600 Mhz Powerful Enough To
- Thread starter 1magus
- Start date
StealthBagel
Member
- Joined
- Oct 17, 2007
- Messages
- 251
Sega Saturn emulators for x86 are slow because they are unoptimized... not because of any inability to handle it. The complexity of the system means it's not a project to be taken lightly and therefore the only emulators have been rather underdeveloped. Yabause is showing promise but its approach isn't suited to the fastest possible operation since it relies on an interpreter. It should, theoretically, be doable for such an emulator to run well on Pandora but the question is, will anyone put in the time to do it. When there's not even a well rounded Saturn emulator for the PC, the odds of someone taking on trying to put together something for this new handheld strikes me as pretty unlikely, at least until there is a working example with acceptable speed on the PC side of things. As it stands only PCs with very high end CPUs can achieve decent performance in any of the existing emulators. Same reason I think it will be a long time before we see DS emulation even attempted on this handheld. It's still in its infancy on the PC side of things and even if you made a fork to a project like No$gba or Yabause you'd have to be maintaining it alongside a constantly-in-development x86 project with massive changes. It'd be a mess..
I feel furthermore compelled to mention MHz is not a measure of performance, it's operating frequency. Different CPU architectures, especially as wildly different as ARM and X86, will perform drastically differently at 600 MHz. If you need something to compare use MIPS and MFLOPS. Even so these are theoretical maximums and real world issues like memory bandwidth, cache and bus contention can all vary how a stressful app like an emulator that taxes the CPU, RAM and GPU performs on actual silicon....
I feel furthermore compelled to mention MHz is not a measure of performance, it's operating frequency. Different CPU architectures, especially as wildly different as ARM and X86, will perform drastically differently at 600 MHz. If you need something to compare use MIPS and MFLOPS. Even so these are theoretical maximums and real world issues like memory bandwidth, cache and bus contention can all vary how a stressful app like an emulator that taxes the CPU, RAM and GPU performs on actual silicon....
I'd like to see the source of that info. Even high end PC GPUs don't operate at such high frequencies. Neither does their RAM in most cases, even operating at a 2x "real" frequency 'effective clockrate' due to being DDR rather than regular SDRAM...... since the graphics hardware in the OMAP3 is about on par with a 3D accelerator somewhere between the GeForce 2 and 3, there's no rational reason for such a high operating frequency.argor said:
Last edited by a moderator:
Yuglooc
Go Go Igo!
Stealth Bagel said:
Last edited by a moderator:
yaustar
UK GP32 & GP2X Owner
Hell. No. The Saturn hardware was heavily customised and as such, a pain to even develop on much less emulate.TylerAW said:
Last edited by a moderator:
StealthBagel
Member
- Joined
- Oct 17, 2007
- Messages
- 251
"Heavily customized"? No, it was complex. The only custom parts are the VDPs (both of them), the SCSP (which is pretty much still just a well documented Yamaha synth), and the SCU, which as I understand it is primarily a fixed point coprocessor that acts like an FPU, and acts as a bus arbiter. There's a few other small things. But for the most part, it's still full of off the shelf parts from Hitachi and Motorola (SH-2s, an SH-1 and a 68K). Every console has custom parts, so the Saturn is no exception, it's just very WEIRD hardware....
Don't confuse hardware complexity with requiring faster hardware to emulate it though. It just means the programmer has a lot more to put up with in coding it all.
Don't confuse hardware complexity with requiring faster hardware to emulate it though. It just means the programmer has a lot more to put up with in coding it all.
Yuglooc
Go Go Igo!
I found the actual source. http://forum.gp2x.de/viewtopic.php?t=4998
In the middle of the page ED says:
QUOTE
Ja, 800MHz ist Standardtakt, soweit ich weiß. Und die GPU hat 1,8GHz...
Which google translate to..
QUOTE
Yes, 800MHz is standard clock, as far as I know. And the GPU has 1.8 GHz ...
In the middle of the page ED says:
QUOTE
Ja, 800MHz ist Standardtakt, soweit ich weiß. Und die GPU hat 1,8GHz...
Which google translate to..
QUOTE
Yes, 800MHz is standard clock, as far as I know. And the GPU has 1.8 GHz ...
MWeston
Internal Development
Yuglooc said:I found the actual source. http://forum.gp2x.de/viewtopic.php?t=4998
In the middle of the page ED says:
QUOTE
Ja, 800MHz ist Standardtakt, soweit ich weiß. Und die GPU hat 1,8GHz...
Which google translate to..
QUOTE
Yes, 800MHz is standard clock, as far as I know. And the GPU has 1.8 GHz ...
I saw that too. Either something got lost in translation or ED was kidding.
Maybe he was comparing it to a GP2X and saying the 3D output was like a GP2X at 1.8GHz? That would make more sense. Incidentally, Pandora is actually like a GP2X running at 1200MHz if you use the architecture comparisons. If you compare 3D, it's probably more like 3GHz!
Last edited by a moderator:
Tobriand
Well-Known Member
I seem to recall one of the stupidly amazing coders (one of Squidge, Exophase, hlide or zodttd) saying that, in principle, the machine at 1ghz ought to have the power to emulate it. But that was if (and only if) someone were to write an emu from the ground up for Pandora, and even then, it'd have to be a very very efficient emulator.
The post to clarify whether they thought it was doable said, essentially, yes, but you'll see monkeys successfully reproducing Hamlet by random letter striking first. Possible, but sufficiently unlikely to render it practically impossible. In theory. And practice is never as easy as theory.
The post to clarify whether they thought it was doable said, essentially, yes, but you'll see monkeys successfully reproducing Hamlet by random letter striking first. Possible, but sufficiently unlikely to render it practically impossible. In theory. And practice is never as easy as theory.
MWeston
Internal Development
You just did. Be careful!Squidge said:
Last edited by a moderator:
whiskthecat
whiskthecat
BUM BUM BUM!
Sorry I couldn't resist.
Sorry I couldn't resist.
Exophase
Nothing good will ever come of Exophase.
1200MHz? So you're admitting that the base clock will be 600MHz, and then doubling it because it's dual issue, right? Pretty bold of you to do so, I don't think any program will pair that well at all. Throw in the hand optimized/dynamically generated ASM that isn't well paired at all and a compiler that may not be that good at it and I think you might get 1.2 to 1.5x the performance per clock as ARM9 (assuming the code doesn't cause a lot of branch mispredicts). Oh, and the memory performance will probably be better, depending on how the L1/L2 cache is and how much the bus keeps up with the CPU.MWeston said:Maybe he was comparing it to a GP2X and saying the 3D output was like a GP2X at 1.8GHz? That would make more sense. Incidentally, Pandora is actually like a GP2X running at 1200MHz if you use the architecture comparisons. If you compare 3D, it's probably more like 3GHz!
600MHz ARM Cortex-A8 is definitely not 600MHz x86. Most 600MHz x86's (Athlon, Pentium 3 Katmai, CuMine Celeron) have benefits like out of order/speculative execution that better aid in keeping the pipes full. Cortex-A8 has an arguably superior and denser (with Thumb-2) ISA.
It's not that no Saturn emulators for PC are optimized either. SSF has recompilers, even for the SCU, last I was aware. Since we don't have the source we don't know the extent of how sophisticated these are. SSF requires SSE2 for some reason so that kind of limits how slow a machine you can run it on. The closest I can site is the commercial GiriGiri - I ran a hacked version of it once on an 850MHz Celeron, running Panzer Dragoon Saga, which is probably one of the more demanding games on Saturn. The performance was not fullspeed, but it was at a level where some would consider it worthwhile. I think that the OMAP3430 architecture with sufficient overclocking can trump the 850MHz (overclocked) Celeron I was using.
I don't know if excellent Saturn emulation will be possible on this. What I do think is that it would be a great endeavor that shouldn't be written off. Even if it didn't run phenomenally it'd be a valuable learning experience for pretty much anyone involved, and would finally offer the foundation for highest performance possible Saturn emulation that could be used in future hardware if not this machine.
I know someone who has been very interested in Saturn emulation for a long time. I'm hoping he'll pick it up.
Last edited by a moderator:
Yrx
the Happy Conqueror
It would basically require an extremely committed developer. And by "committed," I mean it in the asylum sense, lol. It would be awesome, though... Saturn was just such a Frankenstein.
MWeston
Internal Development
Cool analysis, but you read more into it than I would ever let on. I'm pretty picky about not releasing information that should not be out there yet. I was just having fun with numbers to show the leap in technology we have in our grasp now!
TI's development platform for 3430 runs at 550MHz (this is public), so here is what I did:
ARM9 x 1.1 = ARM11
ARM11 x 2 = Cortex-A8 = ~1200MHz
We all know it has been proven to go much faster but we won't be trying it until the kernel is completely stable so the overclocking can be the only thing blamed for system crashes at higher speeds.
I personally think some optimized code will make a very capable Saturn emulator. It truly is a matter of who would be willing to put all the time in to making it.
TI's development platform for 3430 runs at 550MHz (this is public), so here is what I did:
ARM9 x 1.1 = ARM11
ARM11 x 2 = Cortex-A8 = ~1200MHz
We all know it has been proven to go much faster but we won't be trying it until the kernel is completely stable so the overclocking can be the only thing blamed for system crashes at higher speeds.
I personally think some optimized code will make a very capable Saturn emulator. It truly is a matter of who would be willing to put all the time in to making it.
