Snes For The Wiz

[alterdimension]

Do yout think that with the wiz of snes will support star fox to make it full speed, and will it make the ps1 emulator with full speed and mortal kombat II will be full speed and make some rpg's full speed?

 

[anexanhume]

I have no emulator writing experience, so this is a guess:

Star fox- yeah. I think the Wiz should be able to handle all things SNES.

PSX- Better compatability, yes, but I'm sure there will still be games it is not full speed on, at least in the beginning.

 

[Squirrel61]

I can't tell you, but I can do maths so I can make a fair guess.

The F100/F200 are running at 200 MHz, overclockable to 266 MHz. The Wiz will run at 533 MHz, which is exactly the double (in computer math) of the overclocked F100/F200. So you can expect all software to run roughly twice as fast, maybe more when the support chips (graphics, sound) of the Wiz are faster than that of the F100/F200.

So my guess is that any game on any emu that runs over 50% speed on the F100/F200 will run 100% on the Wiz.

 

[WhizzBang]

The F100/F200 are running at 200 MHz, overclockable to 266 MHz. The Wiz will run at 533 MHz, which is exactly the double (in computer math) of the overclocked F100/F200. So you can expect all software to run roughly twice as fast,...

I don't think it works quite like that.

Anyway, I am fairly confident the Wiz will run SNES games full speed. I am not so sure about PS1 but I expect many games will be very playable.

 

[sam fisher]

I can't tell you, but I can do maths so I can make a fair guess.

The F100/F200 are running at 200 MHz, overclockable to 266 MHz. The Wiz will run at 533 MHz, which is exactly the double (in computer math) of the overclocked F100/F200. So you can expect all software to run roughly twice as fast, maybe more when the support chips (graphics, sound) of the Wiz are faster than that of the F100/F200.

So my guess is that any game on any emu that runs over 50% speed on the F100/F200 will run 100% on the Wiz.

What are your credientials that allow you to make this utterly and totally incorrect blind stab in the dark?

Suppose this hypothetical situation. I drive to work in a 60hp car and it takes me exactly 30 minutes each day.
I then replace the engine with a 120hp one and drive to work. It takes me 24 minutes. How does that work, should I surely not get there twice as fast and take at most 15 minutes ?

Or do you suppose that there are other factors that must be considered as well as the speed of my car (or the clock speed of the CPU)?

 

[Squirrel61]

I wrote "roughly" for a reason...

I know it doesn't exactly work that way. But I expect the Wiz to be more than twice as fast in raw processing power. First, the doubled processor clock enables the processor to process twice as much commands per second, unless the commands are more complex and thus require more clock cycles to complete. Second, the design of the Wiz is some years newer, the design of its chips is newer so I expect everything to be more optimized, hardware wise.

On the other side, if the firmware of the Wiz is more complex, you won't get a speed double for most actions, including games that rely on firmware libraries.

 

[Exophase]

I wrote "roughly" for a reason...

I know it doesn't exactly work that way. But I expect the Wiz to be more than twice as fast in raw processing power. First, the doubled processor clock enables the processor to process twice as much commands per second, unless the commands are more complex and thus require more clock cycles to complete. Second, the design of the Wiz is some years newer, the design of its chips is newer so I expect everything to be more optimized, hardware wise.


GP2X is ARM9 and ARMv4, Wiz is ARM9 and ARMv5TEJ. The CPU implementation, being the same generation, is close to the same, so it's not more optimized in that regard. The instruction set adds a few things, but in my opinion only rather special purpose things that don't apply to most programs and aren't going to be what a compiler will generate.

But you're otherwise right, under nominal conditions instructions which do not access memory will take the same number of cycles, and hence run in half as much time as the 266MHz GP2X. What you aren't accounting for are loads/stores that cause cache misses. The Wiz has the same memory clock speed as the 266MHz GP2X but DDR SDRAM, which should mean 2x the bandwidth, but because they've halved the bus width it is at best the same bandwidth, but worse latency. Meaning that the memory performance on a 533MHz Wiz is probably inferior to a 266MHz GP2X.

If a program spends a lot of time in cache misses (which most do, because 16/16 L1 cache doesn't really cut it) then the speed advantage of Wiz is trimmed down tremendously. Unfortunately there aren't automatic performance counters that you can look at w/o extra hardware, but based on what I've seen on similar platforms I wouldn't be surprised if 40% or more of cycles are idle. So let's say there is a GP2X program that spends 40% of its cycles in cache misses, then:

total cycles = 40x + 60x

(where the second part is are the cycles it spends not in cache stalls)

Now on the Wiz it will use at least twice as many cycles, but probably more, let's say, rather arbitrarily, 2.5:

So:

total cycles = 40(2.5x) + 60x = 160x

Vs 100x for the first, of course Wiz is clocked twice as high, so if the GP2X took 100 time units then Wiz takes 80 time units. In this scenario it's not 100% faster, it's only 25% faster.

Actual programs will vary a lot, but it's my opinion that SNES9x tends to be highly cache limited because of its huge frame/scanline buffers for main/subscreen that are pretty much guaranteed cache misses each.

On the other side, if the firmware of the Wiz is more complex, you won't get a speed double for most actions, including games that rely on firmware libraries.

No offense, but this part is just nonsense :\

EDIT: My wall fell on my keyboard and caused this to be submitted early. Don't ask -_-

 

[Squirrel61]

Thanks for the extensive explanation the wall falling down.

On the other side, if the firmware of the Wiz is more complex, you won't get a speed double for most actions, including games that rely on firmware libraries.

No offense, but this part is just nonsense :\
I was thinking along the lines of Windows XP vs. 98 or Vista vs. XP, but apparently I'm mistaken (probably because it doesn't work that way in Linux?)

 

[sam fisher]

EDIT: My wall fell on my keyboard and caused this to be submitted early. Don't ask -_-

I'm sorry but it has to be done....What!?

 

[Exophase]

EDIT: My wall fell on my keyboard and caused this to be submitted early. Don't ask -_-

I'm sorry but it has to be done....What!?


The divider between my two windows spontaneously collapsed and knocked over my monitor and my keyboard >_>

 

[sbock]

@Exophase

What about TCM? Any benefit from this ARM feature?

Regards,
Stephan

 

[Exophase]

@Exophase

What about TCM? Any benefit from this ARM feature?

Regards,
Stephan

I have looked before and found nothing on the Wiz's CPU having TCM. If you have any information on it having some please let me know.

 

[DaveC]

So:

total cycles = 40(2.5x) + 60x = 160x

Vs 100x for the first, of course Wiz is clocked twice as high, so if the GP2X took 100 time units then Wiz takes 80 time units. In this scenario it's not 100% faster, it's only 25% faster.

Actual programs will vary a lot, but it's my opinion that SNES9x tends to be highly cache limited because of its huge frame/scanline buffers for main/subscreen that are pretty much guaranteed cache misses each.

Aw man, a dose of reality. I was hoping that at least SNES being full speed would make up for the smaller screen. I guess faster is still faster though. Maybe at least the software horizontal smooth stretch in SNES will be possible as it has a bit more oomph than the GP2X.

I wonder why they limited the memory bandwith like that?

How do you think the Pandora will be able to run SNES? It is only 100 MHz more so will there be much of a difference?

Thanks for the informative explanation though.

 

[Adventus]

QUOTE
The Wiz has the same memory clock speed as the 266MHz GP2X but DDR SDRAM, which should mean 2x the bandwidth, but because they've halved the bus width it is at best the same bandwidth, but worse latency.
Oh man, here i was thinking the DDR ram was the biggest improvement over the GP2X.

Isn't many of the GP2X applications heavily RAM bottlenecked? Craig's timings certaintly seem to help alot.

Oh well, luckly im already set for Pandora.

 

[trooper]

How do you think the Pandora will be able to run SNES? It is only 100 MHz more so will there be much of a difference?

Thanks for the informative explanation though.

Well, You see, It goes like this...

Nah, I think i let Exophase answer this.

Sorry Exophase. :rolleyes:

I concur on the informativeness of you post. Very enlightening.

Trooper

 

[Exophase]

How do you think the Pandora will be able to run SNES? It is only 100 MHz more so will there be much of a difference?

Pandora has: 32bit DDR instead of 16bit (same clocks though, but I don't know if they can be scaled while overclocking), and perhaps more importantly, 256KB of L2 cache. This cache makes a HUGE difference in the emulators tested with and without it.

SNES emulation has been tested on Pandora and the results have been more than good enough, I doubt you'll be dissatisfied with them.

Wiz is probably good enough too, but we'll have to wait and see. Although bear in mind, SNES emulation doesn't have to be done with the cache killing problems that SNES9x has.

 

[WarmFluffyUK]

Don't forget that Starfox is not your usual SNES game as the starfox cartridge includes a coprocessor called the FX chip used to accelerate the 3D graphics. This is why handheld SNES emulators in the past have struggled or will totally not play Starfox.

If Starfox does become viable at full speed, I will be one happy bunny indeed!

 

[sbock]

@Exophase

The Pollux comes with a ARM926EJ CPU, which features TCM:

http://www.arm.com/products/CPUs/ARM926EJ-S.html

or

http://infocenter.arm.com/help/topic/com.a...js_r0p5_trm.pdf

But I think, you say that the Pollux SoC don't use this feature, right?

Regards,
Stephan

 

[Exophase]

The ARM manuals refers to an available TCM interface bus. You'll notice that it's not internal memory (like cache), but is for low latency external SRAM that can have zero wait-states. So the ARM designs do not include any actual TCM. Pollux is also not using one of those macrocells, or it wouldn't have 16/16KB of L1 cache. If it had TCM it'd definitely mention it here:

http://www.arm.com/products/CPUs/ARM926EJ-S.html

 

[sbock]

The ARM manuals refers to an available TCM interface bus. You'll notice that it's not internal memory (like cache), but is for low latency external SRAM that can have zero wait-states. So the ARM designs do not include any actual TCM. Pollux is also not using one of those macrocells, or it wouldn't have 16/16KB of L1 cache. If it had TCM it'd definitely mention it here:

http://www.arm.com/products/CPUs/ARM926EJ-S.html



Ah, yes. Thanks for clearing this up.

Regards,
Stephan