Pandora, Some Actual Performance Numbers ..


skeezix

Internal Development
Joined
Mar 11, 2003
Messages
8,070
Website
www.codejedi.com
Hey my friends ..

We're all hungry for our release units, and we all know everyone is dieing for more info and news, and we all know the boys are too busy to fill our demands :) But there are a few of us devkit lads on board now, so likely some more info will get leaked.

This is unofficial; I'm not speaking for the devs, and may make errors. I'm not intending to step on the toes of any of the very fine devs for any of the devices (disclosure, I've got dev units of all of them!). Craigs statement about saying something that is true for only 3 days being bandied about for 3 months is ever so true :) It is also tough to post performance numbers without people thinking its a diss to the non-topmost machine. I don't want to say that, since there are other factors (price, portability, screen preferences, add-ons, support, controls, weight, and so on.)

You might recall my past thread http://www.gp32x.de/board/index.php?showtopic=44373&hl= comparing gp2x to GP2x Wiz. See that for some commentary on what is accurate or inaccurate for those tests (ie: optimized code on gp2 versus unoptimized code on the Wiz, including SDL versus framebuffer, meaning the Wiz numbers are artificially lower and the gp2x ones are pretty high, so the Wiz should actually perform higher in relation to the gp2x than I showed.) Defining a 'fair' benchmark is hard since everyone has a different idea (all running at their best or worst? the type of test.. be it math, graphics, bus intensive .. it is easy to have a test that hits a weakness in one but a strenght in another, and is that a fair thing (typical use) or a cheat (atypical use)? Hard questions.)

Suffice to say my number is a _SINGLE_ (yet repeatable) test; a good benchmark is comprised of many benchmark tests so you can get a good picture from many perspectives.. not just a 'throwing this out there' number.

Still, it is a pretty telling number sometimes as well!

All that said..

Given OutcaST (an emu for which many of you have seen and used for some 6+ years from me in various forms), in a emulated CPU-torture test. (ie: Atari ST sitting at the GEM desktop with a stock TOS image will be in a tight CPU grind loop.) Doesn't really hit the ST's graphic system hard or the like, but the ST wasn't really a complex machine for graphics anyway. So it tests the emulation pretty well, in a situation that makes the mu perform worse than average. (ie: If it normally woudl run 60fps for a game, it'd run this GEM situation at 40-45fps at a guess.) Its also easy to auto-boot into, so no keyboard or audio or the like making the test more situational.. its pure out emulation with the machine not even trying to play audio.

GP2x stock F100, 200MHz IIRC. Highly optimized code.

GP2x Wiz dev unit (not final hardware but pretty close); The code is on GPH slow software only SDL. (no acceleration.) 533MHz.

Pandora Mk2 (version 1.1) devboard; 500MHz. Notaz quick SDL port using framebuffer. ie: Not optimized, but I think its prolly better than the GPH version on the Wiz. (GPH stuff tends to be slower for some reason, perhaps they don't compile with optimizatoins on in the compiler, or add debugging stuff, or the mouse support is in the way, or something.)

Sony PSP original FAT version. Not optimized for multi-cpu core, not really using the separate GPU for anything in this test. (ie: alpha blendsing UIs and such are in the code, but not during this test.) Essentially the PSP is chugging along at 333MHz if memory serves.. I don't recall if I'm clocking down or not in this case .. didn't loo back at the code to see.

Latest version of Outcast fro mthe gp2x, converted crappily to SDL on the non-F100 machines.

GP2x 49fps

GP2x Wiz 70fps

Pandora dev board mk2 180fps

Sony PSP 53fps

During actual game emulation the numbers are all higher by about 20-25%, but thats out of scope for this.

It is hard to explain the pandora being so far out in front of the others; my guess is the less congested bus, the processor being a multigenerational gap (the Pandora CPU is a far far newer design than in the GP2x), and the big cache on the CPU in the pandora.

Note my comments above; the 70 versus 49 (Wiz to gp2x) may not seem so impressive in relation to the pandora, but going back to the origiunal post I was pretty impressed. The Wiz _IS_ a snappy beasty (and this is a bit of a torture test, not a typical emu gameplay test!), so it is a very exciting device.

The Pandora is freaking goddamned fast.

jeff
 
Last edited by a moderator:
Wow, and that is at the minimal state of 500mhz? Nice! Any chance of running it at 600 and 900 too? I'd love to see how it compares to itself.
 
Be careful about that (which is why I preluded all that ;) -- the PSP also has a second core (I didn't use), and a separate GPU (not used at all in this test.) The PSP is a pretty amazing piece of kit really, even despite its age. It does show some age though in the GPU I think, though it has plenty of horsepower for its needs IMHO.

jeff
 
Well, whatever gains are made using the PSP's GPU will just be outclassed by the Pandora's GPU, surely?
 
Thanks :D

I'd Love to see how pandora compares with a full overclock. Does the pandora have hardware acceleration of SDL in this test?

Btw...what ever happened to hardware 3D? At one time it was said it would be available in June but we still see nothing running with hardware 3D acceleration.
 
As far as I know, this is a plain naieve port of SDL (no acceleration.)

(I assumed it did have some at first since the numbers are so nutty :)

I think its the cache kicking in, and so on as I say at the bottom.

I am tempted to find a few actual benchmarky benchmarks (good solid numerical processing, mp3 re-encoding, etc) but not a lot of free time to burn.

jeff
 
Stop wasting our time and optimise the damn code!



;) j-k

Impressive - thanks for sharing this.
 
I am tempted to find a few actual benchmarky benchmarks (good solid numerical processing, mp3 re-encoding, etc) but not a lot of free time to burn.

jeff



Do it!

It's probably a better way to compare (number crunching). Ideally, it'd be cool if you found something that worked on:

-iPhone
-PSP
-Old-ish desktop PC (3-4 years old)
-New desktop (less than 1 year old)
-Nintendo DS ( :p )

So we can quickly (and unscientifically) compare them all.
 
Last edited by a moderator:
mazza558 said:
Do it!

It's probably a better way to compare (number crunching). Ideally, it'd be cool if you found something that worked on:

-iPhone
-PSP
-Old-ish desktop PC (3-4 years old)
-New desktop (less than 1 year old)
-Nintendo DS ( :p )

So we can quickly (and unscientifically) compare them all.

After you're done tweaking, compiling, and running a program on five completely different devices I have some laundry that -really- needs to be done, as well as some pretty funky dishes in the sink. Just hit me up once ya get a free minute :D
 
Last edited by a moderator:
well just going from the FPS.... pandora's is 3.6 times greater than the gp2x

so you could underclock to 140mhz and get the same frame rate (maybe)

what is the gp2x clocked at? wiki says 200mhz so you have about 30% better performance due to some combination of HW design improvements and slop in the numbers due to more or less optimization
 
greendots said:
Does the pandora have hardware acceleration of SDL in this test?
No:QUOTE
Pandora Mk2 (version 1.1) devboard; 500MHz. Notaz quick SDL port using framebuffer. ie: Not optimized, but I think its prolly better than the GPH version on the Wiz.
greendots said:
Btw...what ever happened to hardware 3D? At one time it was said it would be available in June but we still see nothing running with hardware 3D acceleration.
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. :)
 
Last edited by a moderator:
... you forget these things are not at the same clock and the pandora is designed to be faster

these numbers should be expected
 
cb88 said:
... 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
 
Last edited by a moderator:
Honestly, I was expecting a '100 or so' .. this just seems so out of kilter! Still, its running right, and not counting wrong (the 'what is a second' count seemed okay, too.)

I could work it over more to be sure, but damn.

jeff
 
Wow that is impressive, thank you for posting. Very interesting even if it is not a fully accurate benchmark. Like fischju2000 said, is it possible to test at 900 mhz ? :)
 
hey

If GunPei2X's numbers are anything to go buy, then at 900 MHz it should hit 324 FPS, obviously this should be taken with a grain of salt, as this assumes the relation scales linearly.

edit
To put this into perspective, if the Pandora ran at 10 MHz, then according to GunPei2X's numbers it would hit 3.6 FPS, which I think is very unlikely, hehe.

another edit
Basically, as the relation is likely not to be linear (you could get a more accurate model of the relationship with the more trials you take, that is run the same test at many different MHz), you could say that those numbers are true only for that instant, that is only when clocked at that MHz used.

cyas
 
Back
Top