Apple A13: Launch 2019-09-10 - Geekbench 4 & 5 scores available (GB4 multi-core 14070)

[Eug]

I just found this entry for iPhone12,1 on Geekbench. If real, that would make it the 2019 iPhone with A13:

https://browser.geekbench.com/v4/cpu/14528580

Single-core 5415, which would be about 12% faster than the fastest A12 score posted.

The multi-core score of 11294 isn't very impressive, though, since it's actually slower than some of the A12 scores.


[ EDIT 2019-09-11 ]

Better scores have shown up for Geekbench 4 at 5472 / 13769

iPhone12,3 - Geekbench

Benchmark results for an iPhone12,3 with an ARM processor.

browser.geekbench.com

Also, Geekbench 5 at 1324 / 3394

iPhone12,3 - Geekbench

Benchmark results for an iPhone12,3 with an ARM processor.

browser.geekbench.com

[ EDIT 2019-09-14 ]

The 14000 barrier has been broken for Geekbench 4 multi-core, and they're almost at 5500 single-core:

5491 / 14070



[ EDIT 2019-09-19 ]

Now the 3600 threshold for Geekbench 5 has been reached, with a single-core score of 1334 and a single-core integer score of 1226 in this case. Highest single-core score so far is 1340.

1334 / 3626

 

[Eug]

 

[Eug]

Matching things up to this article...

https://9to5mac.com/2019/07/23/iphone-11-models-camera-processor/

...it would appear that this leak is for the "R" version of the iPhone 11 series. That means the 2018 XR's 3 GB RAM is bumped to 4 GB RAM for the 2019 11R, but says nothing about the OLED models.

 

[Eug]

According to the creator of Geekbench, this score may be legit.

https://www.macrumors.com/2019/09/02/geekbench-a13-iphone-xr-2019/

The A13's high-performance cores are shown running at 2.66 GHz in today's result, compared to 2.49 GHz in the A12, leading to an approximately 12–13 percent gain in single-core performance for the A13 with a score of 5415, compared to an average 4796 for the A12 in the iPhone XR.

Interestingly, the A13's multi-core score of 11294 is nearly identical to the A12's average score of 11192, although Geekbench's developer John Poole tells us there could be some throttling due to thermal limits as similar situations have been seen with the A12 in the iPhone XS and XR, so we may have to wait for more data to see where the A13 truly tops out.

Careful observers will note oddly low figures for the L1 and L2 caches on this A13, but Poole tells us Geekbench has difficulty telling whether the cache values it reads are for the high-performance or high-efficiency cores, particularly on unreleased hardware for which the software hasn't been optimized.

While we can't confirm whether the Geekbench result is legitimate, as results certainly can be faked, all of the data appears reasonable or explainable and Poole tells us "there's nothing obviously wrong with the result."

 

[Nothingness]

Interesting, thanks for sharing.

The increase in score would be half due to frequency and "IPC".

It's also quite close to the i7-8557U (turbo to 4.5GHz) score found in some of the MacBook Pro: https://browser.geekbench.com/v4/cpu/compare/14530478?baseline=14528580

Except of course the MT score, but that's a phone

 

[DrMrLordX]

For reference, here's one of the faster MT scores for A12 in the GB4 database:

https://browser.geekbench.com/v4/cpu/11533846

There are some curious differences in scores under the MT section. It doesn't look uniform.

 

[Nothingness]

For reference, here's one of the faster MT scores for A12 in the GB4 database:

https://browser.geekbench.com/v4/cpu/11533846

There are some curious differences in scores under the MT section. It doesn't look uniform.

If that indeed is due to throttling I would expect odd fluctuations.

 

[JoeRambo]

Outright scary ST performance @ 2.66Ghz. Integer score is now full 6k versus 6k for 4+Ghz Skylakes and Zens. Apple somehow finds a way to get extra 10+% performance each year on already base performance.

 

[amrnuke]

Outright scary ST performance @ 2.66Ghz. Integer score is now full 6k versus 6k for 4+Ghz Skylakes and Zens. Apple somehow finds a way to get extra 10+% performance each year on already base performance.

When you control the hardware and the OS and the mainboard, and you have capital out the wazoo, and a clear goal, it's amazing what can be done. Also very important, obviously, to remember that their focus is a lot different from Intel/AMD's, where single-core advances are less important in desktop/HEDT/server situations than in a handheld smartphone that has very few things, comparably, going on at once.

It would be silly to assume that Intel/AMD couldn't do this if they built their own custom mainboard, memory subsystem, etc. and basically threw multi-core performance out the window. Even then, Intel's low-power Ice Lake only got 5600 on GB ST (18725 MT).

It would also be silly to assume that Apple couldn't scale up the A13 to be more multicore focused if a few heat/power constraints were removed.

Really excited overall for the advancements!

 

[Eug]

Interesting, thanks for sharing.

The increase in score would be half due to frequency and "IPC".

It's also quite close to the i7-8557U (turbo to 4.5GHz) score found in some of the MacBook Pro: https://browser.geekbench.com/v4/cpu/compare/14530478?baseline=14528580

Except of course the MT score, but that's a phone

How do we measure IPC though? It should be noted that IPC of this alleged A13 test is almost identical to IPC of A12X.

A12X also runs at 2.49 GHz like A12, but scores higher in single-core Geekbench 4 than A12. The highest recorded single-core score for A12 is 4872. The highest recorded single-core score for A12X is 5106, about 5% higher. If you scale that one A12X score up to 2.66 GHz, it's 5455 single-core. So, A12X is effectively the same as A13 single-core (5415).

Outright scary ST performance @ 2.66Ghz. Integer score is now full 6k versus 6k for 4+Ghz Skylakes and Zens. Apple somehow finds a way to get extra 10+% performance each year on already base performance.

Yes, that is a difference. Integer can reach up to about 5475 for A12X at 2.49 GHz, so scaled up to 2.66 GHz it's 5849. However, that A13 score for integer is 6110, about 4.5% higher.

 

[dark zero]

Apple is right now capable to sell their own uArch in their Macs, they have already enough brute force in order to do it.

 

[soresu]

I just found this entry for iPhone12,1 on Geekbench. If real, that would make it the 2019 iPhone with A13:

https://browser.geekbench.com/v4/cpu/14528580

Single-core 5415, which would be about 12% faster than the fastest A12 score posted.
The multi-core score of 11294 isn't very impressive though, since it's actually slower than some of the A12 scores.

Considering Cortex A76 can score up to 3600 single thread, and A77 is supposed to bring a 20% boost, so that is 4320.

Assuming A78/Hercules only hits ARM's conservative 15% estimate for yearly improvement made back in 2018 you get 4968.

That's within 10% of A13 for single thread, they've certainly come a long way since A7 cleaned their clock (ok that does work well as a tech pun....).

 

[soresu]

When you control the hardware and the OS and the mainboard, and you have capital out the wazoo, and a clear goal, it's amazing what can be done. Also very important, obviously, to remember that their focus is a lot different from Intel/AMD's, where single-core advances are less important in desktop/HEDT/server situations than in a handheld smartphone that has very few things, comparably, going on at once.

It would be silly to assume that Intel/AMD couldn't do this if they built their own custom mainboard, memory subsystem, etc. and basically threw multi-core performance out the window. Even then, Intel's low-power Ice Lake only got 5600 on GB ST (18725 MT).

It would also be silly to assume that Apple couldn't scale up the A13 to be more multicore focused if a few heat/power constraints were removed.

Really excited overall for the advancements!

Just looking at what game devs can wring out of PS4 hardware is evidence of what their chips can do in a highly optimised environment.

 

[Eug]

Launch of A13 is imminent. I'm going to take an extended lunch hour today.

Unfortunately, rumours are that no A13X exists. There may be (at least internally) an A12X revision, and an A14X in the works, but no A13X.

 

[Eug]

A13 Bionic launched

"The fastest CPU in a smartphone"
"The fastest GPU in a smartphone"

Whatever that means

 

[Eug]

8.5 billion transistors
2nd generation 7 nm process
4 efficiency CPU cores
Hundreds of voltage domains
Ton more clock gating domains

All cores up to 20% faster
All engines have lower power, eg. GPU up to 40% lower power

 

[DrMrLordX]

Unfortunately, rumours are that no A13X exists. There may be (at least internally) an A12X revision, and an A14X in the works, but no A13X.

Apple seems to skip A##x generations? Like there was no A11x, I think?

 

[Eug]

Apple seems to skip A##x generations? Like there was no A11x, I think?

So far there have been:

A8X
A9X
A10X
A12X

So as you said, no 11X. And possibly there will be no 13X.

 

[Eug]

Pretty vague, but...

Apple A13 Bionic: closer look at the world's most powerful smartphone chip

The world's most powerful smartphone chip, but here is exactly why...

www.phonearena.com

Apple Announces New iPhone 11, iPhone 11 Pro, & iPhone 11 Pro Max

www.anandtech.com

 

[soresu]

Pretty vague, but...

Apple A13 Bionic: closer look at the world's most powerful smartphone chip

The world's most powerful smartphone chip, but here is exactly why...

www.phonearena.com

Apple Announces New iPhone 11, iPhone 11 Pro, & iPhone 11 Pro Max

www.anandtech.com

It's very interesting, but the unfortunate problem with all iOS devices is that they lack the open software ecosystem of the Android based phones they are comparing against on that presentation.

What's the use in all that power when you get gimped by Apple in how to use it?

Mozilla can't even use their own Javascript engine for Firefox Focus on iOS, it's using WebKit/Blink's JS engine (I think?) - that's how restrictive they are.

This is not to say it isn't very interesting from the perspective of possible future ARM ports of big software like Autodesk Maya for when Apple finally make a full desktop switch to ARM, which will in turn benefit those open platforms like Android too.

 

[Glo.]

This is not to say it isn't very interesting from the perspective of possible future ARM ports of big software like Autodesk Maya for when Apple finally make a full desktop switch to ARM, which will in turn benefit those open platforms like Android too.

For that to happen Apple would have to be able to push through their CPUs more than 50 Amps, to compete with desktop x86 CPUs and their WORKLOADS. And their current designs cannot even come close to this.

 

[Ajay]

It's very interesting, but the unfortunate problem with all iOS devices is that they lack the open software ecosystem of the Android based phones they are comparing against on that presentation.

What's the use in all that power when you get gimped by Apple in how to use it?

Mozilla can't even use their own Javascript engine for Firefox Focus on iOS, it's using WebKit/Blink's JS engine (I think?) - that's how restrictive they are.

This is not to say it isn't very interesting from the perspective of possible future ARM ports of big software like Autodesk Maya for when Apple finally make a full desktop switch to ARM, which will in turn benefit those open platforms like Android too.

As an Apple user, I could care less, for the most part, what is under the hood. The performance, for my old 6s is very good for hat I do. I just wish the settings app would be rewritten, it’s gotten very large and unintuitive - thank God for the search function.

That said, I do look for a breakdown on AT by Ian, detailing the architecture of the A13.

 

[Mopetar]

Launch of A13 is imminent. I'm going to take an extended lunch hour today.

Unfortunately, rumours are that no A13X exists. There may be (at least internally) an A12X revision, and an A14X in the works, but no A13X.

Any A13X would only be announce alongside a new iPad Pro. The iPad Pro has closer to an 18 month product cycle, so there's likely not an update this fall and nothing until next spring at the earliest. It's unlikely that they'd put an A13 in a new iPad Pro since it would mean losing two whole performance cores which would have required one hell of a performance improvement in the design of A13 to compensate for. So it's possible that the chip exists, but that it isn't even in production yet.

 

[soresu]

more than 50 Amps

Errr... what voltage do you expect 50 amps to be delivered with??!!

Most extension sockets are only rated to 13 Amps max fuse.

 

[Roland00Address]

I am still amazed how fast the A10 is. The A12 and A12x is what 40% maybe 45% faster in single core. Lets say the A13 is 10 to 20% faster in single thread, in the end that means the A13 60% faster maybe 70% faster in single thread.

The limiting factor of the iPads and iPhones in the older models is the amount of ram that Apple is putting in the devices. That and the battery. Yes these new Apple SoCs are incredible but we are now reaching the point of diminishing returns.