Question Incredible Apple M4 benchmarks...

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

gdansk

Platinum Member
Feb 8, 2011
2,510
3,423
136
None of the Intel chips people are comparing it to implement AMX anyway.
 

mikegg

Golden Member
Jan 30, 2010
1,815
445
136
The last thin tablets were actually the Tab s8/s9 series. They held the record this whole time primarily because iPads were still using LCD displays. The primary cause for the thinness this time was the use of a tandem OLED, it’s makes displays much thinner and the overall case.

Apple has been making their phones and laptops thicker recently and this made the battery removals much easier than the 2016 era MacBooks. I completely agree that the SSD should not be soldered.

Also Apple isn’t the sole pioneer, Dells XPS removed the headphone jacks in laptops while Apple kept it on their laptops.

Anyhow, this is going off topic. If you want to discuss Apples anti-consumer moves on their devices we should move it to the Apple Silicon thread.

Best to let this thread remain for M4 and Intel comparison discussion.
The thinness is not the most important thing.

The new iPad Pro weighs 15% less than the old one. That’s a huge advantage because unlike a laptop, you need to hold the iPad to use it often. And unlike a phone, iPads are much heavier so any reduction in weight is felt more.
 

Eug

Lifer
Mar 11, 2000
23,774
1,346
126
The thinness is not the most important thing.

The new iPad Pro weighs 15% less than the old one. That’s a huge advantage because unlike a laptop, you need to hold the iPad to use it often. And unlike a phone, iPads are much heavier so any reduction in weight is felt more.
The accompanying Magic Keyboard has also been reduced in weight, albeit by an unspecified amount.

One of the biggest complaints I had about the previous iPad Pro 11" when paired with the Magic Keyboard was that it was so much heavier than my 12" MacBook. In fact, the iPad Pro 11" (466 g) + Magic Keyboard (609 g) combined (1075 g / 2.4 lbs) was closer to the weight of the 13.6" MacBook Air (1240 g / 2.7 lbs) than to my 12" MacBook (920 g / 2.0 lbs). Here's hoping the new 11" iPad Pro + Magic Keyboard comes in at under 1000 g (2.2 lbs).

Anyhow, re: Apple chips, I'd be perfectly happy with a 12" MacBook with A18 Pro SoC. That would be >4X as fast as my current Core m3-7Y32.
 
  • Like
Reactions: Orfosaurio

mikegg

Golden Member
Jan 30, 2010
1,815
445
136
No, what annoys us is Apple's walled garden. If they would let us run Windows and our current games library with proper DX12/Vulkan drivers, why would we be angry?
That's a funny way to look at it.

You can look at it as Microsoft not open sourcing DirectX, keeping the AAA gaming monopoly on Windows. In other words, Windows is creating a gaming walled garden.
 

gdansk

Platinum Member
Feb 8, 2011
2,510
3,423
136
Not the best defence of Apple. Vulkan implementations do exist on Windows (not much due to Microsoft but they still do create extensions for it and even have their own wrapper targeting D3D12 upstreamed in Mesa).
There's no Vulkan implementation on MacOS except wrappers like MoltenVK (which Apple also didn't create).
So when it comes to open standards with open implementations, somehow Microsoft - the prime evil of old - is less closed than Jobs and Cook's Apple.

Not that it has much to do with their CPUs but they have a worse attitude than even Microsoft when it comes to anything open. This does show up with their CPUs too - they never talk about them at ISSCC or HotChips.
 
Last edited:

roger_k

Member
Sep 23, 2021
102
219
86
As @Nothingness said, SME is more like Intel's AMX which, as far as I know, isn't supported on client CPUs nor GB 6.3. AVX in its various form is more akin to SVE. They do support matrix math, but SME adds additional instructions/operations that let it accelerate certain matrix math much more when supported in the software. If you want to compare general IPC (or PPC), I would leave the SME enabled tests out of it.

SME is not only for matrix operations, it also includes a subset of HPC-oriented SVE instructions. Apples implementation uses 512-bit vectors, so comparisons to AVX-512 are appropriate as long as one stays within the HPC domain.
 
  • Like
Reactions: Orfosaurio

mikegg

Golden Member
Jan 30, 2010
1,815
445
136
Not the best defence of Apple. Vulkan implementations do exist on Windows (not much due to Microsoft but they still do create extensions for it and even have their own wrapper targeting D3D12 upstreamed in Mesa).
There's no Vulkan implementation on MacOS except wrappers like MoltenVK (which Apple also didn't create).
So when it comes to open standards with open implementations, somehow Microsoft - the prime evil of old - is less closed than Jobs and Cook's Apple.

Not that it has much to do with their CPUs but they have a worse attitude than even Microsoft when it comes to anything open. This does show up with their CPUs too - they never talk about them at ISSCC or HotChips.
Vulcan only exist because Nvidia and AMD support Vulcan. It's not because Microsoft is open to another API.

The difference is that Apple makes their own GPUs and only supports Metal.

Apple supporting Vulcan would not make a dent on AAA gaming. They have to make their platform enticing for AAA game developers. I believe their strategy is "port your AAA game over to Metal and you can run it on all of our devices including iPhone, iPad, Macs, Apple TV, Vision Pro, etc.".

The biggest piece of evidence for Microsoft creating a gaming wall garden is them buying up the biggest gaming studios.
 

gdansk

Platinum Member
Feb 8, 2011
2,510
3,423
136
Vulcan only exist because Nvidia and AMD support Vulcan. It's not because Microsoft is open to another API.
Yeah, it's not open to another API. But merges an implementation of Vulkan to mesa for everyone to use. Ok buddy.
 

poke01

Golden Member
Mar 8, 2022
1,573
1,876
106
This does show up with their CPUs too - they never talk about them at ISSCC or HotChips.
Intel and AMD do because they have to sell their chips. It’s also a great marketing tactic.

Apple isn’t a chip company. The bigger question is will Qualcomm present their X Elite arch since they do sell them to vendors and they have to promote it later this year.
 

mikegg

Golden Member
Jan 30, 2010
1,815
445
136
Yeah, it's not open to another API. But merges an implementation of Vulkan to mesa for everyone to use. Ok buddy.
Yes, to get Vulcan working on Windows Linux Subsystem, which does not have DirectX because it's not open sourced.

Meanwhile, Apple has created a DirectX to Metal kit for macOS.
 

gdansk

Platinum Member
Feb 8, 2011
2,510
3,423
136
Meanwhile, Apple has created a DirectX to Metal kit for macOS.
Which is based on and includes code from DXVK. Yet again Apple taking advantage of permissive license morons and not distributing their derivative source code (of course they have no legal obligation, but simple non-American ethics follows a take and give back model).
 

poke01

Golden Member
Mar 8, 2022
1,573
1,876
106
Yet again Apple taking advantage of permissive license 'tards and not distributing their derivative source code (of course they have no legal obligation, but simple non-American ethics follows a take and give back model).
As far I know it’s not consumer facing and meant for developers. If Apple presented it as a tool for consumers to use then it would be a different story but they presented that tool kit in a developer conference meant for developers only to test their games.
 

poke01

Golden Member
Mar 8, 2022
1,573
1,876
106
In the same way Amazon/Microsoft/Google aren't chip companies and yet they have talked about some of their chips before.
Apple’s never been one to share details. That may seem selfish to some but let’s be honest the level of detail they go thru at hot chips is nothing to the level of documentation companies withhold from presenting.

What would talking about your chips even do for Apple? It’s a dog and pony show for engineers and clients.
 

poke01

Golden Member
Mar 8, 2022
1,573
1,876
106
Which is based on and includes code from DXVK. Yet again Apple taking advantage of permissive license morons and not distributing their derivative source code (of course they have no legal obligation, but simple non-American ethics follows a take and give back model).

The source code is from crossweavers, the only closed part is the D3DMetal wrapper.
 

gdansk

Platinum Member
Feb 8, 2011
2,510
3,423
136
What would talking about your chips even do for Apple?
Unironically some engineers like doing it.
But otherwise not much. If you scroll back to my original post on the matter: Apple is more closed than pretty much all their peers in both hardware and software. And often to the detriment of consumers some of who then bend over backward defending Apple's behavior online thinking that somehow it gives Apple a competitive advantage. It doesn't by the way: they could hire the two main Asahi devs for $500K a year and no doubt there'd be more than 500 people worldwide every year who could then keep buying Apple laptops.
the only closed part is the D3DMetal wrapper.
And what was the only part under discussion and brought up as an example of Apple being open? D3DMetal... :rolleyes:
 

poke01

Golden Member
Mar 8, 2022
1,573
1,876
106
If you scroll back to my original post on the matter: Apple is more closed than pretty much all their peers in both hardware and software. And often to the detriment of consumers some of who then bend over backward defending Apple's behavior online thinking that somehow it gives Apple a competitive advantage. It doesn't by the way: they could hire the two main Asahi devs for $500K a year and no doubt there'd be more than 500 people worldwide every year who could then keep buying Apple laptops.
I do agree that Apple is very closed. Can they be more open? Absolutely.

It’s part of their business model. I would think they need to loosen a bit if want to succeed in the next decade.

Engineers who like to work with AI like publishing their research and won’t come to Apple if that’s not allowed. Over time I do hope that Apples culture changes but for nowthey are closed fisted as ever.
 
  • Like
Reactions: igor_kavinski

mikegg

Golden Member
Jan 30, 2010
1,815
445
136
Here is 12700T vs w5-3435X, Both based on Golden Cove, the latter having AMX: https://browser.geekbench.com/v6/cpu/compare/5330257?baseline=4656765
I didn't pick the fastest scores, but checked both are using the same GB version.

The effect of AMX can be seen on Object Detection. It also looks like Background Blur is affected.
So basically, GB6 has been using Intel's AMX all along. While Apple had AMX, it wasn't utilized by GB6 because you have to use Apple's CoreML library to even target it. CoreML will automatically run your code through NPU, AMX, or GPU so there was no way to guarantee it. Hence, it was left out of GB6.

This is just Apple/ARM having more feature parity with x86 instructions. x86 vs M4 scores for GB6 are valid then.
 
  • Like
Reactions: Orfosaurio

Nothingness

Platinum Member
Jul 3, 2013
2,838
1,560
136
So basically, GB6 has been using Intel's AMX all along. While Apple had AMX, it wasn't utilized by GB6 because you have to use Apple's CoreML library to even target it. CoreML will automatically run your code through NPU, AMX, or GPU so there was no way to guarantee it. Hence, it was left out of GB6.

This is just Apple/ARM having more feature parity with x86 instructions. x86 vs M4 scores for GB6 are valid then.
I don't know if the CoreML library makes use of AMX (I guess it could where NPU lacks support for data types or operations supported by AMX).

OTOH Apples Accelerate library makes use of AMX; see this for instance.

I'm not familiar enough with Apple dev to say if there is a lower level library to access AMX. Though I know for sure AMX was not available outside Apple libraries (which of course didn't prevent people from reverse-engineering it).
 

mikegg

Golden Member
Jan 30, 2010
1,815
445
136
I don't know if the CoreML library makes use of AMX (I guess it could where NPU lacks support for data types or operations supported by AMX).

OTOH Apples Accelerate library makes use of AMX; see this for instance.

I'm not familiar enough with Apple dev to say if there is a lower level library to access AMX. Though I know for sure AMX was not available outside Apple libraries (which of course didn't prevent people from reverse-engineering it).
I think Accelerate targets the AMX with more certainty. Meanwhile, CoreML can use the AMX if it thinks it's better than using the NPU or GPU.