Benchmarking iMX8M-Plus with WebGL FishTank example

[vix]

Is there an official benchmark (in terms of FPS) from NXP for an iMX8M-Plus processoro on https://webglsamples.org/fishtank/fishtank.html with 10 fish and sharks?

This is a screenshot from my laptop

I know there are different SoM maker with the same iMX8M-Plus and different OS makers.

Are there some benchmarks from the users with different SoMs and Operating Systems?

I wonder how much these two things can affect GPU performances.

[Bio_TICFSL]

Hello,

There is no official benchmark to show you the fps numbers we have in the mx8mplus, but for the fishtank you can obtain at least 60fps since is pure opengl.

Regards

[vix]

Hi @Bio_TICFSL 

thank you.

Do you know if FPS can depend on the display resolution?

I mean: the higher is the resolution, the lower is FPS. Or is it independent?

I'm interested in a resolution of 1920x1080.

I don't have an Evaluation Kit for the i.MX 8M Plus Applications Processor in the office at the moment.

Is it possible for you (or someone else) running the official NXP Linux on that board, opening a browser and go to https://webglsamples.org/fishtank/fishtank.html to measure the FPS?

I have a SoM of third party, I get 14 FPS and the answer from the vendor is that this seems the iMX8M-Plus hardware limit.

But your answer is the opposite.

Thanks in advance.

[vix]

Hi @Bio_TICFSL 

do you expect different performances on iMX8M-Mini?

If I'm right:

• iMX8M-Plus has GPU "GC7000 Ultra Lite"
• iMX8M-Mini has GPU "GCNano Ultra"

Thanks

[gary_bisson]

Hi,

Yes the performance depends on the display resolution.

And yes Mini and Plus won't have the same results, Mini having a lower-end GPU the FPS will be lower too.

Regards

[vix]

Hi @gary_bisson 

thank you.

Do you think that a performance of 22 FPS like those shown on this blog post from Foundries.io https://foundries.io/insights/blog/chromium-docker-compose-app-for-imx8/ is the hardware limit for the Mini?

And so 60 FPS is a realistic target for the Plus?

As far as I know, nobody has obtained a performance better than 15 FPS on fishtank example on the Plus.

I trust NXP when it says 60 FPS, but at the moment I have no evidence that this value can be reached.

This is the reason why I ask here.

[gary_bisson]

Hi,

I can't tell you the number, this needs to be tested.

Now, I can ensure you that the 8MP GPU is faster than 8MM, that is for sure. The foundries.io post doesn't mention the resolution that has been used, maybe it was 720p to get those 22fps, so you need to ensure you test at the same resolution.

Also, 8MP supports OpenGL ES 3.x, which might enable more effect on the tests and therefore provide worse fps results although the GPU is faster. To compare apple to apple, you would need to force the test to use only OpenGL ES 2.0 like on the Mini.

Regards

[vix]

Thank you @gary_bisson 

you're perfectly right.

A lot of different settings could be involved, producing different results.

Here is what I get on my Plus, on a display with 1280x1024 resolution:

• 14 FPS on 3D https://webglsamples.org/fishtank/fishtank.html
• 6 FPS on 3D https://webglsamples.org/lots-o-images/lots-o-images-draw-elements.html with 1 item drawn
• 3 FPS on 2D rain shader example (webgl-shaders.com)

From gputop I see that RA (Rasterizer) is up to 100% almost forever, and I don't think this is expected.

Moreover, sometimes, SE (Setup Engine) is up to 100% too.

I use a module without NPU, but with VPU (MIMX8ML6CVNKZAB).

I don't understand if this means tha DPU is present or not https://github.com/nxp-imx/meta-imx/tree/09026f70a64531868ee90ad42f50e9e9cff35d36/meta-imx-bsp/recip...

Because it's not clear to me what is DPU.

But, in any case I think that the above FPS could not be what is expected, and so I wonder where is the bottleneck.

Can the driver be buggy?