Performance Questions

[Optimus]

In all of my demo effect tests in Caanoo I noticed these speed differences.
Let's take the plasma effect test.

In the very first time I tested it, I was on firmware 1.06 and I got 250fps.
When I upgraded to firmware 1.50 it dropped to 212fps.
Today I realized that after I run something on MAME, quit and then run this test it's raised at 230fps. That's probably because MAME did some RAM tweaking when you load a rom and it didn't return back to original after quitting. When I disable RAM tweaking I don't get this 230.

I downgraded to all firmware and tested. Yeah, it's crazy to change your firmware just for performance testing. 1.01 also does 216fps, the best is 1.06. Funny, when I run MAME with it's tweakings even in 1.06 the value drops to 230.

So, did the firmware 1.06 do some faster ram tweaking at start up that maybe it's unstable and so they decided to undo it in later versions?

I am curious if I should keep 1.06 and not upgrade again to 1.50 yet. What are the advantages of 1.50?

p.s. Apparently I am a performance freak, it would look silly to someone to downgrade the firmware because of this. But I find it important. The difference between 212 and 250 wouldn't matter but even in slow effect tests there is the same percentage of difference. Which is quite important.

 

[AJack]

p.s. Apparently I am a performance freak, it would look silly to someone to downgrade the firmware because of this. But I find it important. The difference between 212 and 250 wouldn't matter but even in slow effect tests there is the same percentage of difference. Which is quite important.

Don't feel silly; as you've said, a loss of nearly 40fps is what, a 15% drop? That's pretty significant and well worth looking into.

 

[Exophase]

As far as I'm aware, GPH changed the clocking setup to be more Wiz like in order to allow for Wiz-style overclocking. However, what I thought was that the Wiz clocking was faster, ie, the memory bus at 1/3rd CPU clock instead of 1/4th.

The best thing to do will be to post the actual clock values, RAM timing values, and LCD timing values (refresh rate impacts available bandwidth and therefore performance). You'll need to get the registers from the Pollux data sheet. I can post them later if no one can now.

 

[Optimus]

Well, I don't know any tech stuff about the internals of those handhelds. I would be interested to see what is happening. Maybe I'll try to get the Pollux data sheet next days and see how can I get those values and what are their meaning. Maybe there are some forum threads where you were discussing those memory timings and stuff?

Ok, I'll give it a shot with the Pollux and hopefully I will be able to understand what is happening sooner or later. I might post those values later or have my benchmark programm report some info. Soonish..