I have a Windows7, 64Bit, i7, 8GB Ram, Geforce 460(GX?) and Tesla C2050.
I am doing structured light on this computer. That means I have a projector connected to Geforce which should project a series of patterns. For instance, grey coded patterns.
It is quite important that the projector doesn’t miss patterns. Mostly the projector stalls, which means the same pattern is projected twice, and then the reconstruction gets out of synch.
I had a lot of issues with this. Bottom line, the projector stalls a lot when I run both CUDA computations on my Tesla and project on the GeForce, but on some other GeForce cards it would stall even without the CUDA computation. I think the GeForce 460 was able to project properly when the CUDA computations were off.
I have no idea how to tackle this performance issue. I have set the Tesla GPU to TCC. I don’t have any display on this computer, so no extra 3D rendering calculation. If I use a seperate computer with only a lousy onboard intel GPU, I get PERFECT projections. No stalling at all.
Is it not possible to get good projection on the same computer? Eventhough I have GeForce 460 and Tesla C2050?
As I said, I don’t know how to deal with this. I did what I could and it still act kind of random. Sometimes the stalls and projection misses is frequent and sometimes it’s less frequent. But it’s never perfect or even good.
For the projections I use DX9 and I vsynch the adapter. I tried using DX10 and OpenGL(acutally SFML), but I got the same results if not worse.
I would appreaciate any idea how to deal with this. What else can I do to “multi task” these two seemingly unrelated operaions?(Projecting and computaitions).