I still have high hopes for the next iPad. If they can cram in something thats 2x as fast again as the 5s then we're looking at something that' s quite a bit faster than Adreno 330.
Any hints as to what rogue impelentation is in the 5s yet?
Next-Gen iPhone & iPhone Nano Speculation
- Thread starter Arun
- Start date
I still have high hopes for the next iPad. If they can cram in something thats 2x as fast again as the 5s then we're looking at something that' s quite a bit faster than Adreno 330.
Any hints as to what rogue impelentation is in the 5s yet?
anexanhume
Veteran
Ok, with pure 2x gpu claim, that puts us at 270 MHz for a Rogue 6430/6400. Seems reasonable. With integer multiple for CPU, we are at 1.35, 1.62 or 1.89 GHz. 1.62 GHz seems most logical, given there's also a 2x CPU claim, and 1.89 would be really high for Apple which tends to be on the slower side since the 4S.
Please.. both gpu and cpu claims were mentioned as "up to 2x" - that means in very specific areas considering how inflated the marketing numbers usually are. So maybe it's better to assume like 50% before predicting clocks etc...
Studies (AMD did one for instance) show that there isn't that much benefit beyond 16 registers for general purpose code (technically AArch32 only has usable 15 registers, so adjust that slightly). I know when I've written ASM I've often felt constrained by lack of registers, but the fact that I was compelled to write ASM in the first place probably heavily biases how I feel about it. There's also the matter of something being better vs something being easier - having more registers definitely makes writing the code easier if you're doing it by hand. Which of course almost no one has reason to do anymore.
I don't see any reason to believe that the GPU and CPU are derived from the same clock. That's a very poor design for DVFS. Separate PLLs for major components aren't that expensive. SoCs these days can have several of them.
I still see quad core as a possibility, particularly given the transistor count. Apple said the A5 in iPad 2 would offer 2x the performance of the A4 in iPad 1 - that was moving from 1GHz Cortex-A8 to 2x 1GHz Cortex-A9. Since Cortex-A9 is generally ~25% faster per clock Apple was being pretty conservative with this figure and not just scaling by core count. Since you tend to get less benefit going from 2 to 4 cores than you do from 1 to 2 Apple could be using an even more conservative figure in their estimation. Triple core is also somewhat of a possibility.
Last edited by a moderator:
anexanhume
Veteran
IIRC the "9x" increase claimed for A5 was a FLOP for FLOP increase over A4 SoC. So I don't think it's unfair.
edit: it was. SGX 535 @ 200 MHz = 1.6 GF. SGX 543 MP2 @ 200 MHz = 14.4 GF.
ltcommander.data
Regular
Infinity Blade III's developers did claim they see 5x better performance on the 5S than the 5. As you said, some areas no doubt see more benefit than others. Hopefully they'll be some leaked uploads to GeekBench, GFXBench, and other benchmarks in the near future to get a better sense of performance before launch.
Is it really necessary though? iPhone screen rez is quite limited, unlike the retina iPad, which has an absolutely monstrous screen with far more pixels than even most desktop displays.
2x graphics performance will undoubtedly be just fine. I don't see a big push for technically accomplished phone games anytime soon - do you? Let's not go overboard here, all frothing at the mouth etc over percieved graphics "weakness" which doesn't actually exist in reality...
anexanhume
Veteran
Fair point, but historically it has been. 320 in A6 x4 = 1280. 280 in A6X x5 = 1400. Previous frequencies of 250 and 200 for 1 GHz and 800 MHz CPU speeds respectively. Could be coinky-dink.
Good point about quad/tri core too. I think you're probably right.
Have we ever actually gotten official confirmation on the GPU clock speeds? If not people could flat out be picking ones that nearest fit ratios.
These are just peak clocks anyway, it'd be sloppy design if the CPU and GPU could only run in coarse divisions of a PLL clock you adjusted, unless the master clock was a lot higher than the CPU and GPU peak (doubtful)
anexanhume
Veteran
No, they've been extrapolated from benches I believe. Probably some confirmation bias going on.
Incidentally, it's kind of strange that you argue that they have asymmetric clocking of the CPU cores but symmetric clocking between the CPU and GPU, how would this work exactly?
anexanhume
Veteran
Incidentally, it's kind of strange that you argue that they have asymmetric clocking of the CPU cores but symmetric clocking between the CPU and GPU, how would this work exactly?
It wouldn't. That's why they're not in the same post.
Oh and just by the way (heh heh ), has anyone considered the fingerprint scanner...?
If you're conspiratorially inclined, you'd expect the NSA to have your fingerprint pretty much the very first time you touch that home button. While apple's sworn your fingerprint won't be stored on THEIR systems, they did not say anything about the US gov't did they...
), has anyone considered the fingerprint scanner...?If you're conspiratorially inclined, you'd expect the NSA to have your fingerprint pretty much the very first time you touch that home button. While apple's sworn your fingerprint won't be stored on THEIR systems, they did not say anything about the US gov't did they...
anexanhume
Veteran
edit: what happened to the post I quoted?
It's the 5 plus WTR1605L and different casing. Better front camera.
It's the 5 plus WTR1605L and different casing. Better front camera.
wow, you are speaking about consumer windows.. not professional one ( server ).
In reality, i dont think the "64bits" will do anything ( or much ) in performance for iOS or Iphone/Ipad this generation, but its a good move, maybe right for the next generation. ( It will push developpers in the right direction ). ( I could even believe it is more a marketing thing today ( following annonces of AMD and other about APU/embedded SOC/ server 64bits ARM based ( but anyway i will wait we get more concrete information, keynotes slides, graph and other are.. well.. i will not much comment on them right now. )
Last edited by a moderator:
Thanks, and sorry about the mess, I deleted the post because I found the information I was looking for.
2X GPU performance of iphone5 is only a bit faster than Galaxy S4. It cannot have 100 Gflops. I don't see how 5S can have 270MHz of G6430?
jedibeeftrix
Newcomer
Boom, 64-bit!
2x general-purpose registers
2x floating-point registers
Over 1 billion transistors
102mm2 die size
OpenGL ES 3.0 support
a great foundation for the future of apple mobile computing, i wonder how long it will take other to catch up.
anexanhume
Veteran
270 MHz G6430 gives you 69.2 GF, which is twice 543MP3 @ 320 MHz. Of course, alternative is G6200 @ 540 MHz, but Apple tends to hang on the lower side of clocks, hence my G6430 guess.
edit: I'm using this post to calculate GF:
http://forum.beyond3d.com/showpost.php?p=1770477&postcount=349
Last edited by a moderator: