Strange vertex artifacts on iphone4s, doesn't appear on simulator

Could be an issue with depth buffer precision. Are you drawing with the depth buffer enabled? What are the values for your near/far plane? How far is your sphere from the camera?

What primitive type are you drawing, and what data format are the vertex positions in? What precision are the vertex shader variables involved in transforming the positions?

Is the exact mesh outline exactly the triangle edges, or is there also transparency involved?

Could be that lowp shader variables aren't actually lowp on simulator, which is also dependent on the GPU model on the device.

Definitely looks like there is some problem with the precision of the vertices positions as previous posters imply.

Look at those shading errors, and jaggedness in the inside of the mesh, on that orange edge.

Apart from the data format used, what is the size of it, is it using big or very small numbers in vertices positions?

Wow thanks guys. Turns out I had specified medium precision in my shaders. I changed it to high and the issue was resolved. I didn't realize there was that much difference between the two. Kind of scary actually because mediump really had the verts in the wrong locations. It wasn't an issue on my android phone though. Looking good now though. Much appreciated.

Sometimes it is more about your shader practices than the precision of the position values.
This may reveal that you just have bad coding practices that are simply exposed via medium-precision floats, and raising them to high-precision not only comes at a cost but hides the underlying issue.
For example, do you multiply a single point by multiple matrices or just a single world-view-projection matrix? Unnecessary matrix multiplications on the same point is yet another performance problem that would remain, on top of the cost of higher precision.
What is the range of numbers in use here? You shouldn’t have this kind of issue if your numbers remain reasonably small.


L. Spiro

Sometimes it is more about your shader practices than the precision of the position values.
This may reveal that you just have bad coding practices that are simply exposed via medium-precision floats, and raising them to high-precision not only comes at a cost but hides the underlying issue.
For example, do you multiply a single point by multiple matrices or just a single world-view-projection matrix? Unnecessary matrix multiplications on the same point is yet another performance problem that would remain, on top of the cost of higher precision.
What is the range of numbers in use here? You shouldn’t have this kind of issue if your numbers remain reasonably small.


L. Spiro

I only pass in a single mvp matrix to the shader.

…and the range of visibility and the range occupied by the model?

L. Spiro

My guess is that either the position or texture coordinate attributes are lowp or that one of the variables is. If you'd post the shader code, we could confirm.