Pandora Good X86 Equivelant For Testing Pandora Apps


TrevorBradley

Active Member
Joined
Nov 6, 2007
Messages
732
OK, I know this could easily turn into a flame war about Mhz on different chipsets, but please hear me out.

I'm presently working on an SDL game for Pandora, and things are going really well. It runs great on my dual core AMD with GeForce 8800, and fairly well on an older Sempron 2500 with an ATI 9600

I've got machines here for testing anywhere from an Athlon 1200+ with GeForce 2 to a dual core machine with a GeForce 8800. All run on x86 AMD chips. (except one is an old Fujitsu Laptop with 900Mhz Transmeta chip and an ancient ATI graphics chipset with only 8MB of memory). Most machines in the house are dual boot Linux and WinXP.

I can test that my game uses only a certain percentage of the CPU (say 50%) on these machines and *guestimate* that it would run on an equivalent x86 machine of lesser CPU (say half the mhz).

I'm not so sure about graphic card equivalence... I'm not doing anything 3D with SDL, but I am doing a lot of surface blitting and graphics primitives.

I know this isn't a fair question to ask, but without a Pandora or beagleboard in hand I have to make my best guess and go with it:

What is your **best guess** to an x86 machine and graphics chipset of equivalent power to the Pandora (clocked at 500-600Mhz) for running a 2D, 800x480 SDL application blitting images (rarely) and LOTS of simple graphics primitives (SDL_gfx)?

The answer doesn't have to be accurate, but even a vague answer like "An equivelant x86 machine would be less powerful than Pandora/ARM" would be a help.

Also, are there any tools out there for either Linux or Windows that can force an application to use no more than a certain percentage of CPU?

I promise I'll release more details about my project soon. It's a working game at the moment, but there are many features to add and it needs to become much more shiny before I feel comfortable sharing it.
 
If you look at the performance number thread from a while ago, you could assume the OMAPs performance in integer and memory is equivalent to a Pentium III of the same clock. Floating point is still to be determined, but will likely be a fair bit less than a P3.

An AMD 1200+ would be significantly (think 2x) more powerful, especially in floating point.

However this isn't taking into account the other chip components (SGX, DSP) which SDL may be accelerated with at some point.
 
tbh I don't know, but I do know my eeePC is much more underpowered in comparison to my main laptop, and it has the same resolution screen. So I have been ocasionally trying PandoraPanic! builds on that...
 
Trevor Bradley said:
OK, I know this could easily turn into a flame war about Mhz on different chipsets, but please hear me out.
Brave soul there... ;)

QUOTE

I'm presently working on an SDL game for Pandora, and things are going really well. It runs great on my dual core AMD with GeForce 8800, and fairly well on an older Sempron 2500 with an ATI 9600



If your'e not using 3D, the ATI 9600 won't play into it. If you're using 3D, you're already off into the rough on a close comparison.

QUOTE

I've got machines here for testing anywhere from an Athlon 1200+ with GeForce 2 to a dual core machine with a GeForce 8800. All run on x86 AMD chips. (except one is an old Fujitsu Laptop with 900Mhz Transmeta chip and an ancient ATI graphics chipset with only 8MB of memory). Most machines in the house are dual boot Linux and WinXP.


The Transmeta will weigh in at a bit less than the Pandora. Everything else is nearly twice as powerful or better.

QUOTE

I can test that my game uses only a certain percentage of the CPU (say 50%) on these machines and *guestimate* that it would run on an equivalent x86 machine of lesser CPU (say half the mhz).


It's not a good one in your case. The Athlon's a PIV class machine and has much, much hotter hardware FP than the class of machine you're talking about with the Pandora. If you're at 50% utilization on that machine, you're going to find that it's not going to run well on the Pandora without some optimizations.

The Pandora's roughly equivalent to a PIII of the same clock- this means about a PIII 550-600 would be the off-the-cuff, if it runs there, it'll likely run on the Pandora or Beagleboard comparison.

QUOTE

I'm not so sure about graphic card equivalence... I'm not doing anything 3D with SDL, but I am doing a lot of surface blitting and graphics primitives.



Within SDL, you will not see any boost from graphic card to graphic card. If there's accel, it's going to be largely the same within your context.

QUOTE

What is your **best guess** to an x86 machine and graphics chipset of equivalent power to the Pandora (clocked at 500-600Mhz) for running a 2D, 800x480 SDL application blitting images (rarely) and LOTS of simple graphics primitives (SDL_gfx)?



I'd try to see if your game works "okay" on your Transmeta. It's a weaker execution machine than the Pandora is, but if you work there, you'll work fine on the Pandora once you recompile and clean things up.

The closest comparison would be a PIII-550 with a working 2D card under X11 (Even a RagePRO would do...) at the same resolution you're going to get with the Pandora (800x480). Working adequately on less muscle will ensure it. Working on more may present issues once you get your hands on it- you may need to profile at that point to find where you can shave cycles off.
 
Last edited by a moderator:
Thanks for the replies everyone. I'm starting to get a bit worried about the floating point math. I do quite a lot of it... :)

I'll give the Transmeta a try and see if I can optimize against it.

I'm going to release this to some platform regardless, even if it's just PC/Linux desktop.
 
Trevor Bradley said:
Thanks for the replies everyone. I'm starting to get a bit worried about the floating point math. I do quite a lot of it... :)

As I told you on IRC, standard FP is indeed poor on A8. Each FP instruction will take a few cycles and will block any subsequent FP instruction. So we're talking here about something 5 to 10 times slower than a PIII (IIRC, PIII was issuing 1 FP / cycle).
BTW I hope you followed other advices posted here: don't use double precision, it's even worse :p
 
Last edited by a moderator:
I don't think I'm going to be able to get away with integer trigonometry with this game.. :) I think I can safely go back to floats all around, I don't need the level of accuracy I needed in my attempt at a space orbiter... (In this game the largest difference of scale in FP operations is about 10^6). I'll look at converting back.

My early tests on my old Transmeta chip gave me fps of somewhere between 4 and 16, though there's still a lot to optimize. (And it looks as if most of that is with SDL rather than my math) Thankfully a high or consistent fps isn't a requirement for the game, it just makes things more pretty. I could live with 20fps, if I could achieve that.

In the end, it's going to be a port from x86 to A8, if I'm lucky. There's just no way around it. :)

Thanks again guys...
 
It's a very bad idea to try and come up with an equivalent PC spec for a console. I've seen people make this mistake so many times.
The architecture of an x86 is so different from the ARM. Also the PVR architecture is very exotic so I don't think you can compare that to an ATI card etc either.
If you're going to have to do a lot of FP calculation then I think you should have a serious think about how you can utilise the NEON.
 
Last edited by a moderator:
QUOTE
Any idea how the floating point performance of the pandora compares to the psp's VFPU?

hlide would be the man to ask about this but, assuming you can vectorise well, i suspect the VFPU in the PSP is more powerful than VFP-lite + Neon in the Pandora, even with our 2 - 2.5x clock buffer.

VFPU timings: http://wiki.fx-world.org/doku.php?id=general:cycles

NEON Floating Point Timings: http://infocenter.arm.com/help/index.jsp?t...b/BABHDBIF.html

The VFP-lite timings are given above.

The only thing the Pandora really has over it is a better memory architecture.
 
Trevor Bradley said:
What is your **best guess** to an x86 machine and graphics chipset of equivalent power to the Pandora (clocked at 500-600Mhz) for running a 2D, 800x480 SDL application blitting images (rarely) and LOTS of simple graphics primitives (SDL_gfx)?
check this: www.rasterman.com
there is a performance comparison with core2duo which is 10 times faster then omap at 500 Mhz
 
Last edited by a moderator:
Nice to see rasterman has a beagleboard :)

also don't forget to read his notes at the top of that page stating that he rendering using SW not the HW
 
kukacster said:
Trevor Bradley said:
What is your **best guess** to an x86 machine and graphics chipset of equivalent power to the Pandora (clocked at 500-600Mhz) for running a 2D, 800x480 SDL application blitting images (rarely) and LOTS of simple graphics primitives (SDL_gfx)?
check this: www.rasterman.com
there is a performance comparison with core2duo which is 10 times faster then omap at 500 Mhz


Hmm, more like x12. Gives me a decent baseline though. Thanks!
 
Last edited by a moderator:
Back
Top