one of my machines, and it turns out for nearly two weeks I forgot to Check the "Send Test Applications " box. Hopefully I start getting some soon, but its my slowest machine, and since the only tasks have been the Huge Long Runs, it only searches for work every two days or so.

---

Besides testing a new application, are we doing any other scientific work with these betas?

Actually yes, WUs in betaqueue right now are the first step of the simulations which will be sent to short queue. So this is already the real thing.

I suspended a queued GPUGrid Beta this morning, allowed a running GPUGrid task to finished and then Resumed the GPUGrid Beta, trypsin_lig_127_1-NOELIA_RC3_equ-0-1-RND4222_0 (still running).
It started at 74% GPU utilization, rose to about 76% and then suddenly dropped to 66%. The power usage was about 65%, GPU temp 46°C with the fans fixed at 75%. Core clock was 1176MHz, which is normal. GPU utilization (W7x64) has subsequently dropped to around 55% and is fairly erratic; varies between 50 and 60% (two CPU threads free).

Thanks for fixing the credits ;)

---

FAQ's

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

I have had a beta WU that completed without error:
6 Feb 2013 | 21:58:39 UTC 7 Feb 2013 | 2:17:47 UTC Completed and validated 12,202.02 2,548.72 13,500.00 ACEMD beta version v6.48 (cuda42).
It's about half time faster on my pc (i7, vista ultimate x64, GTX285) but uses approximately 800 seconds more CPU time. Temperature of the card was 60°C while non-beta it is 70-74°C.

---

Greetings from TJ

I suspended a queued GPUGrid Beta this morning, allowed a running GPUGrid task to finished and then Resumed the GPUGrid Beta, trypsin_lig_127_1-NOELIA_RC3_equ-0-1-RND4222_0 (still running).
It started at 74% GPU utilization, rose to about 76% and then suddenly dropped to 66%. The power usage was about 65%, GPU temp 46°C with the fans fixed at 75%. Core clock was 1176MHz, which is normal. GPU utilization (W7x64) has subsequently dropped to around 55% and is fairly erratic; varies between 50 and 60% (two CPU threads free).

Is this much detail useful for the developers? If it is then I would be willing to create a script that would poll the card for this info periodically, save it to disk and ftp the file to some address. It would have the task name of course and each entry would include % completion, fan speed, load, clocks, whatever info would be helpful. I have a library of Python functions that implement the GUI RPC calls and together with the nvidia-settings and nvcontrol apps anything is possible. Would probably run on Windows too.

---

BOINC <<--- credit whores, pedants, alien hunters

Hi,
we cannot test until the beta queue is cleared.

There is a problem with the new app and it has been difficult test it out.

gdf