M3 Chip Generation - Discussion Megathread

[Homy]

Comparing M3 Max to M2 Max on Geekbench 6 OpenCL test:

My M2 Max 38 Core = 87,992
Best M3 Max Geekbench 6 Result = 93,627

6.4% faster?

Who uses OpenCL instead of Metal and why? Just curious.

 

[Confused-User]

I see. There is a possibility that mass production of N3P will be completed in time for A18 in 2024.

N3E mass production is targeted for H2 2023, but actually it will be in 2024, so I thought N3P mass production was targeted for H2 2024, so it would actually be in 2025.

I would also like to hear an answer from Confused-user as to whether he considers N3E to be a node only for iPhones and iPads, but I'm glad to hear an interesting answer from tenthousandthings.

While I'm not the final authority on any of this... I believe I already answered. N3E will be used very widely by many companies for many years. As I and others posted previously, it's the last TSMC FinFET node, and as such it's going to be around for MANY years.

N3B, on the other hand, will be dead the minute Apple stops ordering M3 and A17 chips.

But maybe you were trying to ask specifically if *Apple* will be using it for anything besides iPhones and iPads? If so... hard to say. Apple will very likely switch to N3P for M4... IF it's available in time. I don't have a good sense of whether that's likely or not.

 

[Homy]

Some results for Pugetbench Photoshop for base M3. I know there are higher scores for similar PC systems but it's funny to see a small M3 laptop perform as good as powerful PC laptops or desktops. Take a look at the GPU scores.

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

 

[sv8]

While I'm not the final authority on any of this... I believe I already answered. N3E will be used very widely by many companies for many years. As I and others posted previously, it's the last TSMC FinFET node, and as such it's going to be around for MANY years.

N3B, on the other hand, will be dead the minute Apple stops ordering M3 and A17 chips.

But maybe you were trying to ask specifically if *Apple* will be using it for anything besides iPhones and iPads? If so... hard to say. Apple will very likely switch to N3P for M4... IF it's available in time. I don't have a good sense of whether that's likely or not.

Thank you, Confused-user.

My questions are not very advanced, but even during more advanced discussions, your opinions were reasonable and very meaningful and enjoyable.

Now, I can understand the floor plan of Max from either point of view: Will N3B Ultra/Extreme be created by connecting 2/4 Max, or will N3E or N3P Ultra/Extreme be created with a different design from Max, it seems possible.

 

[altaic]

Some results for Pugetbench Photoshop for base M3. I know there are higher scores for similar PC systems but it's funny to see a small M3 laptop perform as good as powerful PC laptops or desktops. Take a look at the GPU scores.

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

View attachment 2305941

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

View attachment 2305942

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

View attachment 2305943

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

View attachment 2305944

I’m not very familiar with that benchmark, but that looks pretty impressive.

 

[anshuvorty]

Some results for Pugetbench Photoshop for base M3. I know there are higher scores for similar PC systems but it's funny to see a small M3 laptop perform as good as powerful PC laptops or desktops. Take a look at the GPU scores.

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

View attachment 2305941

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

View attachment 2305942

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

View attachment 2305943

PugetBench Results

Build your own computer, workstation, mini pc, or server at Puget Custom Computers.

benchmarks.pugetsystems.com

View attachment 2305944

The power and magic when you own the entire vertical stack on full display....

 

[scottrichardson]

Who uses OpenCL instead of Metal and why? Just curious.

Only did that because there was only OpenCL benchmarks online for the M3 max. No metal scores at the time I did that. Of course I’d use Metal normally.

 

[leman]

I have a more long-range issue with the Pro and Max.

I think that aligning CPU and GPU resources is, in the long run, a mug's game, and that Apple may be pricing itself out of a lot of business. Even business willing to pay Apple's high margins for resources they actually need, because they're unwilling to pay those high margins for resources they really don't need.

I agree. In fact, Apple lost some money on me today. For my work, CPU is more important than the GPU, and I cannot justify the $500 markup just of the two additional P-cores, even though I'd really liked to have them. So in the end I am getting the 10+4/30 core model, which is a shame. I would have even paid $100 extra for the two P-cores.

 

[Chancha]

I agree. In fact, Apple lost some money on me today. For my work, CPU is more important than the GPU, and I cannot justify the $500 markup just of the two additional P-cores, even though I'd really liked to have them. So in the end I am getting the 10+4/30 core model, which is a shame. I would have even paid $100 extra for the two P-cores.

But that $500 also gives you 12GB more RAM plus the additional memory channels, and then 33% more GPU cores.

How much difference does this make in real world we have to see, but in terms of % out of the already premium price of the 14/30, I think there is value.

The problem I have with investing in a laptop is the knowledge of a Mac Studio having the same config is months away, Apple just doesn't want to sell it now since the Ultra isn't ready (or is it?).

 

[altaic]

Who uses OpenCL instead of Metal and why? Just curious.

There are a lot of open source projects that use OpenCL because they don’t have the resources to support a multitude of compute paradigms.

 

[JouniS]

I think that aligning CPU and GPU resources is, in the long run, a mug's game, and that Apple may be pricing itself out of a lot of business. Even business willing to pay Apple's high margins for resources they actually need, because they're unwilling to pay those high margins for resources they really don't need.

This issue is already obvious today.

The issue was already obvious when the M1 was released, because it's a trivial consequence of tighter integration. As the hardware becomes less modular, you can offer fewer different configurations cost-effectively. Integration is great when your needs are common, because then you get better hardware for the same price. But the further away your needs are from the average, the more you have to pay for unnecessary hardware.

 

[leman]

But that $500 also gives you 12GB more RAM plus the additional memory channels, and then 33% more GPU cores.

How much difference does this make in real world we have to see, but in terms of % out of the already premium price of the 14/30, I think there is value.

That's the point. Of course there is value. But it's not value that I can defend in front of my finance director since I don't really need those extra GPU cores or RAM bandwidth.

 

[Macintosh IIcx]

I’m extremely pleased with the look of the M3 Max. Have been wanting to move from my MacBook M1 Max to a Mac Studio and the full M3 Max chip is quite tasty to the point where I might forego to want the M3 Ultra. Hopefully, they will launch the updated Studios early Spring instead of close to the WWDC 2024.

(This is about going from laptop to desktop)

 

[Chancha]

That's the point. Of course there is value. But it's not value that I can defend in front of my finance director since I don't really need those extra GPU cores or RAM bandwidth.

Frankly a $3000+ price tag of the "base" M3 Max is already hard to defend, especially in bulk. Real world benchmark tests may even show diminishing returns for anything above this config. Lately I think Apple strikes a pretty good spot for these Max SKUs on the 14" 16" options, the ones you can buy right off of the retail store are usually the ones to get. Though I am puzzled why this time the 2nd tier 16" M3 Pro with 36GB only has a 512GB SSD, which we know will have gimped speed.

 

[leman]

Frankly a $3000+ price tag of the "base" M3 Max is already hard to defend, especially in bulk.

I think the price tag is adequate. After all, you are getting the performance of a desktop workstation with the footprint and a battery life of an ultrabook. Not to mention a phenomenal display.

It's also priced more than competitively against other workstation laptops. An equivalent Dell or HP would cost almost $2000 more, and that's a completely different form factor with crap battery.

 

[Pressure]

Finally someone did a Geenkbench 6 run with Metal instead of OpenCL.

47651 Metal Score for M3.

 

[diamond.g]

Finally someone did a Geenkbench 6 run with Metal instead of OpenCL.

47651 Metal Score for M3.

Lol I got a website under heavy load try again later message. Looks like everyone wants to look at that score.

 

[eoren1]

While I'm not the final authority on any of this... I believe I already answered. N3E will be used very widely by many companies for many years. As I and others posted previously, it's the last TSMC FinFET node, and as such it's going to be around for MANY years.

N3B, on the other hand, will be dead the minute Apple stops ordering M3 and A17 chips.

But maybe you were trying to ask specifically if *Apple* will be using it for anything besides iPhones and iPads? If so... hard to say. Apple will very likely switch to N3P for M4... IF it's available in time. I don't have a good sense of whether that's likely or not.

I came to this thread after hearing about N3B vs N3E for the first time on a YT post by 'Snazzy Labs'. Looked here then read more about N3B and am confused about whether I should keep my pre-order for the M3 Max at this point.

Is there concern that N3B will be a stopgap chip and the N3E will be a silent refresh of the M3 line when it becomes available?
Is the N3B inherently worse or just more expensive/lower yield than the N3E?
And I did not follow a quote from semianalysis which said 'N3E is not IP-compatible with N3B IP. This means that IP blocks have to be reimplemented' - will that matter to the end user?

I'm moving up from an underpowered 14" M1 Pro with only 16gb of RAM that I jumped on rather than waiting for weeks/months at the time for the 32gb version I wanted. Just need to decide if this is the right chip to jump on or if it is a stopgap and worth waiting longer (I know you can always wait longer in tech...)

 

[Confused-User]

I agree. In fact, Apple lost some money on me today. For my work, CPU is more important than the GPU, and I cannot justify the $500 markup just of the two additional P-cores, even though I'd really liked to have them. So in the end I am getting the 10+4/30 core model, which is a shame. I would have even paid $100 extra for the two P-cores.

So having agreed on all this, I now have a followup question, the answer to which is beyond my limited expertise, but perhaps someone else knows the answer.

Is there a technology coming soon (or already here) which would allow Apple to put together mix-and-match chiplets on a retail basis, rather than paying an OSAT or TSMC to do it in bulk? Obviously, putting them in sockets won't do - that's sort of the whole point of advanced packaging. But it there a possibility of, say, snapping two chiplets together with direct silicon-to-silicon bonding, or with some sort of interposer that doesn't drastically drive up the cost and latency of moving bits around? Or, well, I don't know what else, which is sort of the point?

The idea is, maybe we can no longer expect to put in GPUs on PCIe cards, but maybe Apple could make interchangeable modules of some sort, that would preserve most of the benefits of their current tight integration, while still allowing every machine to be configured separately? Perhaps not by the user, but at least in, say, a low-grade clean room at Apple stores.

I'm not aware of anything that would make this possible, unfortunately, but I may just be too ignorant.

 

[dugbug]

This is really not right.

People keep talking about "just" "some clock speed boosts" but they don't appreciate that there is no "just" about it. It's a MAJOR (to use your term and uppercaseness) effort to make the chip scale up that much in frequency. Being on a new process node makes that possible, but you also need to redo the logic.

I mean are they that different? Not really. They all three are rectangular. They all three have a bunch of cores. Its the same sh*t over and over.

-d

 

[altaic]

I came to this thread after hearing about N3B vs N3E for the first time on a YT post by 'Snazzy Labs'. Looked here then read more about N3B and am confused about whether I should keep my pre-order for the M3 Max at this point.

Is there concern that N3B will be a stopgap chip and the N3E will be a silent refresh of the M3 line when it becomes available?
Is the N3B inherently worse or just more expensive/lower yield than the N3E?
And I did not follow a quote from semianalysis which said 'N3E is not IP-compatible with N3B IP. This means that IP blocks have to be reimplemented' - will that matter to the end user?

I'm moving up from an underpowered 14" M1 Pro with only 16gb of RAM that I jumped on rather than waiting for weeks/months at the time for the 32gb version I wanted. Just need to decide if this is the right chip to jump on or if it is a stopgap and worth waiting longer (I know you can always wait longer in tech...)

No. Also semianalysis is terrible.

 

[Confused-User]

I came to this thread after hearing about N3B vs N3E for the first time on a YT post by 'Snazzy Labs'. Looked here then read more about N3B and am confused about whether I should keep my pre-order for the M3 Max at this point.

Is there concern that N3B will be a stopgap chip and the N3E will be a silent refresh of the M3 line when it becomes available?
Is the N3B inherently worse or just more expensive/lower yield than the N3E?
And I did not follow a quote from semianalysis which said 'N3E is not IP-compatible with N3B IP. This means that IP blocks have to be reimplemented' - will that matter to the end user?

I'm moving up from an underpowered 14" M1 Pro with only 16gb of RAM that I jumped on rather than waiting for weeks/months at the time for the 32gb version I wanted. Just need to decide if this is the right chip to jump on or if it is a stopgap and worth waiting longer (I know you can always wait longer in tech...)

You should completely ignore this. It's utterly irrelevant to you. It's like worrying where the sand came from, that went into making the silicon in your chip. (FWIW though, N3E is not faster than N3B, generally speaking, and is in some ways a slight regression, though it's noticeably cheaper.)

The only question that matters is, will it be a good machine for you? Every indication is that the M3 is a solid upgrade at least, and it may be that on multicore it's a home run - we don't have enough info yet but it's looking extremely good. If you need a new machine, unless you *really* need it in a desktop, and are willing to wait 1-6 months more for that (we don't know when the Studio is coming), you should get it right away.

Like every machine you could ever buy in the recent past or near future, it will be superseded by something better, probably within a year. Whatever, that's life on the tech curve. Buy what you need and don't stress about inevitable progress.

 

[me2icy]

I desperately need to update my Intel based mac and at the moment I have two option both around 2500Eur

new IMac 8/10/16/1tb

or

Mac mini M2pro 10/16/16/1tb + Dell U2723QE + need to buy touchpad, keyboard separately

What do you guys think would be the best decision?

 

[Confused-User]

I mean are they that different? Not really. They all three are rectangular. They all three have a bunch of cores. Its the same sh*t over and over.

I can't tell if you're being facetious or not. If you are, good one. If not... well, the M3 is a lot like an elephant. They're both made of matter and can both be found only here on Earth, in all the galaxy, so, I guess, they're the same sh*t over and over?

It's true but it's a useless point to make.

For those of us interested in chip architecture, on all the levels that matter, the M3 CPU core is quite different from the M2's core.

 

[dugbug]

I can't tell if you're being facetious or not. If you are, good one. If not... well, the M3 is a lot like an elephant. They're both made of matter and can both be found only here on Earth, in all the galaxy, so, I guess, they're the same sh*t over and over?

It's true but it's a useless point to make.

For those of us interested in chip architecture, on all the levels that matter, the M3 CPU core is quite different from the M2's core.

yes I was jokingly indirectly going after the person you quoted. Im a hennessy-patterson guy just like most of us