M3 Chip Generation - Discussion Megathread

[Gudi]

I suspect there are two physical Max dies which share the same design. The smaller Max would be a chop of the full Max, with lower section of GPUs and two memory controllers removed. Apple previously used this for Pro/Max differentiation (with Pro being chopped Max), but M3 Pro has its own unique design. The M3 family must be very expensive to tapeout.

Interesting. You're absolutely right, with 6 instead of 4 E-cores, this year's M3 Pro chip can't possibly be a chop down of the M3 Max anymore. They solved the problem that Pro and Max chips had the exact same CPU performance by adding a third distinct layout. This way the Max chips can advance without repercussions for the vast amounts of M3 Pro Macs. The high-end customers paying $7,199 for a top of the line MBP can feel lucky and the regular Pro customers as well. They benefit mostly from the amazing battery life of a not too shabby 12-core (6P/6E) CPU.

 

[Pressure]

Geekerwan have released their M3 and M3 Max review.

This time they even measure core cluster to core cluster latency, they have really picked up the baton from Anandtech and also used SPEC CPU 2017.

This is only review you really need to watch as they include everything and the kitchen sink!

SPEC CPU 2017, Blender CPU/GPU, Cinebench 2024 CPU/GPU, Cinebench R23, Cinebench, 7-zip, DaVinci Resolve, Adobe Premiere Pro, Adobe After Effect, Adobe Media Encoder, Xcode, 3DMark, Stable Diffusion and games.

Well, they miss some Photoshop tests but I am sure ArtIsRight is ready to deliver that as soon as he receives his units 👍🏼

 

[leman]

Geekerwan have released their M3 and M3 Max review.

This time they even measure core cluster to core cluster latency, they have really picked up the baton from Anandtech and also used SPEC CPU 2017.

This is only review you really need to watch as they include everything and the kitchen sink!

SPEC CPU 2017, Blender CPU/GPU, Cinebench 2024 CPU/GPU, Cinebench R23, Cinebench, 7-zip, DaVinci Resolve, Adobe Premiere Pro, Adobe After Effect, Adobe Media Encoder, Xcode, 3DMark, Stable Diffusion and games.

Well, they miss some Photoshop tests but I am sure ArtIsRight is ready to deliver that as soon as he receives his units 👍🏼

Very thorough review, although they have massively butchered the explanation of Dynamic Caching...

Anyway, the M3 Max RT performance in Cibenebch 2024 is comparable to the desktop RTX 4060 or laptop RTX 4070... not too shabby for Apple's first-gen RT implementation. I was expecting improvement of 50-100%, and they seem to have landed squarely in between. Looking forward to Blender, which should be better optimised.

 

[Chancha]

Tech Chap's video has Blender (matching M2 Ultra).

And he shares some more insight; the thermal behaviour is pretty interesting this time, I wonder how much it has to do with the darker chassis.

Then he managed to get a game to drop performance when the MBP is unplugged.

Both his 14" and 16" are decked out, the 14" falls behind a bit as expected but can match the 16" pretty close in most cases.

 

[diamond.g]

Tech Chap's video has Blender (matching M2 Ultra).

And he shares some more insight; the thermal behaviour is pretty interesting this time, I wonder how much it has to do with the darker chassis.

Then he managed to get a game to drop performance when the MBP is unplugged.

Both his 14" and 16" are decked out, the 14" falls behind a bit as expected but can match the 16" pretty close in most cases.

That is a Total War bug that appears to not have been fixed.

 

[Xiao_Xi]

Dave2D tested M3 Max in Blender 4.0

 

[MayaUser]

Both his 14" and 16" are decked out, the 14" falls behind a bit as expected but can match the 16" pretty close in most cases.

Thats impressive...for the 14" with the M3 Max. Based on the M2 max how 14" was working...i was expecting a bigger fall, but no

 

[yellowhelicopter]

What about Mac Mini M3? Next year I suppose?

 

[MayaUser]

What about Mac Mini M3? Next year I suppose?

for sure, M3 M3 Pro mac mini and probably the M3 Macbook airs

 

[Chancha]

That is a Total War bug that appears to not have been fixed.

Even then I find it interesting; in what way does an app know if the MBP is on battery, and thus deliver less performance? Even if it is unintended?

 

[leman]

Even then I find it interesting; in what way does an app know if the MBP is on battery, and thus deliver less performance? Even if it is unintended?

That’s something software can query from the OS. E.g. if I detect that the laptop is running on battery I might limit the frame rate in my game to conserve power.

 

[diamond.g]

Even then I find it interesting; in what way does an app know if the MBP is on battery, and thus deliver less performance? Even if it is unintended?

That’s something software can query from the OS. E.g. if I detect that the laptop is running on battery I might limit the frame rate in my game to conserve power.

Yeah I think MrMacRight talked to the developer (or porting house) about this and they may have mentioned working on making it a toggle setting. Which I guess hasn't happened yet.

 

[Homy]

The highest score! Can't find any scores breaking 1600 for RTX 4090.

 

[streetfunk]

How important is the RAM on these graphic tasks ?
Whats the minimum on a 16core Max that the RAM won´t throttle the rest ?
simple stuff. single tasks, no long winded bulk work

 

[jeanlain]

The highest score! Can't find any scores breaking 1600 for RTX 4090.

I wonder how After Effects uses the GPU. The M3 Max is a good laptop GPU, to be clear, but there's no way it can approach a desktop 4090 in any performance metric.
Which is normal. The 4090 is rated at 400W or so.

 

[Homy]

I wonder how After Effects uses the GPU. The M3 Max is a good laptop GPU, to be clear, but there's no way it can approach a desktop 4090 in any performance metric.
Which is normal. The 4090 is rated at 400W or so.

I've heard this a few times but apparantly it's faster for whatever reason in this test. 4090 being faster in other cases is not relevant when we talk about this test. So if you want the fastes system for After Effects M3 Max is the one you should buy.

 

[dgdosen]

Would the performance of the M3 family be significantly better if Apple opted for LPDDR5X memory over LPDDR5?

 

[Confused-User]

Would the performance of the M3 family be significantly better if Apple opted for LPDDR5X memory over LPDDR5?

I am assuming you meant LPDDR5X at a much higher speed than the LPDDR5 in the M3 Macs.

The obvious answer is "maybe, for situations where the application is bandwidth-limited". But the question is, what are those applications? And how much does bandwidth influence them? Those questions are way too general for a specific answer. You'd have to ask about a particular application (or benchmark).

For non-GPU-using apps, there will be very few cases where you'd see any difference at all. Depending on the specific details of the design of the 6-core clusters in the Max and Pro - specifically, the bandwidth from the cluster to the SLC - "very few" might actually be "exactly zero".

For GPU-heavy apps... it could make a big difference, but probably won't. Too vague? Sorry, best you can get for a very vague question.

 

[senttoschool]

I wonder how After Effects uses the GPU. The M3 Max is a good laptop GPU, to be clear, but there's no way it can approach a desktop 4090 in any performance metric.
Which is normal. The 4090 is rated at 400W or so.

Wasn’t the M1 Max fastest at certain Divinci Resolve tasks because of its unified memory? Could be the case here as well.

Edit: it was Affinity.

https://twitter.com/i/web/status/1452623920721448963

 

[Chancha]

Yeah I think MrMacRight talked to the developer (or porting house) about this and they may have mentioned working on making it a toggle setting. Which I guess hasn't happened yet.

I know there must be within API that tells apps system-wide that a certain hardware state the Mac is in, particularly as important as power source. While with Intel MacBooks, it was necessary especially concerning the iGPU dGPU switching.

What I find interesting is the fact that the Apple Silicon MacBooks are already proven to be able run at the same clock plugged vs on battery. So an app choosing to toggle itself down to a less demanding state is leaving performance on the table for no reasons? The Total War devs say it is a bug, this sounds like just left over behaviour from when dealing with Intel MacBooks?

 

[Confused-User]

What I find interesting is the fact that the Apple Silicon MacBooks are already proven to be able run at the same clock plugged vs on battery. So an app choosing to toggle itself down to a less demanding state is leaving performance on the table for no reasons? The Total War devs say it is a bug, this sounds like just left over behaviour from when dealing with Intel MacBooks?

It's not really "no reason".

If you're doing work, you have to get the work done, and usually race-to-idle is the right choice- though possibly not, if getting the last 5% clocks pushes power consumption up a lot. (It seems that on Mx Macs, that's probably not the case - see leman's thread about A17 core power usage.)

But...

If you're playing games, the amount of workload is literally unbounded (or, perhaps, bounded by your max display frame rate, 120Hz on these Macs). But unlike work-work you don't actually need to do all of it. You can sacrifice some of it, at a (possibly imperceptible and likely minor) cost of user experience quality, in favor of preserving battery life. And if you're actually on battery, that may be a tradeoff well worth making. Even on Macs.

 

[Chancha]

It's not really "no reason".

If you're doing work, you have to get the work done, and usually race-to-idle is the right choice- though possibly not, if getting the last 5% clocks pushes power consumption up a lot. (It seems that on Mx Macs, that's probably not the case - see leman's thread about A17 core power usage.)

But...

If you're playing games, the amount of workload is literally unbounded (or, perhaps, bounded by your max display frame rate, 120Hz on these Macs). But unlike work-work you don't actually need to do all of it. You can sacrifice some of it, at a (possibly imperceptible and likely minor) cost of user experience quality, in favor of preserving battery life. And if you're actually on battery, that may be a tradeoff well worth making. Even on Macs.

And by extension, it seems the "high power mode" and to some degree the "game mode" are ways for the users to intervene or limit this system behaviour then?

 

[Confused-User]

And by extension, it seems the "high power mode" and to some degree the "game mode" are ways for the users to intervene or limit this system behaviour then?

Yes. But the thing originally being discussed was the behavior of a certain game. That's a specific implementation detail of that particular app, and there's no reason (so far) to think this has anything to do with Apple's code.

 

[Chancha]

Yes. But the thing originally being discussed was the behavior of a certain game. That's a specific implementation detail of that particular app, and there's no reason (so far) to think this has anything to do with Apple's code.

This certainly raised some alarms, as it's being mentioned in a hands-on video with no context, or the reviewer did not investigate. It felt like a new macOS or 14" 16" bug, at least to me.

 

[diamond.g]

This certainly raised some alarms, as it's being mentioned in a hands-on video with no context, or the reviewer did not investigate. It felt like a new macOS or 14" 16" bug, at least to me.

The game does the same thing on Windows laptops, so it is safe to assume it is a carry over feature that Feral didn't remove.