Hello!

I'm working on a custom Game Engine and I've noticed some odd performance issues when drawing objects with a large amount of triangles that I cannot explain, and I need some tips on how I can narrow the problem down.

I've set up some very simple shaders and a scene with a single object (600k triangles). The object itself is off-screen, i.e. no fragments are being written. I have both an integration for Vulkan and for OpenGL and I've compared the exact same situation between the two. Here are the results:

Vulkan

Baseline (just post-processing, no objects being drawn): 1.63ms per frame

Drawing the object: 1.9ms per frame

These measurements are about what I would expect. Now, in comparison, with my OpenGL implementation:

Baseline: 2.06ms per frame

Drawing the object: 13.2ms per frame!

The baseline framerate is again approximately what I would expect, but clearly there is something fishy going on when the object is being drawn.

I've tried running NVIDIA Nsight to find out what's going on:

(I've tried using a non-instanced draw call, it made no difference.)

It confirms that:

a) No fragments are actually being drawn

b) The draw-call is definitely the core issue with a whooping 8.85ms

c) It's a GPU bottleneck and not a CPU problem

Here are the shaders themselves:

Vertex shader:

#version 440

#extension GL_ARB_separate_shader_objects : enable
#extension GL_ARB_shading_language_420pack : enable

layout(location = 0) in vec3 in_vert_pos;

void main()
{
	gl_Position = vec4(in_vert_pos,1);
}

Fragment shader:

#version 440

#extension GL_ARB_separate_shader_objects : enable
#extension GL_ARB_shading_language_420pack : enable

layout(location = 0) out vec4 fs_color;

void main()
{
	fs_color = vec4(1,1,0,1);
}

The C++-code is fairly complex, so posting it here wouldn't do any good.

I have no idea what to even look for, so I'm not sure what other details I could provide, but here's some general information about the OpenGL version and the states:

The vertex buffer is created using glCreateBuffers and initialized with glNamedBufferStorage (with the GL_MAP_WRITE_BIT | GL_DYNAMIC_STORAGE_BIT flags). The vertex data is only being written once.

All I need is some hint for what could be causing this issue, so I have an idea what to look for. Thank you!

can u easily change yr code to use glDrawElements instead and see how this performs ?