Pandora, Some Actual Performance Numbers ..

[krooked]

does this mean we can start to Party?!
/discoparty /dance /dance

 

[JayFoxRox]

http://www.mvps.org/directx/articles/fps_v..._frame_time.htm

GP2x - 20,4ms/F
PSP - 18,8ms/F
Wiz - 14,2ms/F
Pandora - 5,5ms/F

The Pandora is about 2.5 times faster than the Wiz and mora than 3 times faster than the PSP

skeezix: yes, you are right, the PSP is a powerful machine, but the fixed function pipeline limits you in many ways. If the Pandora wouldn't have the ability to use vertex and fragment shaders, I wouldn't have bought it probably.

 

[gaterooze]

yosh64: absolutely, I was only comparing the "stock speed" results between the different devices on a per-mhz basis (not that that really means anything either) as cb88 mentioned that the higher mhz would be expected to be faster - but that's not always the case. I'm definitely not suggesting it correlates to anything else!

 

[Svartalf]

Still no public drivers from Imgtec. We're pretty much at their mercy here. Not that it would have made any difference for an ST emulator.

Actually, you can do hardware blit and scaling with the support there, Chip. We're eating about 10-20% of the peak performance with the software doing the work.

 

[Adventus]

Wow, its good to see some actual numbers that back up what I've been bragging about for a while now. .

The PSP seems to be getting lower than i expected, apparently its got the same bus bandwidth as the Pandora.... so its probably CPU limited. However, The Wiz and GP2X comparison seem to imply a bandwidth limitation.... Maybe its a good test after all.

Thanks Skeezix.

 

[carlgeorge]

skeezix, thanks for taking the time to do these tests. Very interesting.

 

[j.pickens]

Love it, love it.
Best post in some time.
Thanks for sharing the info.

 

[shadow.8]

Thanks for the info, glad to hear our device is doing better than expected!

 

[cb88]

i wasn't even considering the wiz... in my calculations

@gunpei2x while it is not always expected to get higher FPS at higher Mhz.... you are talking about higher FPS on a more advanced processor not an equal or lesser processor so yes my statement was valid

and on top of that I qualified my post with enough maybes and generalitys to make up for my lack of complete understanding of the topic which I do admit though still i probably understand it better than most people if not from a technical POV

 

[GnoStiC]

did you compile it with;
-O3 -fomit-frame-pointer -mfloat-abi=softfp -mfpu=neon -mcpu=cortex-a8 -ftree-vectorize -ffast-math

or what were your flags?

The Pandora is freaking goddamned fast.

 

[cb88]

-O3 -fomit-frame-pointer -mfloat-abi=softfp -mfpu=neon -mcpu=cortex-a8 -ftree-vectorize -ffast-math

I get most of that.... but why -mfloat-abi=softfp ?

heh... I answered myself XD ...except wouldn't hard float be better than any soft float? or do you have to have soft float for neon etc...?

http://gcc.gnu.org/ml/gcc-patches/2008-09/msg00754.htmlQUOTE
@item -mfloat-abi=@var{name}
! @opindex mfloat-abi
! Specifies which ABI to use for floating point values. Permissible values
! are: @samp{soft}, @samp{softfp} and @samp{hard}.
!
! @samp{soft} and @samp{hard} are equivalent to @option{-msoft-float}
! and @option{-mhard-float} respectively. @samp{softfp} allows the generation
! of floating point instructions, but still uses the soft-float calling
! conventions.

 

[lardman]

That just refers to the abi (which means it can then coexist with softfloat code too), the bit that says whether to use soft- or hardware float processing is the next part: -mfpu=neon

This could also be -mfpu=vfp and might be interesting to see what the performance is then.

 

[ldesnogu]

Alas currently gcc does not support -mfloat-abi=hard which could bring some speedup when passing floats to functions.

As far as performance goes, I would say the speedup comes from larger L1 cache, the existence of an L2 cache and faster RAM. Basically the memory subsystem is better.

Skeezix: And for graphics complexity of the ST I agree, though I expect to see how Outcast handles hardware scrolling and border removal demos Also I'd be interested to see profiling results...

 

[fusion_power]

GP2x Wiz @ 533mhz - 70fps

Pandora dev board mk2 @ 500mhz - 180fps

There's a hell of a lot more going on than just pure clock speed, mate

EDIT: To expand on that, these are the fps-per-mhz ratios:

GP2X 49fps/200mhz = 0.245 (this is the only fully optimised version, it should be remembered)
Wiz 70fps/533mhz = 0.131
Pandora 180fps/500mhz = 0.360
PSP 53fps/333mhz = 0.159

Interesting. This says much more than pure speed-comparing because of the different Clockrates of each system. Looks like we could clock the Pandora under 300MHz to get 60FPS in this emulator.

Still no public drivers from Imgtec. We're pretty much at their mercy here. Not that it would have made any difference for an ST emulator.

Instead of waiting till the end of time, I would start to kick some asses until Imgtec starts to rotate! I don't need a Pandora that can't use their Hardware in all of it's glory. And at the Moment it sounds like we will have software-only mode for a VERY long time.

 

[P90Puma]

How about someone write something similar to SuperPi (http://en.wikipedia.org/wiki/Super_PI) and see what the pand can do with number crunching, to provide a _real_ benchmark (at least with regards to processing power).

 

[renegadechic]

This all sounds very intreguing. seeing as how the pandora is looking more and more suitable as a MID i would be quite interested to see how all this compares to a similar test on a netbook like an EEE or an Aspire 1. I dont know if its a particularly valid test but it would be interesting to see if it was closer to them than its gaming competition.

 

[Joshwaan2k]

... you forget these things are not at the same clock and the pandora is designed to be faster

these numbers should be expected

GP2x Wiz @ 533mhz - 70fps

Pandora dev board mk2 @ 500mhz - 180fps

There's a hell of a lot more going on than just pure clock speed, mate

EDIT: To expand on that, these are the fps-per-mhz ratios:

GP2X 49fps/200mhz = 0.245 (this is the only fully optimised version, it should be remembered)
Wiz 70fps/533mhz = 0.131
Pandora 180fps/500mhz = 0.360
PSP 53fps/333mhz = 0.159


I'm really looking forward to a lot of emu's on the Panda. Selling pretty much most of the devices due to the Panda been able to run all my fav emu's. Who knows the Xbox might get the boot hehe.

My personal fav emu I'd love to see some numbers on would be with the MAME emulator. Hope that sucker one day can run my beloved Mortal Kombat 2 full speed, the DSC chip seems to suck the CPU cycles like no tomorrow.

Thanks for taking to time out to provide some insight to what you been up to skeezix, keep it up mate

 

[ldesnogu]

How about someone write something similar to SuperPi (http://en.wikipedia.org/wiki/Super_PI) and see what the pand can do with number crunching, to provide a _real_ benchmark (at least with regards to processing power).

You really think SuperPI is a real benchmark? Its original code is many years old, and doesn't even use latest x86 SIMD instructions. Right now its only use is to test bandwidth, on which it depends a lot.
Anyway for Cortex-A8, I can predict the result: it will be extremely poor since Cortex-A8 is not good at double precision floating point computations.

 

[skeezix]

Alas currently gcc does not support -mfloat-abi=hard which could bring some speedup when passing floats to functions.

As far as performance goes, I would say the speedup comes from larger L1 cache, the existence of an L2 cache and faster RAM. Basically the memory subsystem is better.

Skeezix: And for graphics complexity of the ST I agree, though I expect to see how Outcast handles hardware scrolling and border removal demos Also I'd be interested to see profiling results...

I think the days of OutcaST are perhaps numbered .. it was never a super complete emu; I cut a lot of corners to keep it tight and small (and sitll alot more needs ot be done i that regard.) I don't know much about the bordeer removal and some of those nasty tricks so I never tried to implement them .. I've got partially working HDD emulation and so on, but in the end .. with such horsepower behind pandora, the bigger desktop style emus arenow in scope, so I think Hatari is probably a better overal emu already, and we can work on things like the dewsendants of moron (Saint or STeam or whichever it was, I always forget which is which

I naturally do not use any floats in outcast; it runs 60fps on 130MHz machines!

jeff

 

[ldesnogu]

but in the end .. with such horsepower behind pandora, the bigger desktop style emus arenow in scope, so I think Hatari is probably a better overal emu already, and we can work on things like the dewsendants of moron (Saint or STeam or whichever it was, I always forget which is which

Well SainT is supposedly more accurate, though it's limited to ST and STE, but it's closed source and Windows only.
I will try to get in touch with the author.

PS - I like Hatari more, but making it as accurate as SainT might prove very difficult as it does not handle correctly cycle counting (this requires taking into account two instructions at a time...).