tom w

I'm working on a very large max project that deals with processing multiple video streams, running fluid simulations, particle simulations, blob tracking and so-on, and having just moved the project from my all bells-and-whistles iMac to my MBP retina 15", the performance has taken a massive hit - from 60fps to 25fps in some cases. Nearly all the processing is happening on the GPU, and the even on the laptop the CPU load seems quite low, so I'm assuming the performance problem is to do with the power of the graphics card. I have never paid that much attention to the @type of jit.gl.textures, but wonder if this is something that I should look into? The size of the project means it's going to take some time to jump in and change all the texture types, so before I did this I wanted to find out whether people with experience can tell me whether they've been able to improve the performance of GPU heavy patches by minimising the use of fancy texture types like float32. Will this have an impact? Or am I barking up the wrong tree?

System: MBP Retina Mid 2015, AMD Radeon R9 M370X 2048MB, 16Gb 1600MHz DDR3. running OSX 10.11.6, Max 7.3.2

I'm working on a very large max project that deals with processing multiple video streams, running fluid simulations, particle simulations, blob tracking and so-on, and having just moved the project from my all bells-and-whistles iMac to my MBP retina 15", the performance has taken  a massive hit - from 60fps to 25fps in some cases. Nearly all the processing is happening on the GPU, and the even on the laptop the CPU load seems quite low, so I'm assuming the performance problem is to do with the power of the graphics card. I have never paid that much attention to the @type of jit.gl.textures, but wonder if this is something that I should look into? The size of the project means it's going to take some time to jump in and change all the texture types, so before I did this I wanted to find out whether people with experience can tell me whether they've been able to improve the performance of GPU heavy patches by minimising the use of fancy texture types like float32. Will this have an impact? Or am I barking up the wrong tree?

System: MBP Retina Mid 2015, AMD Radeon R9 M370X 2048MB, 16Gb 1600MHz DDR3. running OSX 10.11.6, Max 7.3.2


performance-and-texture-type

float32 is not really a fancy texture type, it just means the input and output texture values are not clamped between 0 and 1 and have a much higher precision. if you don't need the unclamped values or the precision, you should probably not use float32, as they will affect performance.

if you do need float32 you can try float16 instead, which may reduce the cpu load at the cost of precision.

before you start making changes, i would try and isolate which part of your patch is causing the biggest performance hit.

This is a good time to mention that Jitter would need a probe time debug tool.

Color based or what not , that would help the user see where the heavy process is coming from.

a middle on object could tell us info about the object. like how much GPU memory it's taking and what is the amount of GPU in the entire patch

This is a good time to mention that Jitter would need a probe time debug tool.
Color based or what not , that would help the user see where the heavy process is coming from.
a middle on object could tell us info about the object. like how much GPU memory it's taking and what is the amount of GPU in the entire patch
etc.
you know like you guys did with MSP. 

Thanks for the input Rob - much appreciated. I have been trying to isolate the bits which are causing the bottlenecks, but at the moment it appears to be a general over-working of the GPU and I haven't found a single part which is really causing problems. I'm making some progress though, and will swap out float32s for float16s where I think I can.

Performance and texture type