Hacker News new | past | comments | ask | show | jobs | submit login
First Benchmarks Surface for Apple's ARM-Based Developer Transition Kit (9to5mac.com)
45 points by rbanffy 81 days ago | hide | past | favorite | 41 comments



It's also benchmarks of a slightly slowed down version of a chip intended for an iPad Pro, assembled to support development of porting to a new processor. As they stated in the keynote (or one of the sessions) it's not indicative of an actual Mac chip performance. So it's interesting, but not all that revealing.


For comparison, The Intel DTK was a P4, with an Intel GMA900 GPU, shoved in a powermac chassis. It wasn't indicative of the performance of the core duo & xeons they in the first-gen Intel Macs.

The DTK is a minimum viable platform, not a product.

(That's not to say I'm not hellish curious about some Rosetta2 examples that aren't cherry-picked!)


It's at least interesting that it's close-ish to current entry-level MacBook Air performance, even with Geekbench running under Rosetta. Seems like that would indicate that the real chips they'll be using in laptops and desktops will be pretty performant.


It's curious that it's possibly downclocked from the latest iPad Pro, although only a very little bit and that may not even be accurate reporting by GeekBench running in emulation. But of course the Intel mac mini in the same box has the famous pretty large fan. And the DTK still has the large vent opening on the back. So there is probably the same fan assembly in there — a first for an A_ series device.


It's entirely possible the opening is there because they didn't want to build a completely new back panel


The port layout is a little bit different (there are fewer USB-C shaped ports and HDMI is moved over). So it's different from the for-sale Intel mac mini anyway. They can certainly use the large aluminum housing from the current 2018 model mini. The plastic back-plate is pretty small anyway.


> Interestingly, the Geekbench submissions report that the Developer Transition Kit (which resembles a Mac mini) is a four-core machine. The A12Z chip actually has eight cores, four high performance and four low-power efficiency cores.

My guess as to why this is, is that Apply locks Rosetta to run only on the high performance cores.


Apple likely will have a big incentive to try to get the perf of their processors up to comparable levels with Intel as soon as possible. They will compare native ARM64 apps like Photoshop, chrome, etc to the same apps on Intel chips, but I wonder what will be their marketing / explanation for the difference running apps using Rosetta -- like if a particular app that is quite difficult to transition to ARM64 runs quite poorly, do they just say "well its a really hard problem, so please give us some credit for it?" Like it would seem like Apple would want to figure out a way to make the translation code be multicore, so you could take something that thinks its running on a single core and this translation layer would split it up to multicore and then Apple would just add more cores to make the translation layer fast enough. Anybody have any clue if thats possible?


I think they need to do this as there is no support for asymmetric CPU cores on Intel or macOS prior to Apple Silicon.


The wwdc session on the silicon says they are symmetric (have the same instruction set) unlike the usual big.LITTLE implementation.

That being said presumably Rosetta 2 will be pegged to the faster cores (your app can do hunting as to what sort of load it might create/need).


That's decidedly impressive considering it's non native code too, and shows how much further the art has moved.


Do we have numbers to compare with the Snapdragon-based Windows laptops?

edit: we do: https://www.notebookcheck.net/Qualcomm-s-Snapdragon-8cx-amas...


These are Geekbench 4 results which are not comparable to Geekbench 5


You can still compare Geekbench 4 scores with x86 machines running the same version. This gives an idea of how performant a Snapdragon is emulating an x86.

With that, we have something to compare the A12 with.


Interesting to see the differences in sub-task scores between this[1] and a regular iPhone 11[2]. Though it's hard to tell what might be memory bandwidth and what might be translation overhead.

[1]https://browser.geekbench.com/v5/cpu/2732510

[2]https://browser.geekbench.com/v5/cpu/1000620


Does the DTK support running iOS apps? If so you could, in theory, run the iOS version of Geekbench to get performance numbers outside of Rosetta.


Or you could simply look at the geekbench numbers for the iPad Pro, which uses the same SOC.


Since the power and thermal envelope is a lot higher, it's probably noticeably faster than the iPad Pro.


That is a promised feature, but for existing published iOS apps it seems like it would rely on the Mac App Store in Big Sur providing iOS apps, and that switch may not have been flipped yet.


I also doubt that this is the hardware that will ship to customers. Customer hardware will be 50-100% faster at least.


geekbench is no where near a good measure of performance. Give the phoronix test suite a run and then let's talk.


You want to examine the individual subscores to make sure it's not just hardware acceleration but it's generally as good a measure as any other, at least with version 5. There were previous versions that had different working set sizes for different platforms and that was a major source of error.

EDIT: In the end the only really good benchmark is your particular workload run on the processor. Everything else is just evidence towards what that'll be.


geekbench is exactly what people refer to when they claim the iPad Pro is just as fast as a MacBook: https://www.macrumors.com/2020/05/12/ipad-pro-vs-macbook-air....


It also does well enough on Spec to make that plausible. Though the processor on a MacBook is has a clockrate much slower than the design sweet spot. If Intel had targeted 1.1 GHz operation for their Core processors they'd have much shorter pipelines and cache latencies.

https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x...


I wonder why apple is so persistent with using that awkward "apple silicon" wording more importantly why all blogs and journalists just copy it. Whats wrong with saying arm macs, apple chips, apple CPUs etc? Is anyone alse feeling weird about it?


That is what Apple’s chips with ARM instruction set are called. You don’t call AMD chips, Intel chips because they use intel instruction set.

Likewise Apple’s CPUs are not designed by ARM. They are Apple designs licensing ARM instruction set


I think they want to highlight the custom silicon models that they have so far because we are now at the point that probably every product will have its own processor so you need some kind brand of what the goals are of Apple's strategy in the future.

Also, its really good marketing and allows everyone to have the same talking points because if you want to explain the naming scheme of any processor manufacturer it will take longer than the amount of time that anyone has an actually attention span for people not interested in the subject already.


It think they use the naming because their message is that they don’t slap a few random parts together, but optimize a complete design.

Certainly, for many workloads, what matters isn’t the performance of the CPU but how efficiently it can offload those tasks to the GPU and ‘Neural Engine’. Those two parts also might use up the majority of the transistors.

The CPU also _could_ have a few un-ARM-like extensions that help in emulating x86. It may be interesting to watch what gets committed to LLVM.


For the same reason people dont say x86 Macs. And it is not a CPU so you cant say Apple CPU. If you say Apple SoC not everyone knows what is a SoC. Apple Silicon is much easier to understand for vast majority of consumers, and the name includes everything fro CPU, GPU, NPU, Video Encoder. etc.

You may ask what happen when Wafer move away from Silicon.


Apple Silicon is the brand of the chip, like Intel Xeon.


More like “Intel Inside”, I would think. It separates their hardware from that of other companies, not from future versions of their hardware.

I don’t see them moving to new names or to “Apple Silicon v2” when hardware advances.


For the same reason they wouldn't call the iPod an "Apple MP3 player" or the iPhone "Apple cell phone"— they're reserving the space for their branding which is not yet solidified.


Pure speculation, but when they switched to "intel" they were effectively locked in, and they couldn't then have some intel and some AMD and some of their own x86 stuff.

Now they're going to "Apple Silicon", it might be ARM this decade and who-knows-what next decade, but it will remain as reliable and fast and blah-blah as "Apple Silicon".


This begs the question, if Geekbench developers have the DTK? Since this is not iOS there is nothing that prevent them from releasing an update now on ARM platform.


I believe the NDA forbids benchmarking. That doesn't mean it's not going to happen, but it might hurt Geekbench's case for releasing it. Or even receiving a dev unit.


Just a note that the SoC in that mini is the same as the now two, going on three year old iPad Pro.


The A12x came out in fall of 2018, so coming up on 2 years


> Note that these tests are running under virtualization, using Apple’s Rosetta technology

totally unreliable info, but interesting.


> totally unreliable info, but interesting.

How is this "unreliable"?

Native benchmarks wouldn't be very interesting, since this is the same processor we see in the iPad Pro. What's interesting is how Apple will handle the transition from Intel to ARM, so measuring Rosetta's virtualization capabilities is absolutely what we should be focused on until we actually get access to genuine Apple Silicon desktop chips.


> Native benchmarks wouldn't be very interesting, since this is the same processor we see in the iPad Pro.

Not necessarily. They might be interesting. The processor might have active cooling and be less thermally limited in the Mac Mini form factor. In which case Apple might have clocked it slightly higher.


That's a good point that I hadn't previously considered. Thanks for bringing it up! You're absolutely right!




Applications are open for YC Winter 2021

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: