Apple A9X the new mobile SoC king

[thunng8]

Full results are now on the geekbench site. That single core performance is huge..

https://browser.primatelabs.com/geekbench3/3515465

 

[gdansk]

We need to make an opensource re-implementation of GeekBench. Preferably removing compression and hash algorithms which use dedicated instructions.
Secondly Intel needs to stop selling gimped chips. i5 or above only. And these results suggest the A9X will do better in GB3 than the 5y71.

Samsung Galaxy S6 for comparison:

 

[thunng8]

Here:

http://mobilesyrup.com/2015/09/22/iphone-6s-and-iphone-6s-plus-review/

6s scores 28400 in ice storm unlimitedr

 

[Dribble]

Maybe, we'll see. I don't think you should underestimate Intel, though. They might not be so great at fielding competitive mobile cores/SoCs (frankly the execution here has been dismal), but they still have excellent CPU architects and the world's best manufacturing technology working on their "big" cores.

Intel's big problem is the thing that made them all that money - x86 is a closed eco system. You can't design your own custom chips, or manufacture them. You've got to take whatever Intel give you and pay through the nose for it to maintain Intel's high margins. ARM is open - you can design whatever chip you like with whatever additional modules you fancy and get it manufactured at whoever gives you the best deal.

Hence the moment ARM gets even vaguely competitive with Intel in a market outside of traditional windows x86 PC's (where MS windows legacy means x86 is a requirement) they are screwed. Pretty well all companies will happily loose a few % in performance if it saves them lots of money on chips, lets them build chips that more closely match their requirements, and stops them having to rely on Intel.

 

[Dresdenboy]

For Geekbench 2 John released some python code to create your own scores with individual weighting of subtests. Now it would require some html parsing. Just ignore combined scores.

 

[Rakehellion]

Note that as usual, numbers are skewed by SHA.

By how much?

 

[paffinity]

By how much?

It only skews it by ~200/300 points. A lot of the tests went up by 20+%.

 

[Eug]

It only skews it by ~200/300 points. A lot of the tests went up by 20+%.

If true, that would put the multicore score still above 4000 even adjusting for that, which is very decent, considering it's only dual-core. Those single cores are performance monsters. I wonder how much power each core uses.

It would be interesting to see Apple make a quad-core A10X iPad Pro in 2016. It'd probably blow everything else in ARM in this space out of the water, albeit at higher power usage.

BTW, as mentioned, I'm predicting a tri-core A9X iPad Pro to give us a multi-core Geekbench score of between 5700 and 6300ish, depending upon the clock speed. However, we won't find that out until around November.

 

[paffinity]

Correction, my statement are +20% over A8 at same frequency. At 1.8 GHz, most of the tests are showing +50-60% improvement over 1.4 GHz A8.

 

[freeskier93]

If true, that would put the multicore score still above 4000 even adjusting for that, which is very decent, considering it's only dual-core. Those single cores are performance monsters. I wonder how much power each core uses.

It would be interesting to see Apple make a quad-core A10X iPad Pro in 2016. It'd probably blow everything else in ARM in this space out of the water, albeit at higher power usage.

BTW, as mentioned, I'm predicting a tri-core A9X iPad Pro to give us a multi-core Geekbench score of between 5700 and 6300ish, depending upon the clock speed. However, we won't find that out until around November.

Adobe already let slip that the A9X is quad core...

 

[Eug]

Adobe already let slip that the A9X is quad core...

Since when? Last I checked, they only confirmed it had 4 GB RAM.

However, I could see it being a quad-core part, given Apple says it's up to 1.8X the speed of the A8X. If that's the case though, then my Geekbench 3 prediction of 5700-6300 would be way off, and the actual score would probably be well north of 7000.

Interestingly, 1.8 x 4500 (A8X) is 8100.

If the iPad Pro hits 8000+ in Geekbench 3, that would be shocking for an ARM tablet.

What would be next for A10X? The magical 10000 mark?!?

 

[freeskier93]

Since when? Last I checked, they only confirmed it had 4 GB RAM.

However, I could see it being a quad-core part, given Apple says it's up to 1.8X the speed of the A8X. If that's the case though, then my Geekbench 3 prediction of 5700-6300 would be way off, and the actual score would probably be well north of 7000.

Interestingly, 1.8 x 4500 (A8X) is 8100.

If the iPad Pro hits 8000+ in Geekbench 3, that would be shocking for an ARM tablet.

What would be next for A10X? The magical 10000 mark?!?

Whoops, you're right. 4 GB RAM not 4 cores.

 

[Phynaz]

Intel's big problem is the thing that made them all that money - x86 is a closed eco system. You can't design your own custom chips, or manufacture them. You've got to take whatever Intel give you and pay through the nose for it to maintain Intel's high margins. ARM is open - you can design whatever chip you like with whatever additional modules you fancy and get it manufactured at whoever gives you the best deal.

ARM is not open.

http://www.anandtech.com/show/7112/the-arm-diaries-part-1-how-arms-business-model-works/3

 

[scannall]

ARM is not open.

http://www.anandtech.com/show/7112/the-arm-diaries-part-1-how-arms-business-model-works/3

Anybody can license it and make their own ARM chips. It isn't Open Source', but it's certainly close enough. x86 is closed and proprietary. Surely you understand the distinction?

 

[jfpoole]

For Geekbench 2 John released some python code to create your own scores with individual weighting of subtests. Now it would require some html parsing. Just ignore combined scores.

For Geekbench 3 append ".gb3" to URLs to download the raw JSON result, which you can then parse with our Python scripts:

https://github.com/primatelabs/geekbench-tools/tree/master/geekbench3

 

[Phynaz]

Anybody can license it and make their own ARM chips. It isn't Open Source', but it's certainly close enough. x86 is closed and proprietary. Surely you understand the distinction?

Anybody?

Interestingly there are several companies that have x86 licenses, including AMD of course.

 

[raghu78]

Full results are now on the geekbench site. That single core performance is huge..

https://browser.primatelabs.com/geekbench3/3515465

Monster single core performance. GB3 score of 2500 at just 1.85 Ghz clocks. The A9X is going to be around 2700 on single core and 7200+ on multi core. Those are some monster perf numbers. Skylake Core M will get pounded on GB3 multi core.

Here:

http://mobilesyrup.com/2015/09/22/iphone-6s-and-iphone-6s-plus-review/

6s scores 28400 in ice storm unlimitedr

Given the fact that A8X was 40-50% faster than A8 in GPU performance we are looking at a GPU powerhouse inside the iPad Pro which sports the A9X. A9X is going to thrash Skylake Core M. I am looking forward to head to head benchmarking of both these chips in platform neutral benchmarks.

 

[Arachnotronic]

Monster single core performance. GB3 score of 2500 at just 1.85 Ghz clocks. The A9X is going to be around 2700 on single core and 7200+ on multi core. Those are some monster perf numbers. Skylake Core M will get pounded on GB3 multi core.



Given the fact that A8X was 40-50% faster than A8 in GPU performance we are looking at a GPU powerhouse inside the iPad Pro which sports the A9X. A9X is going to thrash Skylake Core M. I am looking forward to head to head benchmarking of both these chips in platform neutral benchmarks.

Shame that A9 still sucks at 3DMark Physics

But yes, this looks like a really fantastic chip.

 

[.vodka]

If Apple's architecture is able to scale to >3GHz on a less constrained setting (notebook, desktop) and they're able to gracefully handle another ISA change (and gathering enough industry support to get the most important programs out in ARMv8 versions apart from x86), I can see them getting rid of Intel in the future.

Those are some impressive numbers for a freaking phone.

 

[raghu78]

If Apple's architecture is able to scale to >3GHz on a less constrained setting (notebook, desktop) and they're able to gracefully handle another ISA change (and gathering enough industry support to get the most important programs out in ARMv8 versions apart from x86), I can see them getting rid of Intel in the future.

Those are some impressive numbers for a freaking phone.

I think Apple is already at work towards designing their custom ARMv8 core to scale to 3.5 - 4 Ghz speeds. I think we will definitely see Apple move all their products to ARMv8 in the not so distant future. It might happen at TSMC 16FF+ in late 2016 or TSMC 10 nm FINFET in late 2017. TSMC 16FF+ is a high performance process node and is suitable for high performance CPUs. Its the best foundry FINFET node without a doubt when it comes to power, performance and most importantly yields.

 

[Mondozei]

Honestly though, how will the average user notice any difference from the 6 to 6S? Both are monsters but I don't see how it makes much difference in day to day use. Do you need that raw performance aside from e-peen status? Don't think so.

Which is precisely why I agree with raghu; going for this extreme performance must have a longer-term objective. And that objective is unseating Intel, otherwise it makes little sense.

 

[NTMBK]

Honestly though, how will the average user notice any difference from the 6 to 6S? Both are monsters but I don't see how it makes much difference in day to day use. Do you need that raw performance aside from e-peen status? Don't think so.

They'll certainly notice the 2GB RAM, when their apps and web browser tabs aren't reloading all the time.

 

[CHADBOGA]

I was a AMD supporter. But the way things are going I am skeptical that they will even survive to launch Zen.

Has Internet Strongman Juanrga destroyed you so completely that you have totally lost hope with AMD and out of desperation have now jumped onto the ARM bandwagon?

 

[ShintaiDK]

Has Internet Strongman Juanrga destroyed you so completely that you have totally lost hope with AMD and out of desperation have now jumped onto the ARM bandwagon?

Seems its another case of the enemy of my enemy is my new best friend.

 

[raghu78]

They'll certainly notice the 2GB RAM, when their apps and web browser tabs aren't reloading all the time.

walt mossberg in his review for verge said the new iphones are blisteringly fast. The single thread benchmark scores on geekbench 3 back up his statement. The dual core A9 running at 1.85 Ghz is 50-60% faster than dual core A8 running at 1.4 Ghz. the 16nm FINFET process is one of the key reasons as it allows much higher frequencies at same power usage. The other reason which contributes to a lesser level is IPC improvements. The 2 GB RAM makes the phones much faster at general navigation and app loading. Also when browsing with 2 GB you can have more tabs open simultaneously and move very fast between them.