Hi folks,

seriously, I don't know why 6.6.20 became a recommended version. It has serious issues with work fetching, scheduling and debt handling. We expected problems due to this, but as it turns out another of 6.6.20s flaws is hitting us hard:

There are massive complains about WUs taking forever. This is a bug introduced a few versions prior to 6.6.20 and (supposedly) fixed in 6.6.23. So if you want to run a 6.6-series client, upgrade at least to 6.6.23! UPDATE: 6.6.28 is officially recommended by UCB now.

This version still has issues, so if you don't like these I recommend 6.5.0(*), which doesn't have any of them. The only issue I know of is that it doesn't differentiate between cpu and gpu projects regarding the ressource share, so be sure to put in some balanced numbers. For 4 cpu cores and one gpu give GPU-Grid about 20% ressource share, otherwise your cache will be messed up.

You could also use 6.4.7 (the old recommended version), but this may require using a cc_config.xml to use all cpu cores.. so why bother.

All versions can be found here.

(*) not speaking officially for the project

MrS

---

Scanning for our furry friends since Jan 2002

I recommend 6.5.0...

After running in the never endig WU issue, and consulting some fora, I migrated towards 6.5.0. However, this morning I discovered a new never ending WU. Just like the previous ones, It was a KASHIF job...

I re-installed 6.6.20 and the WU resumed again... I'll have to keep a close eye on it, but I wonder if their is an issue with KASHIF jobs (didn't have any problem with IBUCH WU's)...

I don't know if other projects experience the same issues.

There is a strange problem with some "KASHIF_HIVPR" and "IBUCH_KID" WUs. But so far it hasn't been linked to hanging tasks. There have also been rare cases of hanging WUs before, which could be resolved by a simple BOINC restart. In the case of 6.6.20 the restart doesn't help for long, as far as I understand.

MrS

---

Scanning for our furry friends since Jan 2002

I update to 6.6.23 ans I have always the same failure :/

I update to 6.6.23 ans I have always the same failure :/

c:\cygwin\home\speechserver\gpumd2\src\pme\CPME_cufft.cu

I would understand from this that you are running BOINC under a linux emulator? Or are you just running a linux emulator for some other software?

At any rate, that is going to be an issue. Emulators do not allow driect access to hardware which is needed for CUDA, and secondly, when the video driver is "virtualized" it is going to blow up any running CUDA tasks just as if you used remote desktop to look at your PC from remote locations.

Stop using cygwin and the see if the tasks run to completion.

I've had no issues w/ 6.6.20 on Linux 64-bit. Then again, I only have a measley 9600 GSO....:-( Not even SLI, just the one card.

Mike Doerner

---

I had some hanging tasks. Upgrading Nvidia drivers from 182.06 to 185.85 seems to have cured the problem (and the WUs run a bit faster also).

I had a never ending wu on a boinc 6.4.7

it's a IBUCH_KID

this wu

it took me 5 days to realise, as the computing card is far from where I live and work.

@Paul: the cygwin appears relatively often in error messages, so I think it's related to the compiling machine.

@Michael: the bug solved with 6.6.23 was first described as WUs taking about 4 times longer, not really hanging. So one would have to watch the machine or the RAC closely to spot the problems. And I don't know how different the win and linux builds are, the problem may very well not exist under linux.

MrS

---

Scanning for our furry friends since Jan 2002

@Paul: the cygwin appears relatively often in error messages, so I think it's related to the compiling machine.

@Michael: the bug solved with 6.6.23 was first described as WUs taking about 4 times longer, not really hanging. So one would have to watch the machine or the RAC closely to spot the problems. And I don't know how different the win and linux builds are, the problem may very well not exist under linux.

MrS

Yeah, I just crashed one on my other rig. I was interested in that error I had the other day and was curious if it was crashed because of the game or the task. It appears that if the game crashes the driver, and I just did it again tonight. Then, you can also crash the task.. which I did url=KX20708-SH2_US_8-1-10-SH2_US_8670000_0]in this case;/url]. Now what was interesting to me was the fact that the game is pretty stable in the sense that I played a full scenario through (about 20 some hours) and it did not have a problem.

And, yes I got the:

Cuda error: Kernel [fft_data_swizzle_out] failed in file
'c:\cygwin\home\speechserver\gpumd2\src\pme\CPME_cufft.cu' in line 61 : the launch timed out and was terminated.

Hmm, looks like I toasted a couple more tasks ...

Cuda error: Kernel [copy_mul] failed in file 'com.cu' in line 46 : the launch timed out and was terminated.

Seems to be the error if you crashed the Nvida driver kernel. Not a big deal for me ... this is the slower system so it can afford to not have as many tasks queued ...

Anyway,

p20000-RAUL_pYEpYI1205-0-10-RND6536_0
p10000-RAUL_pYEpYI1205-0-10-RND6116_0

Crashed because of the game killing the driver ... sigh ... this stuff should be easier than this by now ...

If you want 6.6.20 to work good, you must do clean install.

Uninstal your previous version of BOINC

Delete all files and folders of BOINC

Instal 6.6.20

Have fun.

I do this and no problem so far.

---

POLISH NATIONAL TEAM - Join! Crunch! Win!

I do this and no problem so far.

You will ... :)

The problem is not the install. It is the debts. People that do an update install do not get the debts reset and that causes issues. A clean install fixes that. So does an install and debt reset using the cc_config file.

The problem is that the LTD calculations are not correct. Not that they listen to me when I prove a bug, even down to the lines of code ... but I have not found where the issue might be (yet) partly because there have been other pressing issues to chase.

Though it may be time to take another look...

Part of the problem is that there are two parts to the problem and I am not sure which one has the most impact. They have already told me I am wrong about part of the first problem (RR SIm has issues, several issues not the least of which is that it does not model the actual system in the way it will process the tasks, I already proved that once though they have ignored my notes)...

LTD calculations also seem to be hammered and there seems to be issues with it heading negative forever ... requires occasional debt resets, the faster the system the more often the resets...

Well yes i reset the debts a few times a month not sure what it does screw up though but works fine for me.

Hello. Sorry i bad speak english. I used boinc 6.6.20 and my work never end. Why? My system: XP x64, GPU: gf 9600GT
last version nvidea driver

Well yes i reset the debts a few times a month not sure what it does screw up though but works fine for me.

Because you reset the LTD it starts fresh all the time as far as allocating time to projects. Which is not a HUGE deal, but means that if a project has been off line for a bit, you lose the debt you "owe" them and don't balance things properly IAW your shares.

I am pretty sure there are more issues with the calculations though so not sure how well it is tracking teh LTD as it is ... I can't prove it yet and am chasing other issues so, don't want to take on another set of problems yet.

I think there are also major problems with RR Sim but cannot prove them yet... again, have not really started to look at that problem either ... though my preliminary look told me it was hammered ...

somebody will prompt that to me to do? upgrade to the version 6.6.23 ?
How many on time occupy calculations on 9600 GT ?

upgrade to the version 6.6.23 ?

That's part of the first post in this thread.

MrS

---

Scanning for our furry friends since Jan 2002

Greetings! I upgrade BOINC to version 6.6.23, GPUGRID do not work! Writes "Waits for the turn" and calculations do not go! ((
Please write what version of the client the working and what version of the driver approaches is better?
P.S. XP x64, GeForce 9600GT, driver: 182.50
Please help

Greetings! I upgrade BOINC to version 6.6.23, GPUGRID do not work! Writes "Waits for the turn" and calculations do not go! ((
Please write what version of the client the working and what version of the driver approaches is better?
P.S. XP x64, GeForce 9600GT, driver: 182.50
Please help

I am not absolutely sure about the 64-bit version of the 182.50 driver, but I do know that is one of the preferred versions on XP 32-bit.

As far as BOINC goes, 6.5.0, 6.6.23, or now 6.6.28 are the versions that I use and suggest.

I am not familiar with the message you describe. are you sure you do not have the task or project suspended?

can you take a picture of the BM window and post a link to it?

I also don't understand what you mean, but I guess this is the answer.

MrS

---

Scanning for our furry friends since Jan 2002