Apple A8 and A8X
- Thread starter ltcommander.data
- Start date
After ~ 100 looped runs, Shield tablet has an average fps of ~ 56fps while iPhone 6+ has an average fps of ~ 32fps. That is a 1.75x improvement.
And FWIW, when framerate capped at 30fps, battery life nearly doubles on Shield tablet.
Battery life comparisons between a phone and a tablet make little sense because a tablet tends to have MUCH higher platform power consumption due to the much larger and more power hungry screen on a tablet. For example, the iPad Air has 3x more battery capacity than the iPhone 6 Plus, and yet achieves "only" 1.6x more battery life in the T-Rex Onscreen benchmark because the platform power consumption is much higher.
Last edited by a moderator:
...and we should NOT compare current generation smartphones with former generation tablets. A8 powered iPads shouldn't take too long to launch also.
Which then bears the question why any "fool" of a smartphone manufacturer hasn't yet used K1 for a smartphone, let alone one with "just" a 2000mAH battery.
Which then bears the question why any "fool" of a smartphone manufacturer hasn't yet used K1 for a smartphone, let alone one with "just" a 2000mAH battery.
loekf
Regular
Next year's A9 in a 14nm process from Samsung ?
http://appleinsider.com/articles/14...to-build-14-nanometer-apple-a9-chips---report
I don't think so honestly. If Intel is able to pull off 14 nm in Q4 with decent yields, best guess is 1.5 more years for the foundries... The A9 should have taped out already if it's up for the iPhone 6S / PlusS next year. 10-12 weeks for processing... 2-3 weeks packaging... 3 months for qualification/testsing... maybe a mask/metal spin.... start of production ramp up in Q2 2015..
Also doesn't make any sense to swap foundries and processes every year.
What Samsung and their Common Platform partners call 14nm is actually Finfet transistors combined with the BEOL from their 20nm node, given that Samsung is already producing SoCs on a 20nm node, their "14nm" node could easily be ready for mass production by Q4 2015. It will be interesting to see whether TSMC can offer Apple their enhanced 16nm Finfet Turbo product, in time for the A9 SoC.
Nebuchadnezzar
Legend
I don't think that people appreciate how far a lead Samsung has on 14nm compared to TSMC
http://www.eetimes.com/document.asp?doc_id=1324165
They've been in risk production since August.
Not sure it really matters. Going from one node to another, there probably isn't very much reuse at the transistor level so it's probably a performance/financial decision.
In terms of process nodes, I found this comparison between Intel 14nm and TSMC FinFet 16nm to be a practical summary.
Note that this describes TSMCs original FinFet and not FF+, which is what will be predominantly used. Given the claimed increases in density and performance of FF+, I could speculate as to what they have changed, but I'm a dilettante in these matters and prefer that someone else steps forward. I haven't found any official statement from TSMC specifying exactly what the process change entailed.
Bottom line - XXnm is a marketing term these days from all players, you'll need to dig deeper if you want to compare foundry processes.
Rumors say that TSMC introduced the 16nm FF+ specifically to strengthen their competitive position for Apples business, which if true suggests that this is the process Apple will use if they go with TSMC. I hope they do, I found the announcements from ImgTech regarding their work on TSVs on TSMC 16nm FF rather promising.
ltcommander.data
Regular
https://www.semiwiki.com/forum/content/3884-who-will-lead-10nm.html
It's interesting that Samsung's processes have had a density advantage for a while that Apple is letting go of by moving to TSMC.
It's interesting that Samsung's processes have had a density advantage for a while that Apple is letting go of by moving to TSMC.
Just out of curiosity? Where the heck do you see in the quoted text any reference to A9 and where it could be manufactured?
As for the real answer to the above there's not much to add to the several replies you've already received on the topic. In fact I'm not even sure anymore if Apple is going to use any 16FF at TSMC after all and not move from 20SoC directly after to Samsung/GF 14nm.
McHuj probably hit the nail on the head that it's at least partially a financial decision to move back to Samsung for manufacturing and from all so far indications it sounds more like they had some sort of one night (or two nights who cares) stand with TSMC and went back to bed with Samsung for manufacturing. IMHO it's both in Samsung's and Apple's best interest that the latter manufactures in Samsung's foundry. For the period Apple is manufacturing at TSMC is Samsung filling those large volume gaps at its foundry manufacturing something else? I haven't heard anything worthwhile for that but I always could stand corrected.
Those tables have some questionable data though.
Note for instance that Samsungs density should have worsened (!) by roughly 25% going from "20" to "14" nm nodes. If however I quote Samsung directly about their new process:
They make direct reference to density improvements vs. 20nm. Which seems reasonable. Having much worse density on the new process seems very unlikely. Also, the tables claim unchanged density from 20nm to 16nmFF+ for TSMC which directly contradicts TSMC claims of modest improvements. Further, the density values contradict published SRAM cell sizes. So.... I don't think we should draw too many conclusions about processes that aren't yet out in the wild.
Based on anecdotal evidence I've heard over the years, this table is wrong all over the place.
I don't know if it's the raw data that's wrong or just this form of density approximation, but I wouldn't take the conclusions too seriously.
As for process selection - how many times did analysts say that X company was going to switch to TSMC before it actually happened? They got it wrong 3 times before they finally got it right - even a broken clock is right twice a day, while analysts are only right twice a decade, it would seem. While it's interesting to speculate about TSMC vs Samsung on 16/14nm and customer traction between foundries, history teaches us it's usually safer to just ignore analyst reports and make a random guess instead.
I don't know if it's the raw data that's wrong or just this form of density approximation, but I wouldn't take the conclusions too seriously.
As for process selection - how many times did analysts say that X company was going to switch to TSMC before it actually happened? They got it wrong 3 times before they finally got it right - even a broken clock is right twice a day, while analysts are only right twice a decade, it would seem. While it's interesting to speculate about TSMC vs Samsung on 16/14nm and customer traction between foundries, history teaches us it's usually safer to just ignore analyst reports and make a random guess instead.
Nebuchadnezzar
Legend
He clearly states Samsung 20nm is just projected by previous scaling factors and they don't have proper numbers. The rest are either measured or publicly available values though.
I see that ZDNET are quoting "sources familiar with the matter" as saying that Samsung has 30% of the A8 production, and has the A9 contract on 14nm.
http://www.zdnet.com/samsung-expects-sales-boost-from-apple-chip-deal-7000034269/
http://www.zdnet.com/samsung-expects-sales-boost-from-apple-chip-deal-7000034269/
One thing about buying SoC from others is, your road map is basically controlled by others.
Samsung was actually quite a good SoC partner for Apple, because almost no one else (except Samsung itself) use that, so in a sense Apple could put some influences on Samsung's SoC design. If Apple buys from Qualcomm, that'd be much less possible.
However, since Samsung became a serious competitor to Apple in the mobile phone market, using Samsung's SoC does not seem to be a great idea. Apple could find another smaller SoC partner, but it's hard to find someone with the qualification. Apple seemed to foresee that problem and bought P.A. Semi to design their own SoC.
The result is that Apple can design their own SoC to fit their own road map (compare this to the situation of Mac, where Apple is unable to make a retina Macbook Air because Intel couldn't make Broadwell on time).
Apple did get a dedicated part from Samsung even for the first iPhone. It was similar to the MBX Lite + ARM11 SoC Samsung was offering to the general market, though.
I'm sure the SoC design expertise of both companies mutually benefited from the collaboration in the early generations.
I'm sure the SoC design expertise of both companies mutually benefited from the collaboration in the early generations.
I copied your reply from the Tegra thread since its getting too much OT there. I don't know where the GX6450s are clocked in A80 but I could imagine that in the 6 Plus the GPU goes slightly over 500MHz. They could go for 700 or even more MHz, but to stay at a similar to iPhone/A8 frequency with a GX6650 sounds way more reasonable from a power budget to me.