Mips Vs Arm


Optimization is the 80/20 - you spend 20% doing a quick port, and if that is sufficient, everyone is happy; if you need to potimize, then it takes 80% of the time to do it well; Usually people are happier with the time being better spent on UIs, compatibility (for an emu), etc etc instead. So don't get me wrong.. optimization can be devilishly fun, but .. we're branching into multiple kinds of performance here now: Performance on obvious port, performance after a bit of tweaking, performance after optimization, performance after fully realizing the potential for the hardware, etc.

You can't compare a direct-port to a oiptimized-year-later port, since obviously the optimized one will win out generally..

The GP32 _required_ a tonne of work to get anywhere, since its basicly a low power off the shelf machine, but that doesn't make devs really happy; sure, a few core people will go through and work their tails off, but everyone would've been much happier without _having_ to do that (thoug of course there are always those who will delve deeper _anyway_ since thats half the fun).

Anyway, we've digressed badly :)

Course, if you want to know my philosphy on optimization, thats a whole other rant; I believe the fun is in the devving, and in the output -- I'm goal oriented, but mostly just interested in developing. I don't use the end produxts most of the time.. no time, when theres more to be done ;)

jeff
 
Is the battery situation really that bad for homebrew on the PSP? Someone told me that the battery lasts around 5-6 hours if you play SNES @ 333MHz. Doesn't exactly sound like the "brick in the wall" situation that you're describing? :unsure:

http://psp.ign.com/articles/574/574557p1.htmlmakes for interesting reading esp considering ridge racer isn't using the full 333Mhz

Sony games head Ken Kutaragi has admitted that the PSP's quoted battery life of four to six hours for games is not entirely accurate

I dont think there is anything wrong with 5-6 hours, but I'm tempted to call that a outright lie.

Its a brick. :)

The original game use cpu(222 mhz), gpu(111 mhz) and sound chip(166 mhz).
The emulator use only the cpu and gpu.
 
Last edited by a moderator:
I’m pretty sure it’s impossible to mention 333Mhz on the PSP with out starting a flame war.

My whole point is looking at what emulators are capable of running at full speed on the psp (at 333Mhz) could gives us a little insight into what to expect out of the GP2X *if* we knew how a 333Mhz MIPS processor compared to a 200Mhz ARM processor. Since I can’t find much of anything on that other then the admittedly lame PDA benchmarks I linked to in the first post, I was hoping somebody with a dev board, or practical experience in dealing with MIPS & ARM could give me some insight into the subject. As of right now with what I’m seeing here it sounds like we can expect a 200Mhz ARM processor to perform like a 250Mhz MIPS processor.

Geez… chill people..

Look at that graph and tell me what can be achived from 206Mhz ARM chip
E-2000 (ARM 206MHz) 427
Simpad_SL4 (ARM 206MHz) 340
What I'm seeing is massive differences from chips of the same Mhz of the same type. Greater if choose to move a couple of Mhz either way. So when I say a figure pulled from the ass I mean it.

CASSIOPEIA E-750 (MIPS 200MHz) 330
is that the about the same less.


If the site I found conclusively showed which CPU was faster I wouldn't be here asking questions. The whole point that’s been completely lost, and I don’t even care if it’s an irrelevant questions- is which is faster per MHz, MIPS or ARM?

Yes, optimization is better
Yes, the supporting hardware will make a massive difference if software is optimized for it
Yes, The benchmarks I found are fairly meaningless

My simple questions answer will give a good indication of what we can expect from quick straight ports of existing software. This is a short sighted what can I expect in beginning of the GP2X’s life span before all the pure assembly written optimized to the extreme software is written. We know how well the PSP is doing with fairly un-optimized software so I choosed it for a basis of comparison.
 
Last edited by a moderator:
There is not really a performance difference between
MIPS r4000, and
ARM9 cpu core.

Both cores can do the same things, in the same time.
So a 200Mhz arm9 is NOT faster, and not slower than a mips r4000 core.

They are ~ 1:1

It more depends on the external bus speed, and how fast the machine can access the memory.

PSP: ddr ram, 32Bit access, connected 256Mbits.
Samsung 1st generation MCP 3.0V/2.5V 32MB 8 bit Uniform Block NAND Flash + 32MB 32 bit 6ns CL3 DDR.
The cpu/ram bus, runs with up to 333Mhz

gp2x: sd-ram ( not ddr-ram ), 133Mhz, 128Mbits.

So from the first view, the PSP ram is about 3x faster than the gp2x ram.

Conclusion. The PSP is at the same CPU clock, ~2x faster.


BTW: Did you know that the PSP also offers a SECOND Programmable CPU CORE ?

CPU1 in the PSP is the main r4000 cpu, with cache.
CPU2 in the PSP is the Media CPU, r400 core, WITH FULL CACHE.
Both cores are programmable, and can run different code.

So the gp2x offers 2x200Mhz, but the PSP offers 2x333Mhz...

So, from the point of speed, the PSP is the winner.
From the Point of "selling" interest, the gp2x is a winner.


cu,
 
There is not really a performance difference between
MIPS r4000, and
ARM9 cpu core.

Both cores can do the same things, in the same time.
So a 200Mhz arm9 is NOT faster, and not slower than a mips r4000 core.

They are ~ 1:1

It more depends on the external bus speed, and how fast the machine can access the memory.

PSP: ddr ram, 32Bit access, connected 256Mbits.
Samsung 1st generation MCP 3.0V/2.5V 32MB 8 bit Uniform Block NAND Flash + 32MB 32 bit 6ns CL3 DDR.
The cpu/ram bus, runs with up to 333Mhz

gp2x: sd-ram ( not ddr-ram ), 133Mhz, 128Mbits.

So from the first view, the PSP ram is about 3x faster than the gp2x ram.

Conclusion. The PSP is at the same CPU clock, ~2x faster.


BTW: Did you know that the PSP also offers a SECOND Programmable CPU CORE ?

CPU1 in the PSP is the main r4000 cpu, with cache.
CPU2 in the PSP is the Media CPU, r400 core, WITH FULL CACHE.
Both cores are programmable, and can run different code.

So the gp2x offers 2x200Mhz, but the PSP offers 2x333Mhz...

So, from the point of speed, the PSP is the winner.
From the Point of "selling" interest, the gp2x is a winner.


cu,


So a SNES emulator will run twice as fast on a PSP than it will on a GP2X? Doesn't sound too promising for the GP2X..
 
Last edited by a moderator:
"So a SNES emulator will run twice as fast on a PSP than it will on a GP2X? Doesn't sound too promising for the GP2X.."

The NES emulation on the PSP, is at 230 MHZ, FULLSPEED, 50-60fps, fullscreen, fills the completet 16:9 display.

The SNES Emulator ( yoyofr release ) is at 333Mhz , at frameskip 1, fullspeed, 40-60fps. 16:9 display full filled.

btw, almost all psp emulators are plain c compiled versions, without any cpu optimisations.

To get the same speed on a gp2x, you need a lot assembler optimisations. And other tricks, like using the second cpu core.
 
So the GP2X is not going to see the rapid porting of many emulators running at playable speeds like the PSP did, unless the emulator being ported was highly optimized for the gp32. If we can expect Mhz per Mhz the GP2X to be slower then the PSP, SNES emulation is going to be a lot harder then many people have currently made it out to be.
 
So the GP2X is not going to see the rapid porting of many emulators running at playable speeds like the PSP did, unless the emulator being ported was highly optimized for the gp32. If we can expect Mhz per Mhz the GP2X to be slower then the PSP, SNES emulation is going to be a lot harder then many people have currently made it out to be.

With Linux and SDL on the flash, the gp2x ports of emulators will come out very fast.
YOu only need some tweaks, and in about 2-3h a port is crosscompiled.
But this fast crosscompiled emulator will be "slow" on the gp2x.

SNES emulation on the gp2x is a VERY hard thing. Some "sold" coders are working on it, to get an handwritten ARM Assembler CPU core. With this core, a fullspeed SNES on gp2x is possible.

But, have a look at the gp32, the machine is now 5 Years old. And at the first year, really nothing impressive was shown.

cu,
 
Last edited by a moderator:
Thats a good point; the GP32 had nothign going on for the first while, and then a flurry of crazyness occurred.

Really, all of this sums up to this; beefy hardware makes things easy; crappier hardwaer makes it less easy.If its less easy, you need good coders.

The gp2x will likely have a smaller dev pool than the PSP (consider the GP32, which to this day is essentially unheard of. The linux kernel being present will multiply the attention hundred-fold.. notice how many /. posts there have already been, for instance.) but that may not matter.. 1 devoted dev is all you need.. the trick is finding them.

So the gp2x will, as the gp32 did, live or die based on developers.

Answer: Many/most of the GP32 devs are movign to the gp2x, and so the gp2x will have _at least_ a small but devoted bunch of devs.. so all the bases will be covered _fast_ (linux+sdl makes it easy), and the core apps will get good over a shortish period of time (since the minum pool of devs is already present.)

Any other device has a much harder time, since no devs are implied.. the gp2x gets a free pool of homebrew devs, plus the Unix/Linux bonus crowd. (ie: Its not just eh linux community.. we cannot count out the FreeBSD/OSX groups, etc.)

jeff
 
Both cores can do the same things, in the same time.
So a 200Mhz arm9 is NOT faster, and not slower than a mips r4000 core.

They are ~ 1:1

Sure, we'll just take your word for it. You can't be wrong. :p

It more depends on the external bus speed, and how fast the machine can access the memory.

PSP: ddr ram, 32Bit access, connected 256Mbits.
Samsung 1st generation MCP 3.0V/2.5V 32MB 8 bit Uniform Block NAND Flash + 32MB 32 bit 6ns CL3 DDR.
The cpu/ram bus, runs with up to 333Mhz

gp2x: sd-ram ( not ddr-ram ), 133Mhz, 128Mbits.

So from the first view, the PSP ram is about 3x faster than the gp2x ram.

Conclusion. The PSP is at the same CPU clock, ~2x faster.

Are you saying that if I simply switch 133MHz memory to 266MHz memory, my PC will start to run twice as fast? <_< That's even worse than those CPU clock speed debates.
 
Last edited by a moderator:
I'm curious why GPH ended up choosing 64MB of SDRAM. From what I understand it was 32MB in the beginning. Why didn't they go with 32MB DDR SDRAM instead of 64MB SDRAM. Wouldn't that would've been a better idea performance wise?
 
I'm curious why GPH ended up choosing 64MB of SDRAM. From what I understand it was 32MB in the beginning. Why didn't they go with 32MB DDR SDRAM instead of 64MB SDRAM. Wouldn't that would've been a better idea performance wise?

Well... there nothing about "DDR ram" in the manual of the MMSP2 so reason is probably a very simple one. However the bus must be 16 or 32 bits wide so atleast there is 200MB/s of bandwidth in the GP2x (at best scenario there would be over 500MB/s). It doesn't matter that much as you will not be using more than 320x240 res anyway (the video can be decoded internally by VP/VPP). The cpus have a cache for a reason too.

So if there would be a hardware 3d accelerator in the GP2x I'd be worried. But there is none so bandwidth is not that important and having more memory is a better idea.
 
Last edited by a moderator:
"So a SNES emulator will run twice as fast on a PSP than it will on a GP2X? Doesn't sound too promising for the GP2X.."

The NES emulation on the PSP, is at 230 MHZ, FULLSPEED, 50-60fps, fullscreen, fills the completet 16:9 display.

The SNES Emulator ( yoyofr release ) is at 333Mhz , at frameskip 1, fullspeed, 40-60fps. 16:9 display full filled.

btw, almost all psp emulators are plain c compiled versions, without any cpu optimisations.

To get the same speed on a gp2x, you need a lot assembler optimisations. And other tricks, like using the second cpu core.


Those fullscreen modes look like crap though, all squished and blurry. That is the only thing I really don't like about emus on the PSP. Other than that it is the best right now. It will be interesting to see how close the GP2X could get.
 
Last edited by a moderator:
Back
Top