Author Topic: RG350, RG350M, GCW0 benchmark results  (Read 1111 times)

elea (OP)

  • Posts: 58
RG350, RG350M, GCW0 benchmark results
« on: June 06, 2020, 10:20:59 am »
Here are some benchmark results comparing the different OS's via Retrogame Handheld discord

It looks like the BLITTING results can vary about plus or minus 50 every time the benchmark is run. That means differences under 100 points aren't necessarily significant.

The RG350 is showing results that are 4% to 5% faster than the RG350M. Video driver differences maybe?

Also interesting that the latest GCW 0 firmware scored so much higher. I wonder if that's legit or due to the buildroot? This version of GPMark was also from that time.



GPMark OPK: https://ozgur.kazancci.com/jutleys/02-GCW%20Zero/GCW%20APPS/gpmark.opk

 
« Last Edit: June 06, 2020, 10:27:33 am by elea »

pcercuei

  • Posts: 1667
    • My devblog
Re: RG350, RG350M, GCW0 benchmark results
« Reply #1 on: June 06, 2020, 04:06:42 pm »
With the 5.7 kernel on the GCW-Zero (996 MHz), I get (software):
900.9
485.0
470.0
550.9
297.0
89.8
13.5

elea (OP)

  • Posts: 58
Re: RG350, RG350M, GCW0 benchmark results
« Reply #2 on: June 06, 2020, 04:20:54 pm »
@pcercuei Thank you! Is that with this specific version of the GPMark?

pcercuei

  • Posts: 1667
    • My devblog
Re: RG350, RG350M, GCW0 benchmark results
« Reply #3 on: June 06, 2020, 04:39:52 pm »
Yes, with the OPK linked above.

elea (OP)

  • Posts: 58
Re: RG350, RG350M, GCW0 benchmark results
« Reply #4 on: June 06, 2020, 06:21:34 pm »
OK, excellent. We're on the same page, so to speak.

Since you only posted the software benchmark numbers, I take it you're still working out some stuff with the GPU (or IPU).

pcercuei

  • Posts: 1667
    • My devblog
Re: RG350, RG350M, GCW0 benchmark results
« Reply #5 on: June 06, 2020, 06:28:03 pm »
Since you only posted the software benchmark numbers, I take it you're still working out some stuff with the GPU (or IPU).
No, that has nothing to do with the GPU or IPU. "Hardware surface" means the app renders directly to the framebuffer. "Software surface" means that it renders to an intermediate buffer, that is then copied to the framebuffer. That's why it's slower.

elea (OP)

  • Posts: 58
Re: RG350, RG350M, GCW0 benchmark results
« Reply #6 on: June 06, 2020, 10:19:22 pm »
Ohhh, sorry! I completely glossed over that part of this discussion:

https://boards.dingoonity.org/dingux-releases/gpmark-for-opendingux/msg48095/#msg48095

So yeah, 3d acceleration tests have never been implemented in GPMark.

Ninoh-FOX

  • Posts: 104
Re: RG350, RG350M, GCW0 benchmark results
« Reply #7 on: June 07, 2020, 05:47:46 am »
the differences between RG350M and RG350 not is exactly the video driver, this is the same, is the panel driver, the driver for lcd screen.

elea (OP)

  • Posts: 58
Re: RG350, RG350M, GCW0 benchmark results
« Reply #8 on: June 07, 2020, 10:14:24 am »
When somebody says "video driver", I think it's usually obvious it includes code that drives a specific display. But it's OK I guess if you really want to consider the panel driver as something separate.

Anyhow, I wonder if there's any way to optimize the way data is processed by the RG350M's 640x480 NV3052C panel?

Or maybe the RG350M just has to have to live with a 4-5% performance decrease for sending 4 times as many pixels to the display.

pcercuei

  • Posts: 1667
    • My devblog
Re: RG350, RG350M, GCW0 benchmark results
« Reply #9 on: June 07, 2020, 12:15:38 pm »
When somebody says "video driver", I think it's usually obvious it includes code that drives a specific display. But it's OK I guess if you really want to consider the panel driver as something separate.
It *is* a separate thing.

Or maybe the RG350M just has to have to live with a 4-5% performance decrease for sending 4 times as many pixels to the display.
The extra overhead is probably caused by memory bus pressure when the data is sent from the IPU to the LCD controller. There is nothing to be done here.

elea (OP)

  • Posts: 58
Re: RG350, RG350M, GCW0 benchmark results
« Reply #10 on: June 07, 2020, 08:45:48 pm »
OK, I gladly stand corrected on the video driver vs panel driver terminology. Learn something new every day!

As for performance, I'll be curious to benchmark the JZ4770-based RG280M with its 2.8" 320x480 display (just like the RG300 2.8" IPS):

<a href="https://www.youtube.com/watch?v=fWYR9h2US0w" target="_blank">https://www.youtube.com/watch?v=fWYR9h2US0w</a>
« Last Edit: June 07, 2020, 08:47:33 pm by elea »