MUST update to 190.38 driver .. or else ???

[veebee]

Just saw this mentioned o n the BS shoutbox... anyone else heard this?

apprently must update to (at least) this driver or we wont be able to process WU's ...

[Wang Solutions]

There was some talk of this when the new drivers came out, but since then most people who have installed them have experienced massive errors so at this stage it seems unlikely.

[Furlozza]

Well, until today's lot of new work, with the 190.38 driver the GTS 250 was performing quite well. However, each wu doenloaded ran for 5 secs and then conked it. I think I have CUDA 2.2....... any way of being sure? cause if I have CUDA 2.3 that would also explain the fun and games. Maybe a drop back to the previous driver?

Any help/answers to this mystery would be gratefully accepted and received.

EDIT: of course, this means that I can now play a game or three for a few hours whilst I try and figure this mess out.

[veebee]

well, I'm still using 180.51 on my single remaining Ubuntu x64 box, no probs.. (except it is slow... 9600GT - even though it is overclocked a bit it's soooo slow )

[Dataman]

Well, until today's lot of new work, with the 190.38 driver the GTS 250 was performing quite well. However, each wu doenloaded ran for 5 secs and then conked it. I think I have CUDA 2.2....... any way of being sure? cause if I have CUDA 2.3 that would also explain the fun and games. Maybe a drop back to the previous driver?

Any help/answers to this mystery would be gratefully accepted and received.

EDIT: of course, this means that I can now play a game or three for a few hours whilst I try and figure this mess out.

You need to have both the 190.38 driver and CUDA 2.3 DLL's. If you didn't install 2.3 then you are on 2.2 or 2.1 depending on when you started doing GPU crunching. You can either drop back on your driver or install 2.3 (recommended).  Go here to get them. If you decide to roll back the driver, I would recommend that you drain your queue and report them before doing so. I have had mixed results trying to go back on drivers and BOINC. It's a bit like a dog chasing its tail trying to align the BOINC version, nVidia driver and CUDA, and the OS. 
Good Luck!

[veebee]

Dataman, just had a look there and 2.2 seems to be the latest CUDA release available (at least it is when I[?b] use that link and select Ubuntu 8.04 Linux 64 bit....

[Dataman]

Dataman, just had a look there and 2.2 seems to be the latest CUDA release available (at least it is when I[?b] use that link and select Ubuntu 8.04 Linux 64 bit....

Sorry, I should have seen the machines in your sig. I only have Windows and don't know much about Linux. Does it show 190.38 as the driver? Either way I would use the current combination they show for your version of Linux.

[Furlozza]

As noted elsewhere, I rolled back to 185.38 and everything is now fine. Crunching as before and crisp game play (just gotta clean mouse's area)

I had a look at those files/drivers and remembering the fun and games with Ubuntu, decided to pull head in until such time as it IS necessary.

Probably this time next week *sigh*

[Furlozza]

And have now had a run of "error in computation" using above driver. I know I have the Cuda 2.2 buuuuuuuuuuutt maybe it is too hot inside the box??

[Dataman]

That is not unusual. When I have all 15 cards at GPUGrid I get 1-2 comp errors every day or two. They are those well documented "unknown error" type errors.  biggrin Yesterday I moved the 9800GT's to Collatz. I am still looking at the results but they appear to be getting more credits/hr there even without factoring in the comp errors at GPUGrid. I left the GTX260's at GPUgrid because they can get the bonus credits where the 9800GT's cannot.
The run time at Collatz is 20 minutes on 9800 as opposed to 16-18 hrs at GPUGrid and it is less GPU intensive. The temps have dropped 5-8C on the cards. The only down side at Collatz is their servers seem to go up and down. They went down at least 5 times yesterday but came back up within 5 minutes.  ???

[veebee]

I have had EVERY WU error out on my 9600Gt (Ubuntu OS) the last cpl days.. enuff is enuff.

will try collatz I think.

[Dataman]

I think that is a good idea as then you will know if the problem is project specific or it is with your machine. Good luck!

[kashi]

.....The run time at Collatz is 20 minutes on 9800 as opposed to 16-18 hrs at GPUGrid and it is less GPU intensive. The temps have dropped 5-8C on the cards. The only down side at Collatz is their servers seem to go up and down. They went down at least 5 times yesterday but came back up within 5 minutes.  ???

Yes when I had a little test of Collatz on my ATI card it only used 50% GPU load and ran nice and cool. It's Slicker's internet connection that goes down regularly. When the connection is lost and then reconnects the server does not always restart. Sometimes it can be down all day here and doesn't come back until Slicker wakes up in the morning and kicks the server, which of course is night here. "This is tomorrow callin', send those Collatz WUs."

[Dataman]

.....The run time at Collatz is 20 minutes on 9800 as opposed to 16-18 hrs at GPUGrid and it is less GPU intensive. The temps have dropped 5-8C on the cards. The only down side at Collatz is their servers seem to go up and down. They went down at least 5 times yesterday but came back up within 5 minutes.  ???

Yes when I had a little test of Collatz on my ATI card it only used 50% GPU load and ran nice and cool. It's Slicker's internet connection that goes down regularly. When the connection is lost and then reconnects the server does not always restart. Sometimes it can be down all day here and doesn't come back until Slicker wakes up in the morning and kicks the server, which of course is night here. "This is tomorrow callin', send those Collatz WUs."

Thanks Kashi! After doing some "finger & toe math" it looks like I am getting ~400credits/hour @ Collatz as opposed to ~325 at GPUGrid on 9800GT's. I didn't even factor in the comp errors at GPUGrid. I compensated for their server problems by increasing the queued work to 0.50 days from 0.01 days. I moved most of the AQUA CPU's back to WCG and immediately lost almost 50K per day.  :cry2: I am trying to make it up on the GPU side. I wish WCG would get their <Beep> together and get their credits in line with the rest of the DC world. They have good science but are so far behind with their technology.

[kashi]

Well there is a new optimised Collatz application being prepared that promises to speed up GPU processing by a large amount, so perhaps the cards will have a greater load resulting in higher temperatures when it is introduced. It also speeds up CPU computation even more so the rate of credit for GPU work may not increase. This is because the credit is based on a benchmark computer which is a 800MHz P3. I was thinking that it would increase quite a lot but I didn't know then that the CPU application would be optimised greatly also.

Yes I hear what you are saying about WCG, I am doing some now while I can babysit it. The credit just crawls, I've been doing it on and off for over a month now and still can't even scrape into the top 100 for the team. I prefer to be in the top 100 so I don't have to go to the next page on the stats sites. :)