Nintendo Sixty Fooooooooooooooooooour!


nice vid. my heart skipped a beat or 2 when f zero x came on.
words cant express how much i love that game.

shame it doesnt play :p
(though im sure it will eventually).
 
Phawx said:
what is the bottleneck for the N64 Emu? The CPU or GPU? Because the iPhone 3GS has a better GPU than the Pandora.
I'm not sure about this. Nobody really is actually (unless the exact chip is finally known??). If the Apple processor is the new Samsung Cortex-A8 chip (S5PC100), then it is about 20% 'slower' (*on paper). The ODROID folks released that datasheet to the public and the specs on paper are lower than OMAP3.
 
Last edited by a moderator:
MDave said:
borgqueenx said:
Yes ofcource i use a 3GS.

Wow, dude you are one spoilt kid! You got a iPhone 3GS and you got your parents to pre order a Pandora for you? :blink:
I wonder about that for a while. My parents would have given me a kick in the ass and if they didn't my friends would ^^ Luckily I'm old enough to pay for my stupid hobbies myself :D
 
Last edited by a moderator:
MDave said:
borgqueenx said:
Yes ofcource i use a 3GS.

Wow, dude you are one spoilt kid! You got a iPhone 3GS and you got your parents to pre order a Pandora for you? :blink:
No i payed for them myself(both devices) but my parent like to control me and everything i do, as long as i live with them.
And i dont wanna move unless i find a nice real girlfriend.
Plus im 19 years of age. Not much a kid anymore.
 
Last edited by a moderator:
MWeston said:
I'm not sure about this. Nobody really is actually (unless the exact chip is finally known??). If the Apple processor is the new Samsung Cortex-A8 chip (S5PC100), then it is about 20% 'slower' (*on paper). The ODROID folks released that datasheet to the public and the specs on paper are lower than OMAP3.

The specs of what, in particular? The CPU in S5PC100 is better than OMAP3530's on paper. I don't know about GPU, but it's known for sure that iPhone 3GS has an SGX 535, which is also better than what OMAP3530 has.
 
Last edited by a moderator:
borgqueenx said:
Vorporeal said:
According to the interwebs, the S5PC100 is, in fact, the chip inside of the 3GS.
Good news! That means 20% speedup plus about 5-10% because the pandora doesnt have osx preloaded.

Whoa buddy, slow down there. You're getting ahead of yourself. Let's not speculate about this - the speed we see in the videos of mupen running on Pandora hardware is the speed that it is.
 
Last edited by a moderator:
Vorporeal said:
borgqueenx said:
Vorporeal said:
According to the interwebs, the S5PC100 is, in fact, the chip inside of the 3GS.
Good news! That means 20% speedup plus about 5-10% because the pandora doesnt have osx preloaded.

Whoa buddy, slow down there. You're getting ahead of yourself. Let's not speculate about this - the speed we see in the videos of mupen running on Pandora hardware is the speed that it is.
Well i already played with n64iphone and if i conpare that with pandora there is about 20% speedup.
But maby n64iphon. Doesnt have all the optimalisations yet...

Anyways im going to get some sleep now. Good night:)
 
Last edited by a moderator:
Exophase said:
MWeston said:
I'm not sure about this. Nobody really is actually (unless the exact chip is finally known??). If the Apple processor is the new Samsung Cortex-A8 chip (S5PC100), then it is about 20% 'slower' (*on paper). The ODROID folks released that datasheet to the public and the specs on paper are lower than OMAP3.

The specs of what, in particular? The CPU in S5PC100 is better than OMAP3530's on paper. I don't know about GPU, but it's known for sure that iPhone 3GS has an SGX 535, which is also better than what OMAP3530 has.
Specs on the GPU only. The Samsung CPU has the 32KB caches which we both know is better. I don't know what kind of memory the iPhone/Touch is using but if it is DDR2-333, that will add more latency on cache misses as well.

Check for yourself. I'm not making any claims beyond what I read: http://dev.odroid.com/wiki/odroid/pds/HardwareInformation/S5PC100_UM_REV101.pdf (WARNING: Big file not for casual download if you are just looking for a couple of pages!)
 
Last edited by a moderator:
MWeston said:
Specs on the GPU only. The Samsung CPU has the 32KB caches which we both know is better. I don't know what kind of memory the iPhone/Touch is using but if it is DDR2-333, that will add more latency on cache misses as well.

Check for yourself. I'm not making any claims beyond what I read: http://dev.odroid.co...0_UM_REV101.pdf (WARNING: Big file not for casual download if you are just looking for a couple of pages!)

This is more evidence that iPhone isn't using a stock S5PC100 then. SGX535 strings can be found in various locations, and at the very least Apple has made it very clear they're using an SGX.

EDIT: Thanks for the PDF.. it's now obvious that S5PC100 does not have an SGX. It's a non-unified shader architecture which can do 4-way float computations. Its fill rate is totally weird.
 
Last edited by a moderator:
what is the bottleneck for the N64 Emu? The CPU or GPU? Because the iPhone 3GS has a better GPU than the Pandora.
I'm pretty sure the CPU is still the bottleneck in almost all games. Not just the dynarec, a significant amount of time is also spent in passing data to the OGL driver especially due to the lack of batching.... Super Mario 64 is at about 300 drawcalls averaging ~18 vertices per frame. Also recognize that the iPhone is only rendering at 320x240 without any upscaling. I tried this on the Pandora and it was noticeably faster, but ofcourse its about 2" x 1" in size due to the pixel density. That's why it will be worthwhile using the 400x240 LCD mode or the scaling hardware.

If glN64 really is doing anything like this then it should really be turned off so it isn't wasting resources switching scaling/clipping inbetween different perspective matrices. It would also not be doing things straight to 800x480, or else Mario wouldn't be off aspect.
As far as I'm aware, There is no FOV hacking occuring. Different resolutions simply change the paramaters passed to glViewport(), clipping is done as usual. As you say, The differences in the mario 64 images are probably due to different roms versions. However the differences are not very noticeable, especially since most HUD elements are bilinearly filtered anyway. I would suggest widescreen for all games except 2D.
 
borgqueenx said:
Vorporeal said:
According to the interwebs, the S5PC100 is, in fact, the chip inside of the 3GS.
Good news! That means 20% speedup plus about 5-10% because the pandora doesnt have osx preloaded.
can we somehow make sure this guy doesnt make posts like this? thats how rumors on engadget get started, and he definitely sounds a little spoiled
 
Last edited by a moderator:
DaMummy said:
borgqueenx said:
Vorporeal said:
According to the interwebs, the S5PC100 is, in fact, the chip inside of the 3GS.
Good news! That means 20% speedup plus about 5-10% because the pandora doesnt have osx preloaded.
can we somehow make sure this guy doesnt make posts like this? thats how rumors on engadget get started, and he definitely sounds a little spoiled
I said: "ofcource i use a 3gs" because n64iphone only works on 3gs iphones and the 3rd gen ipods.
And go some posts back to see why i think 20% speedup.
 
Last edited by a moderator:
Wait wait. So the iPhone *doesn't* have the SGX 535? I knew I saw reports like about what @Exophase mentioned. And even Wiki shows the iPhone3GS as having the SGX 535. (I know, I know)

@Mweston

So the iPhone 3GS is just like the odroid then?

Apologies if I was mistaken.
 
MWeston said:
Specs on the GPU only. The Samsung CPU has the 32KB caches which we both know is better. I don't know what kind of memory the iPhone/Touch is using but if it is DDR2-333, that will add more latency on cache misses as well.
Well, if they are using DDR2-333 instead of DDR-333 then the cycle time would be 12ns instead of 6ns. It probably wouldn't make a huge difference since a 64-byte L2 line fill will still take 48ns (16 transfers x 3ns).

Surprisingly (to me) L2 cache misses aren't that bad in mupen64plus. I actually took out the PLD instructions that I was using since they weren't helping that much and the extra instructions were filling up the L1 cache. The small L1 in the OMAP3 is a problem, though I don't have any S5PC100 hardware to test what the actual speed difference is.

The larger issue is that the dynarec + rsp emulation is using 60-80% of the CPU time at 600Mhz, leaving just 20-40% for graphics. This is why going from 600MHz -> 850MHz yields such a large increase in framerate when frameskip is enabled.
 
Last edited by a moderator:
Phawx said:
Wait wait. So the iPhone *doesn't* have the SGX 535? I knew I saw reports like about what @Exophase mentioned. And even Wiki shows the iPhone3GS as having the SGX 535. (I know, I know)

@MWeston

So the iPhone 3GS is just like the odroid then?

Apologies if I was mistaken.

The iPhone 3GS definitely has an SGX (Apple says so), and almost definitely a 535 (strings in the OS say so). The Samsung S5PC100 definitely does not have an SGX (their documentation clearly contradicts the SGX featureset). Therefore, the iPhone 3GS definitely does not have an S5PC100, although the SoC it's using may be highly similar. People were by and large very quick to assume this, I suppose because they didn't think that Samsung could have been making a custom chip for Apple's several million sold phone.

The Odroid is definitely using the S5PC100, so at least in terms of 3D it's not the same as iPhone 3GS.

Right now one of the reasons why the iPhone 3GS version of Mupen64plus is slower is that zodttd hasn't found a way to map RDRAM to 0x80000000 in the process's address space. This is an optimization Ari64 used in order to avoid having to add something to N64 RDRAM addresses to convert them to emulated RDRAM addresses. zodttd is also currently generating a few too many instructions for this. With the location it's currently landing at it should take him 2 additional instructions to convert it, and chances are good that these instructions can be placed such that it only costs 1 extra cycle. The icache pressure will be a little worse, but the 2x icache should easily offset that.
 
Last edited by a moderator:
Ari64 said:
MWeston said:
Specs on the GPU only. The Samsung CPU has the 32KB caches which we both know is better. I don't know what kind of memory the iPhone/Touch is using but if it is DDR2-333, that will add more latency on cache misses as well.
Well, if they are using DDR2-333 instead of DDR-333 then the cycle time would be 12ns instead of 6ns. It probably wouldn't make a huge difference since a 64-byte L2 line fill will still take 48ns (16 transfers x 3ns).
Hum you are wrong I think: DDR2-333 has the same bandwidth of 2664 MB/s as DDR-333. The reason is that it uses both rising and falling edge for transfers.

As far as latency is concerned, the speed of the RAM isn't always the issue; the memory controller and the interconnect can have a huge impact (and explain why memory latency isn't very good on OMAP3).

The larger issue is that the dynarec + rsp emulation is using 60-80% of the CPU time at 600Mhz, leaving just 20-40% for graphics. This is why going from 600MHz -> 850MHz yields such a large increase in framerate when frameskip is enabled.
MAME is going to dynarec the RSP too. Did you consider it?
 
Last edited by a moderator:
Back
Top