there you go, that was the one, I was just a process ahead of myself !
Next-Gen iPhone & iPhone Nano Speculation
- Thread starter Arun
- Start date
there you go, that was the one, I was just a process ahead of myself !
iMacmatician
Regular
From cnbeta.com: "iPhone 6 will reach 2GHz processor exposure above details" (original).
Also, according to the report, the iPhone 6's A8 will still be dual-core.
I'm not really surprised, since despite an early report stating quad-core I didn't think Apple would need to make that jump given the performance of the Cyclone cores (to be fair, I don't recall the report saying what device the chip would be used in). I think the A8 will have a larger performance improvement in the GPU than the CPU since the iPhone 6 is rumored to come with a higher-resolution display. A G6630 with a similar clock increase as the rumored CPU increase would give up to ~2.3x the theoretical performance of the A7's GPU.
Also, according to the report, the iPhone 6's A8 will still be dual-core.
I'm not really surprised, since despite an early report stating quad-core I didn't think Apple would need to make that jump given the performance of the Cyclone cores (to be fair, I don't recall the report saying what device the chip would be used in). I think the A8 will have a larger performance improvement in the GPU than the CPU since the iPhone 6 is rumored to come with a higher-resolution display. A G6630 with a similar clock increase as the rumored CPU increase would give up to ~2.3x the theoretical performance of the A7's GPU.
There's no legitimate use for quad-core CPUs in most desktop PCs, much less mobile devices. GPU speed is more the limiting factor right now, especially on tablets.
With Metal API, apple will get more use out of the cores they do have, I suppose their reasoning is that is enough.
With Metal API, apple will get more use out of the cores they do have, I suppose their reasoning is that is enough.
iPhone5S CPU = 1.2GHz vs. those hypothetical 2.0GHz = +67% frequency
G6430@450MHz in A7 * 67% = ~750MHz
G6630@450MHz = A7 GPU + 50%
G6630@750MHz = A7 GPU + 150%
Sounds great for a tablet platform but for a smartphone IMHO way over the top. No idea what Apple has done this time, but extravagant frequencies on an already quite "wide" processor, a severely beefed up GPU with equally extravagant frequencies and a bigger display area don't sound like the typical "Apple" iPhone to me unless they also take a 180 degree turn and suddenly go for >2500mAh smartphone batteries all of the sudden.
In order to believe that I'd rather believe to see a Tegra K1 GPU clocked at 750MHz with the CPU at 2.0GHz in a smartphone with a 2500mAh battery and a 4 hour battery life under heavy 3D. Coincidentially the GK20A GPU reaches at 750MHz the same 288 GFLOPs FP32 as a G6630@750MHz.
Yes Apple is now at 20SoC at TSMC but I don't think that one has all that many power benefits against Samsung's 28nm either. Finally if Apple again goes for one and the same SoC for all their i-devices just with frequency differences as with the A7 I honestly cannot understand the reasoning behind it. It'll again create an overpowered iPhone (always in a relative sense) and a quite underpowered tablet/iPad.
iMacmatician
Regular
The iPhone 5S's CPU runs at 1.3 GHz AFAIK (doesn't really change your conclusion though).
Could it be that Apple is introducing some kind of turbo boost into the A8's CPU and the 2.0(+) GHz is actually the maximum boost clock? The translated report says "its main frequency to 2GHz or more" which could means base CPU speed but I don't think it specifically implies that.
Could it be that Apple is introducing some kind of turbo boost into the A8's CPU and the 2.0(+) GHz is actually the maximum boost clock? The translated report says "its main frequency to 2GHz or more" which could means base CPU speed but I don't think it specifically implies that.
Donno I thought more cores = less power (for same performance) thus it benefits mobile (where batteryusage/power is very important) more than desktop (where power comes from the wall)
No idea if accurate, but it states 1.2 here: http://gfxbench.com/device.jsp?benchmark=gfx30&os=iOS&api=gl&D=Apple+iPhone+5S&testgroup=info
iPad Air is at 1.4GHz according to Gfxbench data.
I've read elsewhere another rumor about 2.5GHz which could be a tablet maximum frequency under the above reasoning. I don't know what to think though. Apple never really liked extravagant frequencies this far, but on the other hand Cyclone sounds "wide" enough where one has to wonder if it really makes sense to go any wider at this stage.
For the GPU a 6630 at 600-650MHz could make perfect sense for a tablet; any frequency above that is pushing the envelepe IMHO.
iMacmatician
Regular
Interesting. I got the 1.3 GHz number from AnandTech. Geekbench data also seem to indicate 1.3 GHz.
Perhaps - in an ideal world, but will you necessarily have work for all those cores?
Writing multithreaded software is challenging even for the best of game devs out there, can we really expect slobbish mobile devs to have the wherewithal to manage four, maybe eight cores...? I'm sceptical...
I mean, ppl on desktop's dont normally care as much about power consumption as ppl with mobile because they have an unlimited supply of power coming out of the wall socket
Install one of the CPU moniting apps for your phone, you may be surprised how many of those cores are doing stuff
True (albeit perf/W is a quite important factor these even for the HPC market); that isn't however directly linked to the amount of cores either, because it comes down to how "wide" each CPU core exactly is. If you have something like A7 or even A57 cores why not have a "sea" of them since they're miniscule and yes there threads can be spread over N cores at a time to save power. In the case of Cyclone you have very "wide" cores where the necessity of increasing core count is far less a necessity.
Considering the two Cyclone cores at A7 which clock from 1.2 to 1.4GHz depending on device you could either increase core count from 2 to 4 and only slightly increase frequency or you could just scale frequency. Nothing else changed in the latter case you will get roughly an as high performance increase as the frequency increases at the cost of N% more power consumption which could be absorbed up to a point going from 28 to 20nm. In the first case however going to quad cores power isn't going to increase most likely, while perfornance won't either because it doesn't scale linearly with increasing CPU core amounts.
So if I for example can execute 4 parallel threads on "just" 2 cores faster than on 4 cores what would that tell me then? Isn't a sterile core count completely meaningless under that light?
As it has been so far, single core performance is a lot more important on mobile devices. Just take browsing as an example.
I can understand if Apple isn't vested in more cores than two at this point. There are still gains to be made staying with two wider cores. Double the ram and increase IPC and the A8 will be all they need in mobile devices this year.
I can understand if Apple isn't vested in more cores than two at this point. There are still gains to be made staying with two wider cores. Double the ram and increase IPC and the A8 will be all they need in mobile devices this year.
As iOS GPU drivers, and perhaps the benchmark itself, became more refined with the fill rate test for GfxBench 2.x, the variability in results lessened, so inferring the GPU configuration (TMUs, clocks, etc.) seemed a little easier than with the fill rate test in GfxBench 3.0.
Results for Apple Ax SoCs appeared to be approaching 100% fil rate effectiveness back on that test, and the numbers for the A7's GPU in the iPhone 5s matched up closely with a G6430 @ 433 MHz, which, while not a requirement, would make a good complement to the 1.3 GHz CPU clocks that some people were seeing with some CPU profile tests. When the iPad Air was later measured in the fill rate bench, it scored a marginally higher result than the 5s and one which slightly exceeded the theoretical max for a G6430 @ 433 MHz, so a 466 MHz clock seemed a nice fit for its 6430 and would also make a nice complement to the 1.4 GHz clock being detected in its CPU tests.
Back with the second gen iPod touch and then with the 3GS's SGX535, Apple was fairly aggressive with GPU clock rates relative to other MBX and SGX configs of the time, namely compared to TI and the OMAP platform. I could see Apple being aggressive again with the clocks of a G6630 for A8 with the benefit of TSMC's 20 mn process, though the major challenge they face is that they tend to need one of their highest bins for their highest volume, multi-ten million unit product, the latest iPhone. So, unlike the other SoC makers, they can't headline some extra high clock speed while actually yielding the bulk of their product rated at a more attainable clock target.
Results for Apple Ax SoCs appeared to be approaching 100% fil rate effectiveness back on that test, and the numbers for the A7's GPU in the iPhone 5s matched up closely with a G6430 @ 433 MHz, which, while not a requirement, would make a good complement to the 1.3 GHz CPU clocks that some people were seeing with some CPU profile tests. When the iPad Air was later measured in the fill rate bench, it scored a marginally higher result than the 5s and one which slightly exceeded the theoretical max for a G6430 @ 433 MHz, so a 466 MHz clock seemed a nice fit for its 6430 and would also make a nice complement to the 1.4 GHz clock being detected in its CPU tests.
Back with the second gen iPod touch and then with the 3GS's SGX535, Apple was fairly aggressive with GPU clock rates relative to other MBX and SGX configs of the time, namely compared to TI and the OMAP platform. I could see Apple being aggressive again with the clocks of a G6630 for A8 with the benefit of TSMC's 20 mn process, though the major challenge they face is that they tend to need one of their highest bins for their highest volume, multi-ten million unit product, the latest iPhone. So, unlike the other SoC makers, they can't headline some extra high clock speed while actually yielding the bulk of their product rated at a more attainable clock target.
anexanhume
Veteran
They've increased RAM at least every other iPhone release. With the increased pressure of 64 bit pointers, I think the move to 2GB is a pretty solid bet. They also have to plan for the future and the fact that each iOS release has to support iPhones that are 3 years old. Do they want iOS 11 supporting a phone with only 1GB of RAM?
Real question is if there's any weight to the LPDDR4 rumors or we can expect LPDDR3 again.
What about GX6450? The timetable seems aggressive, but that was also the reason we dismissed 64 bit rumors for A7. Instead, we got a full custom 64 bit ARM dual-core.
Yeah, their drum-beat has been wider, not more or faster. They've not played into the numbers marketing war there and given they're so tightly coupled with their OS and performance metrics, I trust they'll make the decision that works best for performance and battery life.
Last edited by a moderator:
note this is nvidia (and not an independent entity)
http://www.nvidia.com/content/PDF/tegra_white_papers/Benefits-of-Multi-core-CPUs-in-Mobile-Devices_Ver1.2.pdf