Change Log: soon

---

Changes for 6.6.28

- client: enforce_schedule() wasn't starting GPU jobs
- Update Translations

---

When are they going to update the Linux client? We're still stuck at 6.6.20....

---

FOr me it is going to have to wait a couple of days before I will take a look at it, got another gold badge to get on WCG. So, I am going to go to pretty much single project mode until I get that nailed down. Then when I am ready to try it (maybe we will have 6.6.42) I will be ready to give it some acid.

Even if they have work scheduling fixed (I doubt because they just changed the whole mechanism for doing that) there is still work fetch being messed up along with debt calculations.

Ok, running 6.6.28 on my Mac Pro so no idea if they fixed the GPU not running tasks bug...

BUT, the preliminary look says that Dr. Anderson's surgery may have fixed the long standing task turbulence issue. I have only been watching for a couple hours and so don't know for sure yet. I *HAVE* been fiddling with tasks to force the bug into the open and so far things are working as they should ...

Note the *** SO FAR *** caution.

There may still be some other gottcha ...

I asked for another debug so I could be more certain about the internal decision making process, but so far no hint if they will give that to us ...

FOr me it is going to have to wait a couple of days before I will take a look at it, got another gold badge to get on WCG. So, I am going to go to pretty much single project mode until I get that nailed down. Then when I am ready to try it (maybe we will have 6.6.42) I will be ready to give it some acid.

Even if they have work scheduling fixed (I doubt because they just changed the whole mechanism for doing that) there is still work fetch being messed up along with debt calculations.

So much for waiting a couple of days :)

I was going to try it later tonight on one machine (with GTS250 card).

At the moment I have 3 HIVPR results trying to upload, so will wait until they finish before I download another boinc version. I think I will blow my download limit this month if these HIVPR work units are anything to go by.

---

BOINC blog

FOr me it is going to have to wait a couple of days before I will take a look at it, got another gold badge to get on WCG. So, I am going to go to pretty much single project mode until I get that nailed down. Then when I am ready to try it (maybe we will have 6.6.42) I will be ready to give it some acid.

Even if they have work scheduling fixed (I doubt because they just changed the whole mechanism for doing that) there is still work fetch being messed up along with debt calculations.

So much for waiting a couple of days :)

I was going to try it later tonight on one machine (with GTS250 card).

At the moment I have 3 HIVPR results trying to upload, so will wait until they finish before I download another boinc version. I think I will blow my download limit this month if these HIVPR work units are anything to go by.

Put the horses back in the saddle ...

I am trying it on OS-X because of another fix. I have not put it on my i7 where I USUALLY do testing as it is as fast and is certainly "wider" than the Mac Pro with 2 GTX 295 cards ...

Besides, not much point doing more research, developer have been spending an awful lot of time telling me that 1.4 and .7 delineate a 30% plus minus ...


{edit}

Besides I need YOU to tell me if it works for GPUs ...

FOr me it is going to have to wait a couple of days before I will take a look at it, got another gold badge to get on WCG. So, I am going to go to pretty much single project mode until I get that nailed down. Then when I am ready to try it (maybe we will have 6.6.42) I will be ready to give it some acid.

Even if they have work scheduling fixed (I doubt because they just changed the whole mechanism for doing that) there is still work fetch being messed up along with debt calculations.

So much for waiting a couple of days :)

I was going to try it later tonight on one machine (with GTS250 card).

At the moment I have 3 HIVPR results trying to upload, so will wait until they finish before I download another boinc version. I think I will blow my download limit this month if these HIVPR work units are anything to go by.

Put the horses back in the saddle ...

I am trying it on OS-X because of another fix. I have not put it on my i7 where I USUALLY do testing as it is as fast and is certainly "wider" than the Mac Pro with 2 GTX 295 cards ...

Besides, not much point doing more research, developer have been spending an awful lot of time telling me that 1.4 and .7 delineate a 30% plus minus ...


{edit}

Besides I need YOU to tell me if it works for GPUs ...

Well I installed on a quaddie with a single GTS250 at the moment. So far so good. Hasn't swapped anything out. It appears (to me anyway) that its a little faster, maybe something to do with not having to write the state file as often. If this keeps up I might even try it on my i7 with two GTX260's and see how that goes :) It will have to wait until the two GPUgrid wu that its running have finished though.

As for the sqrt(2) I thought that was quite funny.

---

BOINC blog

Well I installed on a quaddie with a single GTS250 at the moment. So far so good. Hasn't swapped anything out. It appears (to me anyway) that its a little faster, maybe something to do with not having to write the state file as often. If this keeps up I might even try it on my i7 with two GTX260's and see how that goes :) It will have to wait until the two GPUgrid wu that its running have finished though.

There is an oddity. Though it is now more stable in the task selection. I think the speed up is because the code is a little simpler which is all to the good. I still cannot convince them,yet at least, that running the CPU scheduler as often as they do is not a good practice, regardless of how flawless the code. Like talking to the wall ...

I have not tried in on my windows machine yet, and not much point in that it is down to two projects now ... WCG and GPU Grid ... trying to get my gold in the latest sub-project ... already have almost 4 days racked up. Good news, bad news, I also have started to get more MW work for some strange reason.

I may go ahead and install it in the next day or so...

One thing that I did note, and not sure why, is that it can cycle as fast as 3 times per second (according to the logs) but I cannot explain that based on the code. I cannot find any explanation as to why it would do that. Most interestingly is that it does it without a flag signaling a trigger event.

As for the sqrt(2) I thought that was quite funny.

Well, that makes one of us ... :)

Well i've put it on all of the machines, seems to be better than 23 at least.

It still has issues with the resource shares, but thats not likely to change until Dr A concedes he needs to change the way it works.

I seem to get tons of cuda work and run out of cpu work before 24 hours is up. I run an extra cache of 1 day, but only have network access between 1am and 7am.

---

BOINC blog

I just installed 6.6.28 and my 2nd video card was listed as " not used " in the boinc startup messages. It would recognize both the 260 and the 9800 but only use one of them. Went back to 6.6.23 and works fine. Win. XP64. 185.85 Nvidia driver.

I just installed 6.6.28 and my 2nd video card was listed as " not used " in the boinc startup messages. It would recognize both the 260 and the 9800 but only use one of them. Went back to 6.6.23 and works fine. Win. XP64. 185.85 Nvidia driver.

you need to add the flag to force the use of the second card. In 6.6.24 Dr. Anderson put in a test and there is something in the video drivers that causes the second card to have 64K less memory reported. In yoiur cc_config file add:

<use_all_gups>1</use_all_gpus> in the options section.

Actually his cards are indeed of different capability, that's not just the memory size bug. Apart from that you're totally correct.

Now everyone sit back and enjoy the marvelous benefits this change has brought us.

MrS

---

Scanning for our furry friends since Jan 2002

i've had a few gpu wu's to error on this release but, it seems ok.

Actually his cards are indeed of different capability, that's not just the memory size bug. Apart from that you're totally correct.

Now everyone sit back and enjoy the marvelous benefits this change has brought us.

Now, now, now ...

You are starting to sound like that Paul D. Buck character ... and you know how unpopular a guy he is ...

Spotted this over at Seti and thought i'd cross-post it as it may be relevant (originally posted by Richard Haselgrove). Might explain some of the errors people are getting.

It seems likely to happen whenever a running CUDA task is pre-empted, and a new CUDA task started in its place. I think it will happen with every version of BOINC from v6.6.23 onwards, because of a bug with new code added then:

Changes for 6.6.23

- client: for coproc jobs, don't start a job while a quit is pending. Otherwise the new job may fail on memory allocation.

If the job start is delayed to allow the old one to tidy up after itself, the next attempt is treated as a re-start, instead of a new start: that's presumably why the work_unit.sah file isn't copied into the slot directory ready for use.

---

BOINC blog

You are starting to sound like that Paul D. Buck character ... and you know how unpopular a guy he is ...

Thanks for the warning, mate! You know how we all hate him. Finding all those bugs, helping people, caring.. we don't need that here. Seriously, this place is not the way it was since he joined!

MrS

---

Scanning for our furry friends since Jan 2002

<use_all_gups>1</use_all_gpus> in the options section.

Typo warning: The first "gpus" is spelled "gups". End tag doesn't match start tag. I just copied and pasted that as no doubt others will also.

Question: On a multi-gpu rig-does the "1" stay a one or is it changed to the number of gpus?

Thanx!

<use_all_gups>1</use_all_gpus> in the options section.

Typo warning: The first "gpus" is spelled "gups". End tag doesn't match start tag. I just copied and pasted that as no doubt others will also.

Question: On a multi-gpu rig-does the "1" stay a one or is it changed to the number of gpus?

Thanx!

Sorry about that, typing when I am in trouble can cause that ... sigh ...

Just put in a 1 and it skips the tests for the GPU comparison. Why some one thought that was a good way to go is beyond me ... this issue was one of the first I brought up when we were talking about GPU scheduling in that you cannot assume that the GPU resource pool will be like the CPU resource pool.

Even in a configuration with C1060 cards there has to be one video card and that could be a GTX260 or better, but will not be like the C1060 cards at all ...

Looks like there is a start bug for gpu. I have loaded 10 seti beta, the first downloaded file started, after the next it started the other got waiting status etc. Unfortunately the same with gpugrid wu, 2 files downloaded, the newest one started, the oldest one started next, first waited, after some seconds it crashed. I have not seen this in earlier versions <=6.6.24