Quatro with NVIDIA drivers + SHH don't work!

[Beettlle]

I bought 2 computers from Dell with NVIDIA Quatro in them. I'm running RH7.2 and I RPMed the NVIDIA kernel and GLX right after installing. Modified the XF86Config-4 and everything works beautifully on both machines.
The problem is when I try to un an OpenGL app from one machine, over ssh, to the other. When I try to run an app that works fine, on both machines when I'm fisicaly on them, over ssh I get a segmentation fault. Running gdb only point me to libGL.so so I'm guessing it's a driver problem.
Any ideas are greatly apreceated.

Thanks!

[bwkaz]

Well, I don't know that it would cause a segfault, but direct rendering, due to a limitation of XFree86, cannot be enabled unless you're physically at the machine that's running the program. You can't, for example, get a direct rendering GL context if the X server isn't the machine that's running the program.

I don't know why, but I would guess that it's because the network latency would be too unpredictable.

[Beettlle]

I didn't know about that. But I have a nother machine that is the same thing except it has a crappy video card therefore all 3D stuff is done by Mesa and it works fine from there. That is why my guess is that it's the video card or the drivers. So why would the machine with no NVIDIA drivers work just fine and the computer with the NVIDIA card doesn't?

[bwkaz]

With Mesa, you can't get a direct rendering context (AFAIK). So if it's something with direct rendering, then that would explain it.

Of course, I don't know, really, I'm just throwing out ideas. But they sound plausible to me... :p

[Beettlle]

So I ran a little experiment. I have 2 other computers with GeFroce3 cards in them. Runnin RH7.2, NVIDIA drivers, etc. The config of these machines is the same as the ones I mentioned before just diferent vid card.

Now for the amazing part. It works just fine with the GeForce3 cards. :confused: . I have no clue. It might just be something with the Quatro cards.

[bwkaz]

Err... wtf? ;)

Hmm... yeah, could be the Quatro, I guess... that's what it looks like at least...

[kappax]

I am unaware that you could even tunnel 3D apps via ssh and still get HW accel. Because all the data has to be snet via a LAN or the net. therefor it woudl be ultra slow :)

[r0gu3]

It was my understanding that indirect rendering (or tunneling over ssh) wasn't accelerated using nvidia's drivers... it was optimized (ie, all matrix transforms were still fast) but all the work is done on your cpu and none of it on your gpu... where as on the sgi boxes you can get accellerated indirect rendering...

If you need remote hardware accelerated rendering, i have had good luck with chromium(chromium.sf.net) but its really made for a whole lot more...

- r0gu3

[Beettlle]

That's the thing, :) . I am trying to build a display wall with Chromium. It worked just fine with the test cluster using GeForce3's and we thought we'd upgrade to the bigger and badder Quatro cards and Chromium just dies. So I ran gdb on chromium and it told me it was dying on libGL.so so I decided to check to see if I could render stuff through the network(be it via ssh or rsh) and none works, :( ! Can you beleive that? Any ideas?

[jamisondaniel]

LINUX> csh
LINUX > setenv CR_SYSTEM_GL_PATH /usr/lib/tls