iPhone 15 Pro Geekbench Scores Confirm Apple's Faster A17 Pro Chip Performance Claims, 8GB of RAM

[Mac_The_Ripper]

The whole world is going straight into IA, and you all think doubling the neural engine performance is bad? Apart from the redesigned GPU that probably will accellerate ML performance even more?

Just think about the possibilities of running a decently performance ligh ChatGPT in your pocket, without disclosing any of your personal data.

Now picture this scaling up to a M3 Ultra, if the current M2 Ultra performance with Lama model is already 12 tokes/sec…. Rivaling a Single A100 (nvidia best performance neural GPU), we could potentially have two A100 (or more) running in a small factor desktop machine.

ChatGPT is the most overrated application I've ever seen. It is great for making lists though, I will give it that.

 

[Darren.h]

this probably means the 3nm Apple Silicon for the upcoming Macs is gonna fall far short of the 3nm performance gains everyone was bragging about. probably just 20 percent for the M3 Pro and Max compared to the 5nm M2 Pro and Max.

the 2nm chips will be the final end of any performance gains. what's left? Just adding more cores.

 

[Darren.h]

meanwhile the cheap i9 13900k hits a blistering 9 GHZ setting an all time record.

Intel's Core i9-13900K Processor Gets Overclocked to 9GHz

By using liquid helium as a cooling solution, the chip was able to break the 9GHz barrier and obtain a new world record.

www.pcmag.com

 

[a.coward]

Is it true ray tracing on the new chip is 30 fps only? Also ray tracing on a small phone screen isn’t gonna exactly gonna blow anyone away.
Seems pointless to me at the moment

Gaming aside, ray tracing should also help with Augmented Reality.

 

[nordique]

Man 3nm is disappointing. Yes the increase performance is nice but I expected it to be higher
Or at the very least if less performance increase then higher battery life - the battery staying the same was a bummer
Now this “less than expected” performance increase
Makes me doubt the m3 chips will really be as impressive as I thought they’d be
Jmo

 

[DavidSchaub]

GeekBench is clearly not running for a long enough period of time?

The limited thermal/power envelope really should be representative in these numbers.

I have difficulty wrapping my head around GeekBench telling me that the iPhone 15 Pro has faster single-thread performance than any Mac (the M2 Mac Studio gets 2803), or that the iPhone 15 Pro has almost as fast multi-thread performance as the iMac Pro?

 

[a.coward]

I don’t know, after all the talk all year about this chip, perhaps I was expecting more?

And can some explain ray tracing to me?

Without going into too much details, Ray Tracing is a bunch of math with models light sources (rays) as it interacts with objects in a scene (tracing light as it bounces off from one object to another object until it reaches the observer).

Typical 3D will model objects as "ambient" sources of light. However, in the real world, most object do not emit their own visible light, but "reflect" off of other sources of light (e.g. the Sun). Ray tracing attempts to simulate these sources of light. This is great for things such a reflections, textures, transparencies and shadows, adding realism.

 

[Lakersfan74]

Nice upgrade from my 12P.

 

[CarAnalogy]

I know some are a bit disappointed with the speed increases here though they aren’t bad. What isn’t talked about and is obviously a big focus for Apple is the Machine Learning part of the chip where they doubled the performance. They obviously feel this is where it makes more sense to invest in performance improvements.

I kind of agree as this can open new features and capabilities not possible before. There’s quite a few new software features being added these days that are driven by those parts of the chips. Things that might help sell more phones vs 10% more CPU performance on a phone that no one I’ve ever come across says is too slow.

In theory but they are locking too many of these behind hardware upgrades. Why have they been touting the neural engine for the past several years, but suddenly on-device Siri and other improvements require a “25% faster neural engine”?

The pinching gesture has existed for several years now and uses all the same sensors that already exist. They claim it’s ML that makes it work, but it doesn’t work on anything except their brand new neural engine.

I will give it to them though, they fixed autocorrect. I turned it back on in the 17 RC and actually left it on and find it to be helpful. Which I haven’t felt since the first iPhone, if ever. They claim that’s thanks to ML too but it’s been such an egregious problem for so long I think even they would hear the backlash for locking that behind a hardware upgrade.

 

[jole]

What isn’t talked about and is obviously a big focus for Apple is the Machine Learning part of the chip where they doubled the performance. They obviously feel this is where it makes more sense to invest in performance improvements.

Locally run, Llama -like, LLMs coming to iPhone and powering the next gen Siri.

Key question is, will A17 Pro be capable enough to run them or do we need to upgrade next year?

The model will also need a lot of RAM. Today quantized 7 billion parameter Llama 2 needs 3.6GB and 13 billion parameter version 6.9GB. Maybe A17 Pro could even run the 13B version today (and definitely can run the 7B version), but they would need to seamlessly page out everything else from RAM while processing Siri request.

 

[smulji]

Man 3nm is disappointing. Yes the increase performance is nice but I expected it to be higher
Or at the very least if less performance increase then higher battery life - the battery staying the same was a bummer
Now this “less than expected” performance increase
Makes me doubt the m3 chips will really be as impressive as I thought they’d be
Jmo

What hasn't been talked about a lot on tech sites but was mentioned in the keynote ( a few times actually) was the idea of SUSTAINED performance which is the processor's ability to maintain performance over a period time before it throttles. If the sustained performance is as improved as was mentioned in the keynote, that alone can be big and explain where a good portion of the 3nm improvements went toward.

 

[TL24]

Yeah, my 14 Pro is doing just fine.

 

[mprime_17]

I think M3 will be better than this 10%/7% boost because with A17 we are looking at the improvement over A16. M2 was likely based on A15, so I think a more reasonable assessment of M3 improvements would be to compare A17 to A15.

The big cores on the A16 improved its performance by just ramping up the clocks... like with the A17 now.
So don't get your hopes that high man.

 

[Ries]

Even with 35 trillion operations per second, the typing and correction for Danish is still worse than a Nokia 3310.

 

[Torty]

Thanks so much, and it is as “interesting” to me too. I haven’t gamed since the Atari in 1981.

So be prepared that the weather app will limit it’s animations for SOCs <A17

 

[chucker23n1]

I have difficulty wrapping my head around GeekBench telling me that the iPhone 15 Pro has faster single-thread performance than any Mac (the M2 Mac Studio gets 2803), or that the iPhone 15 Pro has almost as fast multi-thread performance as the iMac Pro?

Both Macs will sustain their level of performance for much longer.

But, the iMac Pro at this point isn’t very notable any more. And as for ARM Macs: yes, an M p-core is basically an A p-core, clocked higher. The M2 Mac Studio has the same p-cores as the A15 iPhone 13 Pro did. Just clocked higher.

 

[NT1440]

GeekBench is clearly not running for a long enough period of time?

The limited thermal/power envelope really should be representative in these numbers.

I have difficulty wrapping my head around GeekBench telling me that the iPhone 15 Pro has faster single-thread performance than any Mac (the M2 Mac Studio gets 2803), or that the iPhone 15 Pro has almost as fast multi-thread performance as the iMac Pro?

Why is that hard to grasp? Apple’s product segmentation relies on the TDP envelope for differentiation. iOS work tends to be more “bursty”.

 

[SFjohn]

I don’t know, after all the talk all year about this chip, perhaps I was expecting more?

And can some explain ray tracing to me?

Ray tracing (graphics) - Wikipedia

en.wikipedia.org

 

[theotherphil]

I've never understood why people get so hung up on Geekbench scores. Real world use is all that matters. I haven't owned a smart phone that lagged due to the gpu/processor in well over 7 years.

Because Geekbench highly correlates with the SPEC benchmarking (almost perfectly 1:1), which shows what to expect in real-world performance.

 

[goku_vegeta]

Very disappointing gains given the move to 3 nm. ~9% increase in performance core clock speed giving ~10% single core boost and 7% multicore boost. Plus (as has been commented on above), 20% more GPU cores giving a 20% GPU speed boost... Taking into account the rumours from the last year about how the A16 was supposed to have ray tracing but it was dropped late in development due to too much power consumption this is starting to look to me like the A17 is just the original A16 envisioned with the power savings from the 3 nm process sacrificed to the power hungry hardware ray-tracing GPUs. I wonder if battery life will be better than stated unless using ray tracing...

Lots of talk in the Apple Silicon forum about whether M3 is based on A16 or A17 - could it be that A16 and A17 are the same (with just ML cores, ray tracing and USB 3.0 in A17)?

In terms of the iPhone 16 SoC - I've read that the current 3 nm process is a stepping stone before the more scalable second gen 3 nm process and that the differences mean you can't just switch a chip from one process to the other, meaning Apple would have to redesign the chip anyway for the second gen process. Whether they chose to redesign the A17 Pro or just design two versions of the A18 will probably come down to a battle between the financial interests of saving money on the production line (i.e. using binned chips for the regular/pro phones) vs "encouraging" people to buy the Pro models through having a higher SoC generation than the non-Pro.

Battery life seems to be pretty similar according to Apple (their specific testing methodologies). In fact, the only difference between the 6.1" iPhone models is on audio playback where the 15 gets 5 additional hours of playback time. Otherwise the 14 Pro and 15 Pro stack up exactly the same. Real world testing would be interesting though.

 

[Mac_The_Ripper]

Because Geekbench highly correlates with the SPEC benchmarking (almost perfectly 1:1), which shows what to expect in real-world performance.

I don't need Geekbench scores to tell me that I'm not going to experience any lag with a flagship smartphone made by a reputable company though.

 

[sunny5]

That's a disappointing performance despite a new architecture + 3nm + higher clock speed. A15 to A16 also had that amount of improvements.

 

[auwkeung]

Barely any improvement on A16, can't comprehend why going to 3nm can have such mediocre performance without any battery life increase

Something has to give

 

[tripleh3lix]

Interesting..

 

[kkee]

Very disappointing gains given the move to 3 nm. ~9% increase in performance core clock speed giving ~10% single core boost and 7% multicore boost. Plus (as has been commented on above), 20% more GPU cores giving a 20% GPU speed boost... Taking into account the rumours from the last year about how the A16 was supposed to have ray tracing but it was dropped late in development due to too much power consumption this is starting to look to me like the A17 is just the original A16 envisioned with the power savings from the 3 nm process sacrificed to the power hungry hardware ray-tracing GPUs. I wonder if battery life will be better than stated unless using ray tracing...

Lots of talk in the Apple Silicon forum about whether M3 is based on A16 or A17 - could it be that A16 and A17 are the same (with just ML cores, ray tracing and USB 3.0 in A17)?

In terms of the iPhone 16 SoC - I've read that the current 3 nm process is a stepping stone before the more scalable second gen 3 nm process and that the differences mean you can't just switch a chip from one process to the other, meaning Apple would have to redesign the chip anyway for the second gen process. Whether they chose to redesign the A17 Pro or just design two versions of the A18 will probably come down to a battle between the financial interests of saving money on the production line (i.e. using binned chips for the regular/pro phones) vs "encouraging" people to buy the Pro models through having a higher SoC generation than the non-Pro.

I was expecting 90% increase in performance core clock speed giving 100% single core boost and 70% multicore boost... While at it, maybe it can fly to the space and feed cows at the same time.

Oh wait, no? yes, very disappointing indeed.