Rumored M4 MBP geekbench

[Cmaier]

I gather M4 represents a larger microarchitecture change than some thought - the cores are now 10 wide instead of 8 wide. Interesting.

I think we’ve now gotten just about as wide as we’re going to get. My expertise isn’t Arm - I designed schedulers for Sparc and x86-64 - but it’s hard for me to imagine going wider wouldn’t lead to rapidly diminishing returns. Instructions tend to depend on the results of other instructions, and as you get wider and wider, you’ll find more and more of the time you won’t be able to find N instructions without dependencies. And if you’re relying on speculative execution (branch predictions, etc.) the penalty becomes progressively worse for mistaken guesses. Unless, of course, you go to something like multithreading, where you can issue completely unrelated instructions from different threads.

If they add more registers, perhaps, they can maybe reduce the likelihood of dependencies. Would be interesting to model that with real instruction streams.

 

[casperes1996]

I think we’ve now gotten just about as wide as we’re going to get. My expertise isn’t Arm - I designed schedulers for Sparc and x86-64 - but it’s hard for me to imagine going wider wouldn’t lead to rapidly diminishing returns. Instructions tend to depend on the results of other instructions, and as you get wider and wider, you’ll find more and more of the time you won’t be able to find N instructions without dependencies. And if you’re relying on speculative execution (branch predictions, etc.) the penalty becomes progressively worse for mistaken guesses. Unless, of course, you go to something like multithreading, where you can issue completely unrelated instructions from different threads.

If they add more registers, perhaps, they can maybe reduce the likelihood of dependencies. Would be interesting to model that with real instruction streams.

Just to be clear I assume you mean architectural registers not entries in the register file?

Personally I hope they’ll do something akin to amd’s 3D-vcache at some point. I can totally see that as a way to improve performance without needing to go wider or increase clocks too much. Assuming that there are meaningful memory stalls as is of course. Used properly more cache like that might also facilitate more aggressive prediction behavior as it can reduce the cost of a mispredict if neither case need to touch ram but both can fit in cache

 

[Cmaier]

Just to be clear I assume you mean architectural registers not entries in the register file?

Personally I hope they’ll do something akin to amd’s 3D-vcache at some point. I can totally see that as a way to improve performance without needing to go wider or increase clocks too much. Assuming that there are meaningful memory stalls as is of course. Used properly more cache like that might also facilitate more aggressive prediction behavior as it can reduce the cost of a mispredict if neither case need to touch ram but both can fit in cache

yeah, i meant architectural registers.

 

[theorist9]

Looks like we'll get to see the real thing soon. Hopefully this will include the M4 Pro and Max. Interested in seeing if they bump the clocks on the upper-end models.

 

[Cmaier]

Looks like we'll get to see the real thing soon. Hopefully this will include the M4 Pro and Max. Interested in seeing if they bump the clocks on the upper-end models.
View attachment 32161

they probably won’t.

 

[casperes1996]

Looks like we'll get to see the real thing soon. Hopefully this will include the M4 Pro and Max. Interested in seeing if they bump the clocks on the upper-end models.
View attachment 32161

Yeah. Must admit I’m a little disappointed we’re not getting an event. Especially if the Mac mini really is getting a redesign. It also suggests the pro and max won’t do anything that special. I assume exact same core configs as on M3 just with M4 gen cores.

 

[theorist9]

they probably won’t.

You're thinking we'll be seeing just the M4 Pro (for the Pro Mini), with the M4 Max MBP coming later this year? Or that it will be the base M4 only?

 

[NotEntirelyConfused]

they probably won’t.

Interesting. My response would have been "insufficient data". What makes you think that?

You're thinking we'll be seeing just the M4 Pro (for the Pro Mini), with the M4 Max MBP coming later this year? Or that it will be the base M4 only?

I think he's talking about whether the clocks will be higher on the high-end machines, not whether they're coming out at all.

 

[Yoused]

I am going to guess MBP 14 and 16 and maybe Mini/Mini Pro for now. MBAs will sit on M3 for now, because it is plenty good enough. Studio upgrades probably not until at least February.

Seems like a base iPad is due about now as well – it is way behind the curve.

 

[Cmaier]

You're thinking we'll be seeing just the M4 Pro (for the Pro Mini), with the M4 Max MBP coming later this year? Or that it will be the base M4 only?

no, i’m just saying they probably won’t bump the clocks

 

[mr_roboto]

Interesting. My response would have been "insufficient data". What makes you think that?

History is a pretty good guide. M1 Max was a mere 24 MHz faster than plain M1. Based on the data I've searched, there's no clock speed difference between M3 and M3 Max at all.

Anything's possible, of course, but the safe money is on Apple continuing this policy.

 

[theorist9]

Yeah. Must admit I’m a little disappointed we’re not getting an event. Especially if the Mac mini really is getting a redesign. It also suggests the pro and max won’t do anything that special. I assume exact same core configs as on M3 just with M4 gen cores.

They may be different. The base M3 Macs are 8 CPU cores (4/4), while the GB leaks for the base M4 Mac are showing 10 CPU cores (4/6). Not sure about the the GPU cores.

 

[theorist9]

History is a pretty good guide. M1 Max was a mere 24 MHz faster than plain M1. Based on the data I've searched, there's no clock speed difference between M3 and M3 Max at all.

Anything's possible, of course, but the safe money is on Apple continuing this policy.

But for the M2, the base & Pro were 3.5 GHz, while the Max & Ultra were 3.7 GHz. So there's not a consistent policy either way.

Perhaps the policy is to bump the clocks for even-numbered generations only.

 

[Jimmyjames]

I’m more optimistic than some perhaps? I’m not expecting huge changes, but based on what we’ve seen, performance should be great. I’m especially interested in gpu performance and whether there is a significant improvement in ray tracing along the lines of the A18 uplift.

One thing I would love is tandem oled to make its way to laptop and desktop screens, but that’s probably a few years away?

Overall, what pleases me most is the evidence that Apple is trying to keep to an annual update cycle. It should allow these improvements to compound and create even more impressive changes when judged over 3 or 4 years. I would be especially interested to see if the dual issue fp32/16 alu that @leman discussed makes it way to the M5 perhaps.

 

[NotEntirelyConfused]

History is a pretty good guide. M1 Max was a mere 24 MHz faster than plain M1. Based on the data I've searched, there's no clock speed difference between M3 and M3 Max at all.

Anything's possible, of course, but the safe money is on Apple continuing this policy.

The M2 generation had a couple hundred MHz difference.

Really, all history tells me is that they don't have any consistent behavior on this so far.

 

[casperes1996]

They may be different. The base M3 Macs are 8 CPU cores (4/4), while the GB leaks for the base M4 Mac are showing 10 CPU cores (4/6). Not sure about the the GPU cores.

To be clear I expect base M4 to be the same as in the iPad it’s already jib so 10 cores. Then pro and max following m3 core count growth

 

[NotEntirelyConfused]

To be clear I expect base M4 to be the same as in the iPad it’s already jib so 10 cores. Then pro and max following m3 core count growth

Yeah, I'm also not confident of that... though if I were forced to bet, I'd probably bet that way.

I think there's an open question: what size core cluster do they want? 4, 6,... or even 8? I'm a little dubious that they can get a single shared L2 cache big enough to service 8 P cores while maintaining adequately low latency, but what do I know? Does the answer differ between P and E cores? Does it change over time as the P and E cores themselves change?

I think it's entirely possible you'll again see a 6P+6E for the Pro. But I could also see 2x4P, or even 1x8P. And for the Max, similarly, you could see 2x6, 3x4 (probably not), 2x8, 4x4, or even 3x6 if they decide they want to take a big swing at big-bandwidth large parallel problems (possibly because they want something similar in their data centers).

As for the E cores, again, 6E does seem fairly likely all around - after all, they built it for the M4, and cutting it down will save very little. But they could. On the other hand, maybe they're looking at Intel's swarm of E cores and thinking "we can do really big things with really small cores". Maybe they do 12P+12E... or 12P+24E?

Giant numbers of E cores would definitely be a departure for them. Not a lot of existing code on Macs would be able to take advantage of it. So as I said, if I had to I'd bet against it. But the AS team at Apple has been operating under Steve Jobs rules better than any other part of Apple for a long time now ("skate to where the puck's going to be!") so I wouldn't rule this out.

The complete lack of solid info about this less than a week before launch is also REALLY interesting and unusual these days.

 

[dada_dave]

Yeah, I'm also not confident of that... though if I were forced to bet, I'd probably bet that way.

I think there's an open question: what size core cluster do they want? 4, 6,... or even 8? I'm a little dubious that they can get a single shared L2 cache big enough to service 8 P cores while maintaining adequately low latency, but what do I know? Does the answer differ between P and E cores? Does it change over time as the P and E cores themselves change?

I think it's entirely possible you'll again see a 6P+6E for the Pro. But I could also see 2x4P, or even 1x8P. And for the Max, similarly, you could see 2x6, 3x4 (probably not), 2x8, 4x4, or even 3x6 if they decide they want to take a big swing at big-bandwidth large parallel problems (possibly because they want something similar in their data centers).

As for the E cores, again, 6E does seem fairly likely all around - after all, they built it for the M4, and cutting it down will save very little. But they could. On the other hand, maybe they're looking at Intel's swarm of E cores and thinking "we can do really big things with really small cores". Maybe they do 12P+12E... or 12P+24E?

Giant numbers of E cores would definitely be a departure for them. Not a lot of existing code on Macs would be able to take advantage of it. So as I said, if I had to I'd bet against it. But the AS team at Apple has been operating under Steve Jobs rules better than any other part of Apple for a long time now ("skate to where the puck's going to be!") so I wouldn't rule this out.

The complete lack of solid info about this less than a week before launch is also REALLY interesting and unusual these days.

The current rumor is that the Pro is going back to being a cut die Max a la M1/M2. To me, that means to me the two most likely configurations for the P-cores are 1x6 for the Pro and 2x6 for the Max or 2x4 for the Pro and 3x4 for the Max with an extra CPU cluster appearing on the added die as well as an the extra GPU cores. I lean towards the 2x4 and 3x4 as it differentiates the Pro and base die more though maybe Apple doesn't care about that and wants to differentiate the Pro and Max more. So I could make a case for either if the rumor that the Pro isn't getting its own bespoke die is true. Actually even if it isn't true those are still the most likely configurations I'd argue - in addition to the cache point you brought up, each cluster also gets its own AMX unit and therefore cutting down on the number of cluster cuts down on the number of AMX units so having 8 P-cores potentially feeding a single AMX unit could make that into an extra bottleneck than it already is with 6. I do think they'll keep the 6 cluster E-core design for both the Pro and the Max if they share a common die area. If not, the Max might go back down to 4. Like you, I think it less likely that they'll go for the E-core swarm. They simply don't need one as their P-cores are pretty space efficient already, but maybe.

Not only is the lack of solid info interesting, but also the supposed week long rollout with no singular big press announcement is a bit odd.

 

[NotEntirelyConfused]

Not only is the lack of solid info interesting, but also the supposed week long rollout with no singular big press announcement is a bit odd.

It is even possible that they will surprise us with a big video intro announcement after all. Again, not betting on it, but... things feel *really* weird right now.

The rest of your analysis makes sense to me. But that doesn't mean it's right. I think there's some big-picture stuff we're not seeing yet. No idea if we'll be seeing more of it next week or not.

 

[dada_dave]

It is even possible that they will surprise us with a big video intro announcement after all. Again, not betting on it, but... things feel *really* weird right now.

Yup.

The rest of your analysis makes sense to me. But that doesn't mean it's right. I think there's some big-picture stuff we're not seeing yet. No idea if we'll be seeing more of it next week or not.

Oh absolutely. My post was "what makes sense to me", but it's all based on rumors and past history which may not be accurate or prologue.