Rumored pro chip

[Unregistered 4U]

By the end of 2022 we'll have M2X. Apple will be for sure refreshing its chips every year. I wouldn't be so sure about Intel.

Plus, it really doesn’t matter how fast Intel’s processors are anymore. All each successive generation of M series chips has to do is be faster than the prior one. If FCP, Logic, Xcode etc. is more performant by a decent margin, then it’s a win.

I’m almost certain that someone somewhere will be able to overclock a stressed Intel processor such that it’s faster than an M series chip at one specific benchmark or another, and it literally doesn’t even matter!

 

[AgentMcGeek]

As you said yourself, the best CPU in an ultra compact form factor. MBP aren’t desktop remplacements. Even the 16” remains thin. And we know what happened when they put an i9 in a 5mm chassis. That’s why the perf per watt ratio of the M series is incredibly well suited for MBP.

 

[richinaus]

They are not, but they are competing with other laptop manufacturers in the premium market segment. As Apple say themselves, they don’t sell parts, they sell product experience. And performance is an important part of that experience.



I disagree. Apple has long made best effort to deliver fastest possible CPU in a ultraimpact form factor. Their new entry level CPU core competes toe to toe with much more expensive fastest available desktop designs. As I’ve said before, I see no world where Apple will be satisfied in holding the second place.

Regarding GPUs, yes, Apple will be ok with delivering mid-range to lower-high-end performance, but they will do so in a form factor and with power consumption others can only dream about.

We will see what they deliver. But my $$ are on a great laptop that isnt the fastest around, but doesn’t overheat, become loud, and is at the right weight to be transportable. That is what it needs to be.

 

[Falhófnir]

Anything is fast enough to run basic Mac applications. But when I am running my data analysis scripts or building software I need top single core performance. And if Intel beats Apple in this department it would be really embarrassing, especially after the extremely strong start that M1 had.



Current mobile Tiger Lake H is pretty much toe to toe with M1, with later being a tad faster in SPEC. This is quite embarrassing for Intel really, with M1 being a low-power SoC and all. But if the prosumer silicon retains the same single-core performance, and Alder Lake is 10% faster than TL, then Intel gets slightly ahead.



Of course not. Don’t be silly.

Hmm, well if nothing else that's another reason it's going to be extremely awkward if Apple really are planning on updating the M chips on some sort of >12 month cycle like the iPad AX chips...

 

[AgentMcGeek]

Intel may catch up slowly as it starts leveraging TSMC nodes. We can’t rule that out. But the inherent advantages of ARM over x86 will remain.

 

[leman]

That’s why the perf per watt ratio of the M series is incredibly well suited for MBP.

There is litte doubt that Apple will retain an insurmountable lead in performance/watt, which will also give them the best sustained performance at any given TDP target.

I am however talking about peak single-core performance, which has traditionally been very important for Apple and where Apple CPUs have always been exceedingly strong. I believe that Apple will want to have a commanding lead in both single-core and multi-core performance, which is not possible with current Firestorm@3.2. If the prosumer CPU uses an enhanced Furestorm @4.0 ghz, that’s a different story entirely.

 

[AgentMcGeek]

When you say Apple CPUs, you mean iPhone / iPad A-series chips? If so, I don't see why Apple could not maintain its lead, since the M-series CPU cores are based on A-series architectures. The A chips have been consistently improving year-on-year, due to both their excellent engineering teams and TSMC lithography.

 

[Unregistered 4U]

Intel may catch up slowly as it starts leveraging TSMC nodes. We can’t rule that out. But the inherent advantages of ARM over x86 will remain.

Intel’s bigger problem is that it’s going to be harder for them to shrink due to their inherent complexity. I’m convinced if their processors were just “simpler”, they might have hit their targets. If they could make a break from x86, redesign from the ground up some new amazing architecture, I think they could… but the moment they did, AMD would just keep producing the same ol’ same ol’ and crater Intel’s method of money making.

If AMD wasn’t around, there MIGHT be an opportunity to see what Intel is truly capable of. As it is, they have no choice but to continue “the same thing as yesterday, just faster”.

 

[cmaier]

Intel’s bigger problem is that it’s going to be harder for them to shrink due to their inherent complexity. I’m convinced if their processors were just “simpler”, they might have hit their targets. If they could make a break from x86, redesign from the ground up some new amazing architecture, I think they could… but the moment they did, AMD would just keep producing the same ol’ same ol’ and crater Intel’s method of money making.

If AMD wasn’t around, there MIGHT be an opportunity to see what Intel is truly capable of. As it is, they have no choice but to continue “the same thing as yesterday, just faster”.

The last time Intel tried something new we got Itanium.

‘Nuff said.

 

[Unregistered 4U]

The last time Intel tried something new we got Itanium.

‘Nuff said.

Welllll, yah. In my heart of hearts, I still commend the attempt to rid themselves of the cruft, but the entire PC industry will always want “the same as yesterday, just faster”, so it’s locked in a cycle of incremental almost inconsequential tweaks.

And since that time, their breadwinner has gotten more and more crufty, to the point where I think however a bad job they’d do on “new-Itanium”, it’d be better than the woefully inadequate last 8 years. Then again, yeah, they weren’t even able to create a decent 5G chip which was essentially “trying something new.” If Qualcomm comes out with something cheap that can eat into the i3/i5 space, it’d be like someone winning 80 of the mid-size jet market. At that point, it’s not even profitable for the competition to keep making the planes… In that view Intel might have to give up the masses of i3/i5 sales.

 

[CWallace]

The last time Intel tried something new we got Itanium.

‘Nuff said.

Kind of irconic that when Netburst hit the wall with thermals, the Core microarchitecture "saved the day" by performing as fast, but running so much cooler.

Now Core has hit the wall with thermals, yet Intel does not appear to have anything to save them again.

 

[Kpjoslee]

Kind of irconic that when Netburst hit the wall with thermals, the Core microarchitecture "saved the day" by performing as fast, but running so much cooler.

Now Core has hit the wall with thermals, yet Intel does not appear to have anything to save them again.

Well, Alder Lake is first real attempt at Big-Little architecture (Lakefield was nothing more than a test project lol) so we will find where Intel stands against the competition when it launches. Other than that, Intel's first radical change won't happen until Meteor Lake which features foveros with multiple chiplets.

 

[mac1z]

How reliable is CPU-Monkey results? They claim to have some details about the apple silicon m1x/m2 processor. What do you guys think about this?

Apple M2 - Benchmark (cpu-monkey.com)

 

[EntropyQ3]

Well, Alder Lake is first real attempt at Big-Little architecture (Lakefield was nothing more than a test project lol) so we will find where Intel stands against the competition when it launches. Other than that, Intel's first radical change won't happen until Meteor Lake which features foveros with multiple chiplets.

"Chiplets" doesn’t mean "low power draw" though. If anything, these solutions draw more power because of having to go off-chip for communication and data flow between the components, as opposed to keeping it on the same die. Those costs have been reduced, but fundamentally the benefits of using chiplets lie elsewhere (and it’s questionable if those benefits are relevant for Apple products stack).

 

[Admiral]

I can't believe how many people think that the M1X is going into the redesigned MacBook Air.
This is laughable at this point.

I can't either. Apple has to achieve some kind of product differentiation; the current lineup, where MacBook Air, MacBook Pro, iMac, and Mac mini all have the same CPU and nearly identical performance, won't be tenable over the longer term. Apple will do this by shunting the higher-performance CPUs into the more "pro" machines while leaving the MacBook Air, iMac 24", and the low-end Mac mini as entry-level models.

Apple Silicon has a decade-long history of getting about 15-20% year-on-year uplift in performance. We can expect that the next-generation low-end machines are going to get this base 15-20%, while the pro machines are going to get some wild improvements as Apple adopts more cores (M1X) and possibly multi-CPU configurations — where we might see 70-350% increases in performance.

 

[thenewperson]

Assuming we get cores based on Firestorm, is increasing the decode-length a feasible way to get more single-core performance?

 

[cmaier]

Assuming we get cores based on Firestorm, is increasing the decode-length a feasible way to get more single-core performance?

Not really. They already get very good pipeline utilization, and your premise is that they are using the same firestorm, with the same number of pipelines.

 

[Joelist]

If they do an M1X where they use Firestorm and Icestorm but with more cores then single core would be pretty much the same but multicore could radically jump depending on the core counts. If for example they did the rumored 8 Big 2 Little setup it could essentially double.

 

[Kpjoslee]

"Chiplets" doesn’t mean "low power draw" though. If anything, these solutions draw more power because of having to go off-chip for communication and data flow between the components, as opposed to keeping it on the same die. Those costs have been reduced, but fundamentally the benefits of using chiplets lie elsewhere (and it’s questionable if those benefits are relevant for Apple products stack).

Of course not. I never said it will dramatically lower the power draw or anything like that lol. But it will give them much better flexibility on performance scaling.

 

[cmaier]

Of course not. I never said it will dramatically lower the power draw or anything like that lol. But it will give them much better flexibility on performance scaling.

Why? The reticle is big enough to handle anything they are rumored to do, without having to stick power-sucking interchip drivers in the middle of critical paths.

 

[Kpjoslee]

Why? The reticle is big enough to handle anything they are rumored to do, without having to stick power-sucking interchip drivers in the middle of critical paths.

At a lower cost. If the cost wasn't the issue, then AMD wouldn't have gone chiplets and Intel wouldn't be hell bent on working on foveros for their future processors.

 

[cbum]

I haven't seen this discussed recently, so maybe it's obvious, but will the M1X solve the RAM limitations?

 

[macsplusmacs]

I haven't seen this discussed recently, so maybe it's obvious, but will the M1X solve the RAM limitations?

Seems yes. as most of the rumors mention 32gig or 64 gigs of ram. Beyond that. who knows.

 

[cmaier]

At a lower cost. If the cost wasn't the issue, then AMD wouldn't have gone chiplets and Intel wouldn't be hell bent on working on foveros for their future processors.

AMD has to sell chips to customers. Apple doesn’t. Any increase in cost (which could only be due to yield fall-out) is easily absorbed in the price of the devices Apple sells.

 

[CWallace]

I haven't seen this discussed recently, so maybe it's obvious, but will the M1X solve the RAM limitations?

The real "RAM limitations" is that almost all on-package smartphone RAM are modules at or below 8GB in capacity, which is why M1 launched at 8GB (2x4GB) and 16GB (2x8GB).

Production availability of 16GB, 24GB and 32GB modules is improving so this should allow "M1X" (and M1/M2, for that matter) to offer 32GB and 64GB capacities, but supply might be very constrained (especially in the 24GB and 32GB capacities) which could be impact the release of machines.