2D benchmark

[StephC]

This benchmark is testing 2D performances of pocketpc and is currently in beta test, you could download it

Unzip the file, copy the exe and data.pak in the root directory of your device, launch it (the exe) and return your opinion about it in this forum.

Thanks..

[okku]

[StephC]

[DillRye]

[StephC]

[Digby]

[StephC]

beta 2 is

[Johan]

StephC: What exactly is it you are measuring? If you are measuring raw GX fill backbuffer performance, why not measure it in kB/sec or something similar?

[StephC]

[Digby]

I'm sorry but I disagree. Pure memory bandwidth is exactly the number you want. You also need the cost of a cache miss and some idea of the CPU execution speed.

Just because your test runs 30% slower on an e750 compared with an iPaq (for example), that doesn't mean my code is also going to run 30% slower. However, if you give me the relative difference in raw memory bandwidth I'll have a pretty good idea of how much slower/faster my code will execute.

[StephC]

[StephC]

some results :

h2215


h5555


h3970


asus a620


xda

[Digby]

Memory bandwidth is far more useful to the developer of a blit routine or a 3D pipeline than an end-user "typical" scenario would be.

Knowing that the memory bandwidth is less on one device over another gives me a good indication of *why* my blit routine is slower on that device. The test you've written only tells me that it is slower (provided I write my blit routines the same as you do).

Anway, I certainly don't want to start a big argument over this. I think what you've written is a good thing, especially if I were out shopping for a new Pocket PC device. I'd like to hear what Johan, Dan, Refractor and others think. If there is enough interest in "raw" benchmark data such as memory bandwidth, I'll do a brain dump here.

[refractor]

Haven't got time to comment now (my 45GB boot drive - an IBM "DeathStar" has dropped dead ). Treat this as a placeholder for comments I'll write tomorrow at work.

[pacomac]