That's 42% of the 70% CPU improvement. That leaves ~20% improvement by microarchitecture. That's pretty astonishing.
Cheers
Apple A9 SoC
- Thread starter Entropy
- Start date
That's 42% of the 70% CPU improvement. That leaves ~20% improvement by microarchitecture. That's pretty astonishing.
Cheers
Haha wikipedia, uncited at that. But if that 2 GHz were true, and IPC hasn't decreased, it would rival the browser benchmark performance of the Macbook 2015 and its 14nm Core M 
, using data from Anandtech's reviews.
When Apple talks about performance improvements over the iPhone 6, I wonder if they make the comparison on the same base iOS version? Because browser perf in iOS 9 seems to have improved considerably. To that extent I hope that review sites re-test the iPhone 6 on iOS 9 before making any benchmark comparisons, seems logical but it rarely happens.
iPad Air 2 8 vs 9
https://infogr.am/ios_9_und_os_x_1011
, using data from Anandtech's reviews. When Apple talks about performance improvements over the iPhone 6, I wonder if they make the comparison on the same base iOS version? Because browser perf in iOS 9 seems to have improved considerably. To that extent I hope that review sites re-test the iPhone 6 on iOS 9 before making any benchmark comparisons, seems logical but it rarely happens.
iPad Air 2 8 vs 9
https://infogr.am/ios_9_und_os_x_1011
Nebuchadnezzar
Legend
Let's wait for some actual benchmarks before we base our estimates on Apple's 70% number.
If it can be done on ancient x86/64 it must be more easy on ARM.
Er, what makes you think "it can be done on ancient x86"?
On a completely different note, isn't it interesting how we had a spreadsheet worth of performance leaks before the launch of the new products, and complete silence after, when production has to be in full swing along with initial distribution?
On a completely different note, isn't it interesting how we had a spreadsheet worth of performance leaks before the launch of the new products, and complete silence after, when production has to be in full swing along with initial distribution?
Hasn't Apple been fairly truthful with their numbers in the past? Do we have a specific reason to doubt them this year?
I was 'refering' to IPC.
ltcommander.data
Regular
http://arstechnica.com/apple/2015/0...ce-preview-a-1-5ghz-apple-a8-with-2gb-of-ram/
So in addition to the bump to 2 GB of RAM, the A8 in the iPad Mini 4 has also been up-clocked to 1.5 GHz. This is the same as the A8X in the iPad Air 2, missing a core of course.
EDIT: Not sure the best thread to be posting this iPad Mini news, but it might have been better posting this in the Apple TV thread, since it's not A9 related.
So in addition to the bump to 2 GB of RAM, the A8 in the iPad Mini 4 has also been up-clocked to 1.5 GHz. This is the same as the A8X in the iPad Air 2, missing a core of course.
EDIT: Not sure the best thread to be posting this iPad Mini news, but it might have been better posting this in the Apple TV thread, since it's not A9 related.
Last edited:
Honestly, even with a A9 I wouldn't see much of a reason to upgrade, unless you just must have the latest and greatest (albeit it should be said the mini 4 should be more of an upgrade on the gpu side than on the cpu side). Even that "old" cpu still has the best single-thread performance out there for arm based tablets with that size.
But you have to agree the mini 4 provides quite some nice updates over the mini 2, unlike the mini 3 did (just pay extra for the touch sensor?). Some faster cpu (and more on the gpu side), that (important) 2GB memory upgrade, better camera, lighter (and you still get that touch sensor...). The mini 2 is still sold, but imho the mini 4 would be definitely a better buy. Apple still sells the original Air too (at the same price as the mini 4), which still only has a A7.
The other upgrade is the mini 4 has the laminated screen so it's thinner too.
I just received a BB flash sale email for $150 off the mini 3 too.
Unless iOS 9 really slows it down, I'll just hang on to the mini 2. In those Ars benchmarks, the iPhone 6 is faster than the mini 4 and I have an iPhone 6 Plus.
Then I guess mini 5 will get the A9. That is if they upgrade the mini next year.
I just received a BB flash sale email for $150 off the mini 3 too.
Unless iOS 9 really slows it down, I'll just hang on to the mini 2. In those Ars benchmarks, the iPhone 6 is faster than the mini 4 and I have an iPhone 6 Plus.
Then I guess mini 5 will get the A9. That is if they upgrade the mini next year.
Apple A8 already have pretty good IPC (its decoder is able to decode up to 8 instructions per cycle). If A9's clock rate is, say, 50% higher, then to improve or even maintain its IPC would require quite a lot upgrades on non-ISA related parts, such as branch predictor, cache, memory controllers, etc. These are no easier to do on a RISC CPU than on a x86 CPU.
You may pardon the OT but I was looking the other day for a decent 7-8" Android tablet to recommend to a friend at a low price. There really isn't anything compelling there to buy these days either anymore unless someone is willing to settle for ultra-crappy specs of a Sofia based super-low priced tablet. If you even go above the usual chinese OEMs and go for more expensive Android offerings you'd be damn lucky if you find anything with comparable CPU and GPU performance even to an Apple A7. That shouldn't mean that the mini4 wouldn't had been better off with something like a die shrunk A8X, but it's not like you can find in that class way faster offerings either elsewhere.
So you mean to say the A9 will perform better on not well written/optimized code (of which of course there is a lot) ie branchy code, code where prefetching is not effective (you can argue that this can not be avoided, but algorithms can be adapted to minimize this IMHO, but most programmers have no such deep notion of how a CPU works indeed)
For well written/optimized code the A9 should not be much faster as the A8, at same clock frequency.
Well in the UK, the iPad mini's start at £219, thats for an iPad Mini 2, which has an A7 which is still a pretty decent processor and will be for another few years. Looking at the new iPad Mini 4, it's £100 more expensive, sitting at £319. I'm not sure it's worth the £100 difference for basically having an A8 and TouchID.
I managed to pick up the Mini 3, 4 months ago for £249, when it was only £10 more expensive than the Mini 2, which I thought was worth it for only the TouchID.
In short though, for £219 you are getting a pretty damn good tablet with the Mini 2.
Again sorry for going OT.
Which tablet to choose depends which eco system you prefer.
Myself I'm not tied to a single eco system, but use all of them.
If you are a single eco system user, than you have either limited or more choice.
Personally I've an iPad air 2 and a Galaxy Tab S.
Both are great devices. The latter has better value for money.
Myself I'm not tied to a single eco system, but use all of them.
If you are a single eco system user, than you have either limited or more choice.
Personally I've an iPad air 2 and a Galaxy Tab S.
Both are great devices. The latter has better value for money.
Well, you can say that to almost any processor. For example, Intel's CPU have roughly the same peak IPC starting from Nehalem. However, almost all real world application runs much faster on Haswell than on Nehalem at the same clock rate, even without considering those new AVX instructions, so it's not just because those codes are not well written/optimized.
I don't know how well A9 will perform (I hope we'll have more information about it shortly
), but what I mean is that it's very difficult to improve real IPC once you have most of the low hanging fruits taken (which A8 sort of did as its IPC is already very good). If Apple managed to improve A8's IPC by 20% and runs at a much higher clock without needing to burn a lot of power, that'll be very very impressive.I'd say achieving 20% IPC improvement without increasing power draw is something intel would very much like to achieve. The improvement they got going from 32nm planar to 22nm FinFET was very modest indeed and that's the closest analog to the change Apple is doing now and claim a 70% (!!) total improvement from.
Most tests seem to put the combined IPC improvement of the last two generations (Haswell-Broadwell-Skylake) at less than 10% overall.
If intel could achieve 20% higher IPC at a given power draw, why wouldn't they deliver just that?
In reality it's a complex balancing act. You can raise IPC, but it costs gates, power and typically frequency. New lithography typically helps with cost per gate and power draw, which is one reason why intel has drawn out a perfectly straightforward set of increases in resources over several generations. Link to table. And note that at the top end frequency hasn't moved much for the last four/five generations, in spite of the advances in process tech. So for Apple to increase the performance of their cores by 70 or 80% over last year strains credibility. Raising both clock and IPC to that degree, when your starting point is arguably best in class as it is, well, I would call it borderline impossible. Which is why the oldtimers here speculated in increased number of cores and optimistic estimates of the benefits that would bring behind Apples numbers. If A9 is indeed a two-core design, and the performance is even within spitting distance of +70% over A8, then there will be a number of highly technical people itching to know just how the !?#%*?! Apple achieved that. In depth probing can't come soon enough.
The thing that strains credibility more than this would be them lying about it to the entire internet in their live-streamed broadcast.