PDA

View Full Version : Sub D's crashing a lot...


dotTom
09-19-2004, 08:30 PM
OK, so I load the Rhino model and try subdividing a few times just to stress out the app. Here's what happens (polys/memory), the first entry is just the standard model (I should stress this is done starting modo cold, i.e. no other models loaded or cleared modeling sessions):

4569/297MB
18150/363MB
72600/662MB
Crash/1.73GB

I'm running on a dual 3.6GHz Xeon with 2GB of physical RAM. I'm less concerned about there being an effective poly limit than the app just burning out underneith me. You might think that further sub-dividing a 72600 poly model is an odd thing to do, but what if you hit the 'D' key by accident? Bang, you lose your work. Is anyone else getting behaviour like this? Until Lux get the first patch out (OK it is a 1.0 product and frankly I'm totally in love with the UI and tool pipe) perhaps someone could come up with a script that at least warns as you approach what I take to be some kind of memory barrier.

bradpeebler
09-19-2004, 08:54 PM
Ive not seen such behavior, but I will have Dion give it a heavy dose of testing. I know we have loaded models with well over a million polys before. If it is OK with you, could you email us directly so we can work with you to find out all relevant info?

Thanks!
BP

dotTom
09-19-2004, 09:08 PM
Ive not seen such behavior, but I will have Dion give it a heavy dose of testing. I know we have loaded models with well over a million polys before. If it is OK with you, could you email us directly so we can work with you to find out all relevant info?

Thanks!
BP
Sure thing, you can contact me via my profile or give me an email address and I'll contact you. I'm happy to send over a machine spec, even let you Remote Desktop if you want. I'm running on a DELL Precision 470 with a ATI V3100 graphics card.

JacquesD
09-19-2004, 09:32 PM
I just loaded a 417408 freezed polys this afternoon and with a P4 and 1Go of ram, I still have +- 300 Mo of memory left.
Besides my NVidia driver problem the refeshing is still relatively fast.

http://users.skynet.be/fa008785/heavymodel.jpg

Cheers,

The ripper.

MallenLane
09-19-2004, 09:36 PM
Ripper: Curious what nvidia card/driver do you have? I can't do anything in the program without an eventual crash in nvoglnt.dll, and vertex arrays are already off.

I have a 6800gt

JacquesD
09-19-2004, 09:41 PM
I've got a GForce FX 5700 128 Mb and my latest driver version is 4.36.20.23.00

The ripper.

MallenLane
09-19-2004, 09:47 PM
Well, its really frustrating for me, because I can't use a driver revision that low; they don't have support for the card. I've tried every driver revision that the card can use, w/o any luck. The card works fine in every other program w/o any problems.

I sent in a report to Lux about it, wanting to know if this was a known issue that the program just won't work on a 6800 series card at all, but no response.

Its really hard to justify keeping a program that crashes every 5-10 minutes without any info as to why, or when it might be resolved.

kursad_pileksuz
09-19-2004, 09:57 PM
well you should note that, there are so many cards and drivers outthere. i had a older geforce and to use modo i upgraded to ati 9800 pro, which is not the best card but holds fine so far. crash may not be a modo issue, but a graphics card. since you have told what is ahppeneing to luxology , my suggestion is that just wait a day or 2 for a response from them. during the beta testing people solved their crash problems through graphic cards drivers.

Pazur
09-19-2004, 10:05 PM
i had very dense crashes until i switched back to 42.51 drivers for my GF4. now it's all gone. I know this is not solution for You but it does seem to be drivers issue. from what I heard, Quadro series drivers are free from this bug.

Dion Burgoyne
09-19-2004, 11:17 PM
Yes, the NVidia driver issue is known by NVidia and a fix is on the way, generally speaking turning vertex arrays cleans up the issue considerably. There is still some instability with NVidia however, and the best thing to do is to turn on autosave with a few iterations. Until NVidia can fix the issue...

However I could not repeat the issue of subdividing a model and having a crash, I've got a model that has a million polygons when subdivided and no crash, and this machine only has 1 gig of ram, there may be a video card driver issue with your 3100 are you sure you have the latest drivers?

jangell
09-19-2004, 11:25 PM
Well, its really frustrating for me, because I can't use a driver revision that low; they don't have support for the card. I've tried every driver revision that the card can use, w/o any luck. The card works fine in every other program w/o any problems.

There is a known issue with all consumer (ie: GeForce) NVidia cards using any driver version above 42.51. Unfortunately, many newer cards need the more recent drivers to work.

As Brad posted elsewhere, NVidia is aware of the problem and has a fix in testing. The new drivers should be public in the next few weeks.

-- Joe

AmbiDextrose
09-20-2004, 01:33 AM
If you're using Windows, there's a 2GB limit on every process. It looks like MODO tried to access more than the memory limitation of it's process on a 32-bit operating system. I have this same issue when I try to render large models in Rhino/Flamingo. If the Windows version of MODO can run (and utilize) the 64-bit version of the O/S, that would probably prevent this from happening. Otherwise, you have to be mindful of the amount of memory consumed by the application.

dotTom
09-20-2004, 07:02 AM
If you're using Windows, there's a 2GB limit on every process. It looks like MODO tried to access more than the memory limitation of it's process on a 32-bit operating system. I have this same issue when I try to render large models in Rhino/Flamingo. If the Windows version of MODO can run (and utilize) the 64-bit version of the O/S, that would probably prevent this from happening. Otherwise, you have to be mindful of the amount of memory consumed by the application.Yes, of course it is possible to give an application access to more User mode memory in 32 bit Windows. Assuming the application is cleanly written and doesn't use the top bytes of an address to store flags or something, then you can use the Win32 SDK tool "editbin" with the /LARGEADDRESSAWARE flag. This modifies the binary so the Win32 process loader will split the address space User(3GB)/Kernel(1GB), rather than the usual User(2)/Kernel(2). You also have to specifiy the /3GB switch in your boot.ini. There's more information about this here:

http://msdn.microsoft.com/library/default.asp?url=/library/en-us/memory/base/4gt_ram_tuning.asp

Before anyone rushes off to do this it does require that:

1. LargeAddressAware binaries don't assume anything about pointers, i.e. play silly tricks with them.

2. It's only really useful in a real world sense if you have at least 2GB of physical memory.

3. If you cram the kernel's working space down to 1GB you can impact IO etc.

That said it does have it's uses and our 32 bit app ships with the /LARGEADDRESSAWARE flag set, I'm actually in the process of porting our app to Win64 - my workstation dual boots to Windows Server 2003 Service Pack 1 x64.

In fact it is possible for an app under Win32 to access more than 4GB of memory, but the programming model re-introduces segmented (non-flat) pointers. Anyone who's ever coded under Win16 will have a shiver running down their spine at this point.

The modo issue I'm seeing looks a lot like a 2GB maximum assumption somewhere in the code. Now whether this "bug" is in Lux's code or ATI's drivers I don't know (sort of attaching Visual Studio and stepping into the code - which is probably illegal). For the record I usually buy nVidia and as soon as PCI-Express nVidia cards are widely available and I can get a 6x00 series card I'll be throwing ATI out. I've never heard anything but grief when it comes to ATI and drivers. Their hardware is great, their driver team is either very over worked or needs more "encouragement" :-)

dotTom
09-20-2004, 07:20 AM
Yes, the NVidia driver issue is known by NVidia and a fix is on the way, generally speaking turning vertex arrays cleans up the issue considerably. There is still some instability with NVidia however, and the best thing to do is to turn on autosave with a few iterations. Until NVidia can fix the issue...

However I could not repeat the issue of subdividing a model and having a crash, I've got a model that has a million polygons when subdivided and no crash, and this machine only has 1 gig of ram, there may be a video card driver issue with your 3100 are you sure you have the latest drivers?
I tried installing the latest ATI V3100 drivers from their site and all that did was to introduce redraw artifacts under the mouse cursor and crashed near the 2GB limit (it got up to 1.87GB this time), rock on ATI. So I've rolled back to the drivers that DELL shipped with the box. Both XSI 4 and Maya 6 behave perfectly on this box so I'm kind of at a loss about what to do. I'll keep my models <= 70000 polys until either I get a 6x000 based PCI-Express card or Lux or ATI sort it out. It won't stop me using modo, I just wish it would throw a "can't complete the operation" dialog rather than just crashing and burning.

CGTalk Moderation
01-19-2006, 08:00 AM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.