iPhone 3G S - How much faster?

Rob Evans

Newcomer
I've just noticed a comparison speed test between an iPhone 3G S and a second gen iPod Touch at http://blackpixel.com/blog/244/early-iphone-3g-s-opengl-test-results/

The first test, which was both GPU and CPU intensive, showed a 2x improvement in speed.

But the interesting test was a CPU only one:

iPhone 3Gs: 6.42 Mflops (small dataset) 5.86 Mflops (large dataset)
2g Touch: 5.21 Mflops (small dataset) 4.86 Mflops (large dataset)

Only a 20% improvement.

The tests are fairly simple, but made me wonder if the CPU really is a 600MHz Cortex A8, as has been claimed.

Rob.
 
Well isn't the Touch about 20% faster than the original iPhone as well?
 
The only clear(ish) benchmark comparison I can find is at http://www.anandtech.com/gadgets/showdoc.aspx?i=3587

Generally the 3G S is around 40% faster than the 3G.

However, according to most reports the 3G S has a 600Mhz Cortex A8 compared to the 3G's 412MHz ARM11. The clock speed difference is 45%.

According to http://en.wikipedia.org/wiki/ARM_architecture the Cortex should also be significantly faster than the ARM11 - around 75% faster at the same clock speed.

I don't think the rumour of the iPhone 3G S having a 600Mhz Cortex A8 can be correct - I would expect a much bigger increase in overall performance over the 3G if it was.

Rob.
 
The guys who took the iPhone 3G S apart already determined that is has a Cortex-A8 CPU. But I think the extra RAM and possibly faster storage plays a role in the overall performance increase as well.

And yes, the 2nd gen iPod touch CPU already is ~30% faster than the iPhone 3G's. And it's also quite possible that the OS is not fully optimized for the new CPU architecture just yet. I mean it's not as Apple had to optimize anything to make most people happy in regard to speed this year... they can do the optimizations for next years iPhone (which in my opinion will keep the same SoC, maybe @45nm and higher clocked; a 2 year exterior/interior cycle seems very plausible to me).
 
But the interesting test was a CPU only one:

iPhone 3Gs: 6.42 Mflops (small dataset) 5.86 Mflops (large dataset)
2g Touch: 5.21 Mflops (small dataset) 4.86 Mflops (large dataset)

Only a 20% improvement.

The tests are fairly simple, but made me wonder if the CPU really is a 600MHz Cortex A8, as has been claimed.
i assume that test is fpu-centric, in which case i'm surprised the 3GS comes out faster at all : )
 
The guys who took the iPhone 3G S apart already determined that is has a Cortex-A8 CPU.
Give me a break. The idiots being quoted all over the place couldn't tell apart a 68K microcontroller from a quad-core Itanium. The 3GS chip has nothing to do with the S5PC100 whatsoever, and any real teardown analyst who actually looked underneath the package will tell you as much: http://i.cmpnet.com/eetimes/news/09/06/image6_061909.jpg

FWIW, it nearly certainly does have a Cortex-A8 though. But that's an entirely separate issue, and the teardowns won't help you there. Being able to calculate percentages correctly does help though ;) (come on guys, it's not because Anand originally couldn't do it right that you need to follow his lead in even stranger directions - no need to reinvent algebra! :p)
 
Give me a break. The idiots being quoted all over the place couldn't tell apart a 68K microcontroller from a quad-core Itanium. The 3GS chip has nothing to do with the S5PC100 whatsoever, and any real teardown analyst who actually looked underneath the package will tell you as much: http://i.cmpnet.com/eetimes/news/09/06/image6_061909.jpg
I don't know if they are idiots, but this is the teardown article I read:
http://www.phonewreck.com/2009/06/19/iphone-3gs-teardown-and-analysis/

FWIW, it nearly certainly does have a Cortex-A8 though. But that's an entirely separate issue, and the teardowns won't help you there. Being able to calculate percentages correctly does help though ;) (come on guys, it's not because Anand originally couldn't do it right that you need to follow his lead in even stranger directions - no need to reinvent algebra! :p)
412 MHz x 1.3 = 535.6 MHz. I think that's close enough to 533 MHz to let my "~30%" slide ;)
 
I don't know if they are idiots, but this is the teardown article I read:
http://www.phonewreck.com/2009/06/19/iphone-3gs-teardown-and-analysis/
And that says "Apple has chosen to go from the ARM 11 Samsung S3C6400 to the ARM A8 Samsung S5PC100 with this device." - now *that's* evidence! I sure as hell believe it now. </sarcasm> ;) They didn't go deeper than the photo you see, and there's nothing useful printed on the package, so how exactly do they know? (hint: those people are just repeating what they read elsewhere, not realizing that original source is really just speculation).

412 MHz x 1.3 = 535.6 MHz. I think that's close enough to 533 MHz to let my "~30%" slide ;)
Oh sure, I was mostly thinking of Rob's claim that "Generally the 3G S is around 40% faster than the 3G." - no it's not. It takes ~0.45x (i.e. 45%) the time of the 3G to render webpages on average, which means it's ~2.22x as fast or ~122% faster. There's no way an ARM11 could be that fast with no software changes.

Of course, it's not strictly impossible that the main performance benefit actually comes from software changes - i.e. more of the rendering being offloaded to the SGX. In that case it still could be a 600MHz ARM11 - but that's not the most likely explanation IMO.
 
Give me a break. The idiots being quoted all over the place couldn't tell apart a 68K microcontroller from a quad-core Itanium. The 3GS chip has nothing to do with the S5PC100 whatsoever, and any real teardown analyst who actually looked underneath the package will tell you as much: http://i.cmpnet.com/eetimes/news/09/06/image6_061909.jpg

FWIW, it nearly certainly does have a Cortex-A8 though. But that's an entirely separate issue, and the teardowns won't help you there. Being able to calculate percentages correctly does help though ;) (come on guys, it's not because Anand originally couldn't do it right that you need to follow his lead in even stranger directions - no need to reinvent algebra! :p)

The photo you refer to appears in this video teardown, done by Portelligent/Semiconductor insights and Teardown TV (all part of the EEtimes group). One assumes they can tell a 68K from an Itanium :)
The voiceover at the time of the picture being shown states that its an S5PC100, I assume the video is from the people who de-capped the chip and hence took the photo.
http://www.eetimes.eu/218100442;jsessionid=GQII2OD1GHM2OQSNDLRSKHSCJUNN2JVN

Are you saying its a Cortex A8, produced by Samsung, but has nothing at all to do with Samsungs only known Cortex A8 processor ?


Hopefully when it gets jailbroken, someone will download GLbenchmark onto it, and then we'll get a more reliable speed check against the previous iphone, which is already in the database.
 
Last edited by a moderator:
The photo you refer to appears in this video teardown, done by Portelligent/Semiconductor insights and Teardown TV (all part of the EEtimes group). One assumes they can tell a 68K from an Itanium :)
The voiceover at the time of the picture being shown states that its an S5PC100, I assume the video is from the people who de-capped the chip and hence took the photo.
http://www.eetimes.eu/218100442;jsessionid=GQII2OD1GHM2OQSNDLRSKHSCJUNN2JVN
Funny - I meant that remark at the other teardown guys, not Semiconductor Insights which I usually respect, and I was using that pic to prove my point it wasn't a S5PC100.

But the voice-over does claim that's what it is - with a catch. It says "before we even decapped it" - i.e. I suspect that audio was recorded before they checked the die markings (which were just edited in the video later). I doubt they'd have been so clearcut otherwise.

Anyway, just to make this clear for the last time: it's... not... a... S5PC100. Heck, that chip doesn't even have a SGX! As I said in another thread, I talked with someone from Samsung at Mobile World Congress 2009 where they were showing that chip and I was told it was based on an in-house 3D solution.

Are you saying its a Cortex A8, produced by Samsung, but has nothing at all to do with Samsungs only known Cortex A8 processor ?
Produced by Samsung, designed by Apple, yes. I have no idea why people think Apple's semiconductor team has no other goal than making a custom ARM core - they're also responsible for custom SoCs and have been for much longer.

I hate repeating myself so much, but people really should read this paper (especially the last page) and put 2 and 2 together by themselves: http://www.samsung.com/global/busin...foundry/downloads/foundry_brochure_200611.pdf
 
Anyway, just to make this clear for the last time: it's... not... a... S5PC100. Heck, that chip doesn't even have a SGX! As I said in another thread, I talked with someone from Samsung at Mobile World Congress 2009 where they were showing that chip and I was told it was based on an in-house 3D solution.

Produced by Samsung, designed by Apple, yes. I have no idea why people think Apple's semiconductor team has no other goal than making a custom ARM core - they're also responsible for custom SoCs and have been for much longer.

Samsung's *published info* on their ARM11 parts doesn't show anything that has MBXlite in it, but we know they supplied the chip for that, and as Apple didn't have any access to MBX licensing at the time, Samsung would have been the major architect (sp?) and it wouldn't be a suprise to me if that processor born a close ressemblence to Samsungs only published ARM11 parts, the non-mbx S3C6400/6410.

I certainly never suggested it was an S5PC100, I have guessed/suggested that its a customised version of it with SGX530/535 graphics and possibly IMG derived video IP. Purely based on the timeline of the S5PC100 being announced, and the timeline of various SGX milestones and licence announcements.

Articles describing it as an S5PC100 are just plain silly/lazy

I guess it'll all come out eventually one way or another.
 
I don't think the rumour of the iPhone 3G S having a 600Mhz Cortex A8 can be correct - I would expect a much bigger increase in overall performance over the 3G if it was.

Rob.


It is definately Cortex ... otherwise I wouldn't be able to run armv7 code which it currently does :)
 
Samsung's *published info* on their ARM11 parts doesn't show anything that has MBXlite in it, but we know they supplied the chip for that, and as Apple didn't have any access to MBX licensing at the time [...]
But Samsung did have access to it for their foundry design business! All this proves is it can't be Apple which synthesized that MBXLite, it must be Samsung. And since I'm not claiming Apple even bothered with any synthesis work at all for that first SoC, and possibly not even the new one, I don't see how that contradicts anything.

Why wouldn't Apple use a Samsung SGX license again then (or ask them to license it if they didn't do so already)? Probably to negotiate better pricing and to reduce long-term dependence on Samsung.
 
I totally failed to notice (or it was added later), that there is a page "2" to that EEtimes article, where they state they will be investigating the App Pro in detail and measuring various aspects of the die to see if they can make any positive IDs.
 
When compiling code for 3gs , the compiler does generate Neon code so to do a proper comparison one would have to have two sets of binaries.

As far as the GPU is concerned ... my own (2d) test gave me about 5 times speedup ( which in this case would mean 5 x fill rate)
 
I don't know ... my test was based on some older code designed for OpenGL ES 1.x and since 1.x is is emulated within the driver ( with shaders) on the latest iPhone .. perhaps that had something to do with it, hard to say.
 
Back
Top