Repeatable bug (Client Error)
21 Oct 2005 17:25:04 UTC
Topic 190041
(moderation:
Hey,
I've been using BOINC and Einstein@home for some time now and I've got that warm and fuzzy feeling from helping out but I discovered something weird some time ago and finally found out what the problem was yesterday.
I like to keep my drivers updated, even with Beta drivers when I need to, and though I don't always reboot after installing those new drivers (for different reasons) I was able to replicate a bug twice because of that.
Here's how to do it:
Update display drivers to latest version but don't reboot. Start BOINC Manager, chose a WU and click on Show Graphics. From what I can see and understand the call is made to display the WU's work being done but the drivers gives an error and thus the client errors out and cancels the WU. I think this is wrong, computation is still working as intended, CPU doesn't make an error and it should still continue working even if the graphics don't display.
Any take on that?
NOTE: I use nVidia drivers (latest Beta for now (81.85) as of yesterday). More info available on request.
Thanks.
EDIT: Forgot to add that only the selected WU will give the Client Error. The other units are unaffected by this.
Repeatable bug (Client Error)
)
You talk about using Beta graphics card drivers but you don't seem to be using the Einstein Beta application. If you are trying to view Einstein graphics and you are using Einstein 4.79 with BOINC 4.45 or later, you are vulnerable to a well known bug (no matter how up to date your graphics card drivers are). The Beta app fixes the 4.79 graphics bug. Check the link under More Information on the E@H home page.
Why would I want to use
)
Why would I want to use Einstein's beta client to run while I reboot? That just doesn't make sense. I use those drivers because they fix issues about games I play. It's as simple as that. I'm almost 100% sure it would still error out if I'd use non-beta drivers.
The fact that the client gives error when trying to display the graphic output from an inbetween installation (not rebooted) shouldn't be relevant to the client I'm using.
All I'm saying is the error message is wrong. The computation is correct. The error is given because the computer haven't been rebooted after changing the drivers. Think about it, it only errors out AFTER/WHEN you click on Display. If the computation would be erronous, I would get errors without having to click on the button.
MadJack, What Stick was
)
MadJack,
What Stick was referring to is a well-documented problem, a conflict with the combination of v4.45 Boinc and v4.79 of the Einstein app. This problem occurs often when someone runs the Einstein screensaver or clicks to "Show Graphics", and it errors out the Workunit. This behavior is entirely independent of your vidcard driver. If you haven't had a WU fall victim to it yet, consider yourself fortunate. You will find complaints about this problem all over the message boards here, if you do a message search for "graphics", and also the recommended fix, which Stick has suggested to you - upgrading to the 0.18 beta version of the Einstein app. This beta has been tested pretty thoroughly for the past six weeks or more, following it's predecessor, v0.03 beta. The only downside found so far is that it slightly lengthens crunchtimes, varying by which platform it runs on, from 1-2% on Intel silicon to around 10% on my OCed AMD Barton. If you decide to give it a try, we'd be happy to have your feedback. I personally tested 0.18 for a week or so, gave my report, and then reverted to my 6-month-old v4.37 beta (which didn't have the graphics conflict), solely because I never use the "Show Graphics" or the screensaver, which I long ago deleted from my Windows/System32 folder, and I didn't want the 10% crunchtime penalty I'd seen.
What you opt to do is your concern entirely. You reported the issue as an Einstein problem, and you have been given the facts concerning the graphics issue with the Boinc/Einstein combo you crunch with. Whichever way it goes, your contribution to the science of the Einstein project will be appreciated.
Regards,
Michael
P.S. - Nice rig you have there. I'm going to be building one along that line soon.
microcraft
"The arc of history is long, but it bends toward justice" - MLK
Hey Micheal, That's the
)
Hey Micheal,
That's the thing. I'm not using the screensaver. This problem will only happen after I update my video drivers and I hit Display before I reboot from the Manager. Except for that, the app works flawlessly and it works fine if I reboot the computer and click Display. I guess I should've said I wasn't using the screensaver.
Anyway, I don't know if it's a bug or not in that case since I'm not using the screensaver but since I'm rarely looking at the display I might change the app to run as a service instead.
As far as my rig is concerned, thanks :) but that's far from showing everything ;) BTW, I'm slowly climbing the Top Computers list :D Two or three days I wasn't showing up but now I'm at around 750.
On a side note, on WinXP64 the display is not smooth but updates every other second or so, whereas on WinXP, it's smooth though it might be because I'm on SLI but I doubt it. Not an issue really since I rarely look at the graphic representation of E@H.
Anyway, thanks for posting :)
RE: Why would I want to use
)
Sorry, I misread your original post.
RE: Sorry, I misread your
)
No probs Stick :)
I guess I could've have been clearer... *shrug*
As I said, I'm just wondering about the error reported by the client. The way I see it, this is a false positive on the client side and it saddens me because both WU it barfed on were over 95% done. One was at 5 mins to completion :(
I'll make sure not to click on Display from now on, just to be on the safe side :)
RE: RE: Sorry, I misread
)
Turnabout is fair play and maybe you misread what I was trying to say (also ineffectively). I read somewhere that, because Einstein 4.79 was built before BOINC 4.45, it is out of date and not totally compatible with it. That is, BOINC 4.45 uses the "new" BOINC Library - changed from when Einstein 4.79 was built. The Einstein Beta, which uses the new BOINC Library, seems to clear up "Client error" problems related to graphics. It could be that you have just found a new way of triggering the same old problem.
MadJack, Does this happen
)
MadJack,
Does this happen every time you run the Display Properties applet or just when you update the drivers?
Walt
RE: Turnabout is fair play
)
Oh I see now! hehe
Well, to be quite frank, it is very possible though my knowledge of BOINC and E@H is very basic at best. We might be onto somethere here... .: looks left and right :.
RE: MadJack, Does this
)
Hey Walt,
Here's how I was able to do it:
Install a new display driver package from nVidia but don't reboot.
Double-click the BOINC Manager icon in the tray to bring the application up. Click on the Work tab, chose a running Project and finally click on Show Graphics button on the left hand side. If it behaves the way it did here, you should get an error from nVidia drivers and a client error msg scrapping your WU. Note that I haven't tried to get a similar result any other way.
If you look at all the WU I have crunched you will find two client errors that were made with the above method.
HTH