V2 - Benchmarking Jitter's CPU/GPU performance on your computer - V2

@DTR: I am familiar with your solutions, thanks, tried similar things myself. But for my sort of workflow and needs I wasn't able to improve the performance much. Also, once you have 12 cores available (as with my current processor), running 12 Max applications has it's obvious disadvantages. However, I would be very interested to hear which functions of Max use multi-threading - I don't remember ever coming across this info (I know only about the audio version of poly).

I there! I built this test patch years ago (2014?) merely as a demonstration of what could be a benchmarking tool for jitter users with the intent of it being worked on and expanded by the community before being used as a reference point. I feel somewhat "ashamed" that it still being used to this day without any update/modification (aside from the inclusion of an initial @cache_mode attribute that was missing and the use of jit.gl.material instead of the fixed-function rendering pipeline in v2). Here's a quote from the original thread where the idea was initially developed:

Anyway, I think the benchmark should measure at least 4 components:CPU processing:Regular jitter matrix operations Physics processingGPU processing:geometry rendering shader processing

These should be as isolated as possible. The resolution could be in HD (1280 x 720) in order to accommodate both laptops and desktops easily. The routine could be automated, cycling through each test for x seconds and capturing an average of the framerate for each one. These values could be stored in a text file or a GUI that could be easily copy/pasted to the forum including captions.

The 4 tests effectively developed were the following:
CPU: Jitter regular objects usage (CPU only).
GPU Geometry 1: uses many instances of the same geometry (jit.gl.gidshape / jit.gl.multiple)
GPU Geometry 2: uses 3 high poly-count objects and animates them with jit.animdrive
GPU Pixel Shaders: GPU texture processing using a series of jit.gl.slab effects

This test was built in Max 6, before all the Max 7 Jitter functionalities. I think it's time for a community effort in building new benchmark scenes to test things not currently included such as:
GPU video decoding
Jitter Physics
Jit.mo objects (animation)
Render to texture and texture to matrix operations (bringing the rendered images back to CPU/RAM)
MRT (Multiple Rendering Targets) effects such as shadows, motion blur, among others.
...

@Pedro: Thanks for these insights! I asked the question because I've been following the thread for a long time and still don't fully understand what the numbers actually signify in terms of their effect on Max and how it runs on my Mac versus someone else's computer.

I've admired your Max coding - something I've recently gone back to University to study. As a professional designer who now needs to understand help menus and tutorials, I have some thoughts on how help can be more successfully communicated by Graphic Design rather than pure coding - which to non-programmers can be quite opaque. If you or a group of members decide to work on this benchmark together I'd be happy to contribute my design knowledge. That would involve your answering questions about precisely what is being measured and the insights it provides to users.

I was in Porto in 2011 for the Alliance Graphique Internationale (AGI) Congress. Attached is a poster I made for this about my design process. If nothing else it may illustrate the dogged determination of the designer :)

@Bob: Thanks for the compliment, but there are far more knowledgeable members worthy of praise in these forums. Coincidentally, since you've mentioned it in your post, I'm from Porto, Portugal.

Regarding the benchmark, in addition to the suggestions I mentioned above, I also think that a new benchmark should tend to value scene and rendering complexity instead of merely frame rate, because there are other bottlenecks when pushing for very high frame rates (200+) and it wouldn't be a practical scenario anyway...

CPU: 249.5
GPU Geometry 1: 232.4
GPU Geometry 2: 498.2
GPU Pixel Shaders: 200.1

Lenovo P72
Windows 10
Processor    Intel(R) Xeon(R) E-2176M CPU @ 2.70GHz, 2712 Mhz, 6 Core(s), 12 Logical Processor(s)
Ram: 16gb
GPU: NVIDIA Quadro P4200

I had the task manager open during one of the tests and at no time did the cpu or gpu perfomance graph ever get above 40%.

MacBook Pro (Retina, 15-inch, Mid 2015)
Catalina 10.15.6 (19G73) [64-bit]
2.2 GHz Quad-Core Intel Core i7
16 GB 1600 MHz DDR3
Intel Iris Pro 1536 MB

Max 8.1.5

OpenGL Engine: gl2
Video Engine: vddll

OpenGL Engine: gl2
Video Engine: avf
CPU: 213.5
GPU Geometry 1: 25.9
GPU Geometry 2: 79.0
GPU Pixel Shaders: 59.6

OpenGL Engine: gl3
Video Engine: avf
CPU: 186.0
GPU Geometry 1: 14.9
GPU Geometry 2: 72.6
GPU Pixel Shaders: 54.3

Imac (27" 2020)
Catalina 10.15.6
3.1 Ghz 6-core Corei5
32 GB 2667 MHz DDR4
AMD Radeon Pro 5300

Max 8.1.5

CPU: 424.6
GPU Geometry 1: 88.7
GPU Geometry 2: 289.5
GPU Pixel Shaders: 280.7

It is disappointing that GPU Geometry 1 performance of this brand new mac is worse than that of almost a-decade-old PCs. This should be noted somewhere in the Max documentation (that on a Mac with certain Jitter operations, you'll get SIGNIFICANTLY worse performance than on a PC)

Mac mini (M1, 2020)
Chip: Apple M1
Memory: 16GB

CPU: 243.6
GPU Geometry 1: 7.3
GPU Geometry 2: 254.6
GPU Pixel Shaders: 245.8

MacBook Pro 13 from mid 2017, Touch Bar 3.1ghz, Big Sur, eGPU, Radeon 5700 XT

Running on eGPU only:

CPU: 156.3
GPU Geometry 1: 62.6
GPU Geometry 2: 127.5
GPU Pixel Shaders: 295.7

Only internal:

CPU: 155.5
GPU Geometry 1: 26.5
GPU Geometry 2: 89.5
GPU Pixel Shaders: 84.5

Haven't tried on bootcamp but its probably a bit higher there.