Comments on: Instance culling using geometry shaders

By: Dmitry Duka

Dmitry Duka — Fri, 17 Jun 2011 12:02:47 +0000

GTX 465 – 240 fps on average, rarely falling down to minimum of 150 fps for a second. Nice demo

By: Daniel Rákos

Daniel Rákos — Fri, 19 Nov 2010 15:55:24 +0000

Yes, you should do it that way.
Actually that can differentiate an efficient and an inefficient OpenGL app that how much you care about the parallel nature of the client-server architecture.

By: Dude

Dude — Fri, 19 Nov 2010 12:18:59 +0000

Ahh I see, I have always used only one query object. I should have an array of them, one for each block? And maybe I should also do some trivial task between drawing and culling, like drawing the skybox?

And yeah it helps, makes sense now, it was geting me really frustrated, I am so used to thinking that everything needs to be sequential.

By: Daniel Rákos

Daniel Rákos — Thu, 18 Nov 2010 16:03:41 +0000

Sorry, don’t really read the youtube comments.

Actually querying transform feedback is a cheap operation, the only problem is that the query blocks by default until the results are available. So for you it seemed expensive only because you waited for the latecy between the geometry culling pass and the query results retrieval. In practice you should fill that time with other rendering tasks so you don’t have to stall the CPU (and inherently the GPU) by using blocking query readback.
In your case when you have several blocks to cull I would do it in the following way:

for i = 1 to block.count do cull_with_geometry_shader(block[i]) endfor

for i = 1 to block.count do inst_count = get_query_result(i) draw_instanced(block[i], inst_count) endfor
I hope this helps.

By: Dude

Dude — Thu, 18 Nov 2010 15:36:00 +0000

Hey dude, I sent you a message some time ago on youtube, asking for the models you used in this demo, which was quite silly of me because they were actually included along with the binary. Anyways, I have a slight issue: I generate terrain blocks with the GPU using a density volume, and as suchm i have quite a few blocks(4*4 minimum, excluding coarser(larger) blocks for lod), and for each of these is do another geometry shader pass to generate points for instances(currently just a sphere with 75 vertices or so) and I found that doing the query to retrieve the culled instances for each block was quite expansive(performed once for each block I know there to be instances), so I wound up puting all those instances into a huge buffer and doing just one query. Is querying a transform feedback really that expansive?

Btw I ran your demo with my GT9500, and while I can not remember the fps, it ran quite fluently)

By: Daniel Rákos

Daniel Rákos — Tue, 07 Sep 2010 12:08:18 +0000

Since that I’ve been optimizing the algorithm a bit, so most probably also on the 2600XT it would work faster, however I already replaced my video card with a 5770. Anyway it is possible that the algorithm performs better on NVIDIA cards.

By: Yours3lf

Yours3lf — Tue, 07 Sep 2010 10:54:24 +0000

Hello, I’m also from Hungary, I achieved 50-60 FPS on a 8600GT, which is similar to 2600XT I cant understand that 6 FPS

By: Pepa

Pepa — Tue, 17 Aug 2010 09:23:17 +0000

Thanks Erwin (for this code and Bullet). This is my new code:

GLuint shader = glCreateShader(shaderType);

ifstream fileIn(filename);

if(!fileIn.good()){ std::cerr << "Load shader – ERROR: " << filename << std::endl;}

string stringBuffer(std::istreambuf_iterator(fileIn), std::istreambuf_iterator()));

string source1 = stringBuffer;

const GLchar *tmp = static_cast(source1.c_str());
glShaderSource(shader, 1, (const GLchar**)&tmp, NULL);

glCompileShader(shader);

Thanks Daniel for this tutorials. Demo works:

~120FPS from Geforce GTS250, Intel I3 530, Windows 7

By: Erwin Coumans

Erwin Coumans — Mon, 16 Aug 2010 23:50:34 +0000

I had the same problem. This version of loadShaderFromFile works for me:

http://codepad.org/tAt1EoYR

By: Cat

Cat — Fri, 09 Jul 2010 07:09:01 +0000

~400 FPS on a GTX 470