Apple A12 & A12X *** Now A12Z as well *** Now in a Mac mini

[PeterScott]

Top score to top score I'm seeing 13% in Int and 15% in FP. In multi-core the difference is 6% for Int and like 2% for FP.

Geekbench is probably the most optimistic as it runs for a few seconds per sub-test.

Fitting for quick burst of clock speed. Really I doubt the actual CPU core changed, more like clock speed/cache changes.

The bigger boost will be in GPU and Neural stuff.

 

[Mopetar]

there's the whole "obvious" issue. it's a dock. one dock is as obvious as any other.

Not to derail, but patents don't work like that at all. There are patents for objects that are all functionally the same, but have very minor implementation differences. You would really need to read the entire patent to see exactly what Apple is claiming. It doesn't matter if 95% of it already exists or is "obvious" as long as they've got something new that isn't covered by other patents or prior art. It's a bit like saying that you couldn't patent a rotary engine since one engine is as obvious as any other.

And that is precisely what we see here. Both Integer and Floating point scores increase by 15%.

Typically when Apple gives performance quotes they cherry pick one benchmark that does exceptionally well or saw a big performance gain so they can claim up to 50% or something like that. The average performance is usually good, but no where near what they claim. Though to be fair, I didn't watch the presentation so I don't know if they claimed average performance or what wording they used.

Really I doubt the actual CPU core changed, more like clock speed/cache changes.

The bigger boost will be in GPU and Neural stuff.

I can see a case for minimal CPU changes since they probably needed to spend time fixing any specter vulnerabilities along with some focus going towards their neural engine. The GPU also probably was a larger focus since they're still relatively new to that and probably had a lot of room for improvement. Also, their CPU performance is good enough where focusing on that (even if they do get really good performance gains) just provides diminishing returns. Even my 3 year old phone still feels plenty fast.

 

[jpiniero]

Fitting for quick burst of clock speed. Really I doubt the actual CPU core changed, more like clock speed/cache changes.

It appears that the clock speed is only 100 Mhz higher (2.5 vs 2.4). The CPU improvement is mostly due to arch changes. It also appears that they had originally intended to go higher but dialed it back because TSMC missed the quality target and/or to improve yield.

 

[Eug]

Typically when Apple gives performance quotes they cherry pick one benchmark that does exceptionally well or saw a big performance gain so they can claim up to 50% or something like that. The average performance is usually good, but no where near what they claim. Though to be fair, I didn't watch the presentation so I don't know if they claimed average performance or what wording they used.

IIRC, usually Apple does not do that much these days. They provide optimistic benchmarks, but usually it's within reason. They aren't going to pick one anomalous subtest to base their marketing around, I suspect because that would just backfire for them anyway.

I can see a case for minimal CPU changes since they probably needed to spend time fixing any specter vulnerabilities along with some focus going towards their neural engine. The GPU also probably was a larger focus since they're still relatively new to that and probably had a lot of room for improvement. Also, their CPU performance is good enough where focusing on that (even if they do get really good performance gains) just provides diminishing returns. Even my 3 year old phone still feels plenty fast.

I just updated to iOS 12 today, and suddenly my iPhone 7 Plus feels like a brand new machine. I'm very impressed with their under-the-hood optimizations. I don't know if some of that is some sort of UI trickery, but there are definitely real performance improvements.

In fact, it's so fast that I won't be upgrading from that phone this year. However, I will pick up an iPhone XR for my wife, to upgrade from her old iPhone 6s with that old Apple A9 CPU. My A10 Fusion is about 50% faster in Geekbench 4 CPU than the A9, and it's about 25% faster for Geekbench 4 Metal. However, A11 is a whopping 70% faster than my A10 for CPU though, and about 20% faster for Metal. And then A12 adds to that. So yeah, I think they're good for a while for CPU. They may be spending more time with their GPU and Neural Engine.

It should be noted that while the XS and XS Max seem to have enough stock to roughly meet demand, they did delay the XR, which some think may sell in even higher numbers. All three have the same SoC. I don't know if the XR delay is purely a marketing ploy to push early adopters to the MUCH higher priced XS and XS Max, or if it had to do with component shortages. I did see some rumours about an LCD yield issue for the XR, but one wonders if TSMC couldn't produce enough of the 7 nm A12 Bionic to support all three models at the same time.

 

[Thala]

Top score to top score I'm seeing 13% in Int and 15% in FP. In multi-core the difference is 6% for Int and like 2% for FP.

Geekbench is probably the most optimistic as it runs for a few seconds per sub-test.

Indeed the multi-core score is a combination of big.LITTLE and is not conclusive if we want to reason about the big cores.
As it stands with the current results we have around 4.2% clock increase and roughly 10% IPC increase for the big cores.
The little cores typically run at lower clock speed anyway, which is not reported by Geekbench.

In addition i expect the average score will increase slightly as we get more entries into the database. Top-score to top-score discrepancy is guaranteed to increase even more as the variance increase with more entries. So in about a month or two we will have a more complete picture.

 

[ElFenix]

Not to derail, but patents don't work like that at all. There are patents for objects that are all functionally the same, but have very minor implementation differences. You would really need to read the entire patent to see exactly what Apple is claiming. It doesn't matter if 95% of it already exists or is "obvious" as long as they've got something new that isn't covered by other patents or prior art. It's a bit like saying that you couldn't patent a rotary engine since one engine is as obvious as any other.

i know the patent system is broken, yes. "minor implementation difference" shouldn't be enough to get a patent.

 

[Headfoot]

i know the patent system is broken, yes. "minor implementation difference" shouldn't be enough to get a patent.

This is really pretty OT. I practice in patent law space and it is significantly more complex than that. An entirely separate topic.

 

[deathBOB]

i know the patent system is broken, yes. "minor implementation difference" shouldn't be enough to get a patent.

This is really pretty OT. I practice in patent law space and it is significantly more complex than that. An entirely separate topic.

The first law of discussing patents is that those with the least knowledge will have the strongest opinions.

 

[AkulaMD]

The first law of discussing patents is that those with the least knowledge will have the strongest opinions.

Couldn't agree more. It seems easy and straightforward to those who understand the least.

 

[Eug]

In Geekbench 4 multi-core, A12 is 11% faster than A11 (11515 vs 10357).

https://www.tomsguide.com/us/iphone-xs-max,review-5747.html

 

[StinkyPinky]

The iPhone Xs and the iPhone Xs Max also shined in other real-world tasks, such as video editing. It took Apple's phones just 39 seconds to transcode a 2-minute 4K clip to 1080p. The Galaxy S9 took 2:32, and the OnePlus 6 finished in 3:45.

How about opening apps? The iPhone Xs took 20.8 seconds to open Fortnite, 4.9 seconds for Pokémon Go and 6.17 seconds for the Asphalt 9 racing game. The Note 9 was slower across the board at 35 seconds, 7.2 seconds and 9.1 seconds, respectively. The older iPhone X was also slower than the iPhone Xs at 26, 7.2 and 10 seconds for the above apps.

Yikes.

 

[Eug]

Yikes.

I'm not sure what to make of the transcoding benches, since there is no mention whatsoever of how the test was done.

BTW, here are the graphics benches:

To test graphics performance, we ran 3DMark Slingshot Extreme, and the iPhone XS Max scored 4,339 and the iPhone XS hit 4,244. Both of those numbers are below the Galaxy Note 9's 4,639, but they're a lot higher than the 2,227 turned in by the older iPhone X.

In a separate graphics benchmark that has a more intense workload, GFX Bench 5, the iPhone XS Max crushed the the Galaxy Note 9. Apple's phone scored higher Aztec Ruins test (1,604.7 frames to 851.7 frames) and on the Car Chase test (1,744.44 frames to 1,103 frames).

 

[Eug]

https://wccftech.com/apple-iphone-xs-a12-antutu/

Within a week of its launch, today the iPhone XS is the subject of an Antutu score. Apple’s 2017 iPhone X performs quite well in the benchmark app. The smartphone scores between 210,000-and 230,000 points. This score range is laughable when we consider today’s results for the iPhone XS.

The smartphone scores 363,525 points in Antutu and matches Apple’s claims for improvements. At its launch event last week the company claimed up to 50% performance increase for the A12’s GPU. This gain reflected in today’s Antutu run, as the Apple A12 scores 150,931 points in graphics.

The score marks for a one-third increase over the A11. They also beat current 10nm processors from Android. The Samsung Galaxy S9+ with Qualcomm’s Snapdragon 845 scores around 260,000 points. Antutu is not a good indicator of daily performance as users who’ve experienced the S9+’s overheating will attest.

 

[Mopetar]

The iPhone Xs and the iPhone Xs Max also shined in other real-world tasks, such as video editing. It took Apple's phones just 39 seconds to transcode a 2-minute 4K clip to 1080p. The Galaxy S9 took 2:32, and the OnePlus 6 finished in 3:45.

How about opening apps? The iPhone Xs took 20.8 seconds to open Fortnite, 4.9 seconds for Pokémon Go and 6.17 seconds for the Asphalt 9 racing game. The Note 9 was slower across the board at 35 seconds, 7.2 seconds and 9.1 seconds, respectively. The older iPhone X was also slower than the iPhone Xs at 26, 7.2 and 10 seconds for the above apps.

Yikes.

Maybe that's the real reason that Apple isn't working as hard on the CPU gain. Their next best competition is really just last year's product. Hopefully they don't get too complacent. Intel was bitten badly by that and Apple is hardly immune.

 

[DrMrLordX]

Hopefully they don't get too complacent. Intel was bitten badly by that and Apple is hardly immune.

I guess it depends on where Apple wants to take their CPUs. If all they want is to dominate the phone/tablet space, they won't do much. If they want to expand their designs into the notebook/PC market and beyond, they need to keep pushing harder for more gains.

 

[jpiniero]

https://www.notebookcheck.net/Apple-iPhone-XS-Smartphone-Review.333546.0.html

More benchmarks. Roughly 18% faster in JetStream, Kraken and Octane. For some reason it's slower than the iPhone X tested in WebXPRT 2015 though.

 

[Eug]

Geekbench 4 single core:

Geekbench 4 multi-core:

As of next month, I will have both the Xr (10943) and the iPad 2 (575) in active use at home. The Xr is 19X as fast as the iPad 2.

I also have the 7 Plus as my main driver for my phone. It does not feel slow at all, esp. in iOS 12 (which included performance improvements over iOS 11). I know Geekbench 4 may not be the best cross platform benchmark in the world, but nonetheless it does give us an idea. With that caveat, it's not surprising the 7 Plus doesn't feel slow, since it actually scores higher than the 2012 Core i5 13" Retina MacBook Pro, as well as faster than the 2015 and 2016 Core m3 12" MacBooks.

 

[Mopetar]

It’s weird looking at that chart. I still use an Air 2 regularly and it still feels quite snappy.

It’s hard to imagine a phone at 3x that.

 

[Eug]

I was in the store today comparing the iPhone Xs, Xs Max, 8, 8 Plus, and my 7 Plus.

The Xs/Max felt just a touch snappier overall than my 7 Plus, and maybe even more snappier for the camera, but for surfing it really felt similar. It seems the bottleneck for surfing was internet access most of time, not rendering speed. The 7 Plus was already a fast enough renderer, at least for average web pages.

I'm sure that will change in a couple of years though.

OTOH, the iPad 2 (not Air 2) is basically unusable for surfing, probably not just because of rendering speed, but also because it only has 512 MB RAM. When the iPad 2 came out in early 2011, it was actually an OK surfer.

 

[Eug]

A12 Bionic gets 3435/8129 in low power mode.

https://browser.geekbench.com/v4/cpu/9986297

Dunno clock speed though, since it reads as 0 MHz.

---

Fastest multi-core speed at full power is 11714.

https://browser.geekbench.com/v4/cpu/10012187

---

With that kind of speed in lower power mode, you could run it full time, and it'd still be plenty fast, about 35-40% faster than a 7 Plus.

 

[bakyt115]

https://twitter.com/dhh/status/1043277162676072449

I can't believe that phone chip faster than desktop chip

 

[Eug]

https://twitter.com/dhh/status/1043277162676072449

I can't believe that phone chip faster than desktop chip

Hmm...

 

[Eug]

https://www.macworld.com/article/3308541/iphone-ipad/iphone-xs-iphone-xs-max-benchmark-results.html

 

[Hans de Vries]

Apple A12 floorplan: So many unidentified units.....

 

[Eug]

Apple A12 floorplan: So many unidentified units.....

A12:   83.27 mm2 TSMC     7 nm
A11:   87.66 mm2 TSMC    10 nm
A10X:  96.4  mm2 TSMC    10 nm
A10:  125    mm2 TSMC    16 nm
A9X:  147    mm2 TSMC    16 nm
A9:    96    mm2 Samsung 14 nm
A9:   104.5  mm2 TSMC    16 nm
A8X:  128    mm2 TSMC    20 nm
A8:    89    mm2 TSMC    20 nm
A7:   102    mm2 Samsung 28 nm
A6X:  123    mm2 Samsung 32 nm
A6:    97    mm2 Samsung 32 nm
A5X:  165    mm2 Samsung 45 nm
A5:    69.6  mm2 Samsung 32 nm
A5:   122    mm2 Samsung 45 nm
A4:    53    mm2 Samsung 45 nm