Rosetta 2 Benchmarks Surface From Mac Mini With A12Z Chip

[jlocker]

I'd say that it's at least possible that these benchmarks provide a close approximation of how non-native apps will perform (Running in Rosetta)?

If nothing else, this should add to the pressure from customers on developers to deliver native apps asap? I know two programs that I use daily that were not likely to be updated in a timely matter, based on the transition from PPC to Intel, so I welcome any additional pressure for them to update?

I'm not worried, as I've already budgeted the purchase of at least one "backup" Intel system to provide the best possible performance with those important apps that I doubt will make the transition to native in a timely matter, if at all?

I'm just sitting back and enjoying the show at this point?

I agree if you buy a MacBook Pro today and get Applecare you use it for three years and then you take the old model to the Apple store and you turn it in for the model with the new Apple processor and we are at Apple 11.3 with all the bugs and problems worked out and more native applications.

 

[mdriftmeyer]

Someone posted his own benchmarking results for iPad a few posts earlier. His observation was that the small core performance is 10% of the big core. If this is correct, adding 4 small cores would add 10% performance at best. The impact of Rosetta is hard to estimate.

It's not a linear summing relationship in performance, by adding more cores.

 

[mr_jomo]

So that's pretty cool - even through a translation layer the A12Z manages to outperform the 2020 i3 MBA and approach the i5 version!

On beta OS, a DTK and 6 months before launch of the optimised OS and actual silicon.

This bodes well on all parameters.

 

[mdriftmeyer]

Apple hasn't confirmed anything.

Currently Apple sells Macs with 2, 4, 6, 8, and 12 to 28 cores. Common sense is that there won't be _one_ ARM chip for all Macs. Common sense is that the chips going into the lowest end Macs (replacing two cores) will be whatever mobile chip Apple has in six months time, clocked to the maximum that the chip can handle. That will be a huge improvement for dual core Macs even with Rosetta. Take these benchmarks, add 5% for an improved chip, 5% improvement in Rosetta, and 40% improvement due to running at 3.5 GHz, and another 50% for running ARM code.

Common sense is also that Apple will package two or four of these chips into one package, at which point we will have a huge improvement for all the Macs with four to eight cores. We will have a huge improvement running x86 codes through Rosetta, and native code will fly.

Apple will have to have a back plane controller ala AMD Infinity Architecture and/or interposer with a chiplet design [already patented to the nines by AMD and to a lesser extend Nvidia and Intel] to deliver on your hope. I expect they will have their own implementation and that's what patents are for--novel solutions to an obvious problem that isn't already implemented, or they will have to license from the big three just mentioned--and what is the motivation from them to do so?
[automerge]1593457324[/automerge]

Is your 9-year-old PC running pre-release software on an underclocked tablet CPU under emulation?

Of course not, but their 9 year old CPU has hundreds of processes and nearly two thousand threads running at all times for a baseline OS whether it's Linux, OS X or Windows.

 

[polaris20]

1) down-clocked slower than iPad Pro!
2) Running benchmark in rosetta
3) Only using 4 out of 8 cores for some reason
4) not the chip that will be used in macs

These benchmarks mean absolutely nothing.

If anything, I'm impressed. For it to be still running that well whilst being converted, that's awesome. They should run the iOS version on there, since it should be compatible with this new Mac.

 

[Jamie0003]

Apple needs to do better than this for desk top performance. I am a little scared now.

EDIT: I retract this statement due to all the negative reactions. But, Apple still needs to do better than this for desk top performance. I am not scared though.

Did you actually read the comments?...

 

[MacCheetah3]

Apple needs to do better than this for desk top performance. I am a little scared now.

EDIT: I retract this statement due to all the negative reactions. But, Apple still needs to do better than this for desk top performance. I am not scared though.

Really underwhelming results, makes me wonder if it was the right time for Apple to do this, or maybe they should have waited a few more years for the silicon team to catch up to Intel, or maybe they should have just gone with AMD.

Edit: I retract what I said, clearly I didn't know enough to make a comment. Thanks everyone who corrected me.

You're allowed to state an opinion, however, I don't know why you're disappointed. Apple decided to quickly create a tool by just swapping in their most recent, publicly available tablet SoC, which as already mentioned probably lacks a lot of software-level optimization when paired with macOS.

Looks like these are only running on 4 cores, probably because apple didn’t bother to teach the thread scheduler how to optimize for a12z which will never be in any real product.

And also as already surmised...

Way better than I expected. Already scoring broadly in line with the scores for a 2018 Mac mini with a 4 core i3 at 3.6 GHz, and as lots of other people have said, this isn't a desktop chip, it's drawing less power, runs at a much lower clock speed, has half it's cores turned off and is running emulated code.

Despite these limitations/restrictions, plus cooling and possibly others, it appears the DTK is on par with the 3.6GHz quad-core Mac mini. So, it has essentially already matched an offered desktop performance -- let's not nitpick "desktop." Additionally, if you compare these two CPUs on a very simplistic scale, such as clock rate, the A12[X/Z] offers 50% more performance than Intel, 2.4GHz vs 3.6GHz with the same general processing score. Again, I don't understand how that's something to scoff at.

 

[Taz Mangus]

I am curious if the Apple DTK has an active fan in it. It being down clocked and the iPad Pro not having a fan.

 

[Madhatter32]

Did you actually read the comments?...

Yup. How could I not? It's all across my notifications.

 

[Macbookprodude]

We all knew that this would happen. However, we also know that the real Apple Silicon Macs will use a completely different chip, no doubt modified and optimised in ways we don't know about yet. While this is interesting (and I'll read all the news articles that come up about this), its going to tell us next to nothing about what's coming.
[/QUOTE

Just call it PPC V2 on steroids.

 

[dmccloud]

This is a developer kit for developing apps - it is NOT a production unit. These benchmarks mean nothing, especially given that the actual production processor hasn't been revealed yet.
[automerge]1593460867[/automerge]

The move away from Intel is a solution in search of a problem to solve.

You mean besides the recurring yield issues with Intel CPUs, deep-seated security issues with multiple Intel products, and cost?

 

[Gerdi]

Wow, if these benchmarks are legit - they are totally impressive!

This means it outperforms the latest Intel SoC Lakefield, which is also rated 7W TDP - and this is under emulation!!!
It also means it is on the same level as Cortex A76 from ARM (aka Microsoft SQ1 7W TDP) running native ARM code - again under emulation!

Quick calculation shows us, that it is more than 60% native speed. (830/1300)

 

[JPack]

I can understand not using the "efficiency" versus the "performance" cores, but why would Apple underclock them in a Mini compared to an iPad Pro? There's more room in the Mini enclosure for air flow so there ought not be a thermal reason and it's always plugged in to a wall socket so saving power ought not be an issue either.

Power delivery reasons.

With sufficient cooling, the processor will ramp to higher clock speeds which exceed the power delivery design of the logic board. Anandtech found the A12 crashed whenever they tried to run 3DMark on a cool iPhone Xs.
[automerge]1593461755[/automerge]

I am curious if the Apple DMK has an active fan in it. It being down clocked and the iPad Pro not having a fan.

Fan most likely exists. Keep in mind the iPhone and iPad quickly thermally throttle under load.

 

[mannyvel]

Fan most likely exists. Keep in mind the iPhone and iPad quickly thermally throttle under load.

I've never heard of an iPad or iPhone thermally throttling. iPhones will shut down if the battery can't handle it.

Do you have anyplace else that discusses this?

 

[JPack]

I've never heard of an iPad or iPhone thermally throttling. iPhones will shut down if the battery can't handle it.

Do you have anyplace else that discusses this?

Look up sustained vs. peak performance for the iPhone. Anandtech explores this in detail for the previous several iPhones.

The SoC conumes about 6W under heavy load. There is no way to passively cool this without a heatsink and without throttling.

 

[dogslobber]

Then the developer is responsible for having let an unauthorized third party have access to the dev kit machine. Also against the agreement they signed.

I don't think many people would refer to their mom as an "unauthorized third party".

 

[magbarn]

I've never heard of an iPad or iPhone thermally throttling. iPhones will shut down if the battery can't handle it.

Do you have anyplace else that discusses this?

From Anandtech's 2018 iPad Pro 11 review:

"Sustained Results
As hinted to above, Apple offers significantly more peak performance from the GPU than sustained, which can be beneficial for computational GPU in the OS for various functions. But for gaming, sustained performance is important.

GFXBench offers a battery rundown mode where Manhattan 3.1 is looped consecutively. For a comparative basis, if you run Manhattan 3.1 on the iPad the result is 100.79 frames per second on a single run, which is 6249 frames.

​

The longer test starts out high, but then decreases in performance significantly in the first couple of runs, and then starts to taper off near the end. The lowest framerate in this roughly 30-minute test is 3960.9 frames, which works out to 63.9 frames per second. That’s about a 36% decrease in performance over time."

Apple is going to either need to use a very large heatsink or more likely use active cooling unless they want to artificially cap performance with games so you don't get the drop in fps as the heat builds up.

 

[CarlJ]

I don't think many people would refer to their mom as an "unauthorized third party".

And yet, if you signed an agreement saying that only you would have access to the machine, that’s what it’ll reasonably look like to Apple. And their lawyers.

 

[dogslobber]

And yet, if you signed an agreement saying that only you would have access to the machine, that’s what it’ll reasonably look like to Apple. And their lawyers.

These signed agreements are worthless.

 

[andersenep]

These early numbers seem pretty promising for Rosetta 2 imo. I am curious how the ios version of geek bench (arm native) would match these chips to their iPad pro counterparts. Can't big sur run ios (ipados/whatever...)apps natively?

 

[buchhla]

Yea, about the same proformance as I get on my 2015 macbook pro 15", so not looking very good here! It will be very interesting to see what they really come up with with final silicon and native performance. And we better be getting a 2x inscrease is actual usable battery life for the more than likely performance trade-off...

 

[fafner]

That is quite impressive, the thing that many people are forgetting is that the A13's eight-core Neural Engine with 5TOPS, an A13 is about 1000% faster than a Nvidia jettson (Pascal). And that the performance of co-processors built into the A-series chip will make a huge performance boost for a lot of tasks.

so how much neural engine performance can they integrate in a desktop chip?

EdgeAI Performance Test on Apple A12, A13 and NVIDIA Jetson TX2

Summary

medium.com

 

[gnasher729]

It's not a linear summing relationship in performance, by adding more cores.

That was a measurement I made - on an iPhone XR with 2 + 4 cores, running a benchmark using 1, 2, 3, 4, 5 and 6 threads. The amount of work done in a fixed time was quite exactly 100, 200, 210, 220, 230 and 240 work units. Interestingly, GCD seems to move more than two threads between fast and slow cores, so six threads given the exact same amount of work would finish quite exactly at the same time. With 4 + 4 cores, I'd expect the work done by 1 to 8 threads in the same time to be 100, 200, 300, 400, 410, 420, 430 and 440 work units.

 

[brucebrendon]

Imagine if there is no new hardware, but you can run BigSur off an iPad/iPhone Pro... with a keyboard, display, eGPU etc – finally a reason for the silly iPro name? #BYOKM2

 

[Tech198]

it may say "Developer Transition Kit" on the box, but it still looks like a hacked Mac mini to me. (by Apple this time)

Imagine if there is no new hardware, but you can run BigSur off an iPad/iPhone Pro... with a keyboard, display, eGPU etc – finally a reason for the silly iPro name? #BYOKM2

Apple would be fooling us all