Hard Truths About The Apple Silicon Replacement to the Mac Pro (2019)

[dmccloud]

Remember Optane Memory in Intel machines? Could AS end up adding something like that to the SoC's?

Intel killed off the Optane line last August.

Intel Kills Optane Memory Business, Pays $559 Million Inventory Write-Off

3D XPoint at the last crossroad.

www.tomshardware.com

To be honest, it never made sense in my mind to use an M.2 slot for Optane when most use cases would benefit more from a fast SSD instead. It was a solution waiting for a problem to resolve instead of a solution TO a problem.

 

[kpluck]

The fact of the matter is nobody knows, outside of Apple, what features the Mac Pro will ultimately have. Definitively stating likely things as 100% fact is seems pointless to me.

-kp

 

[ksj1]

I'm a bit puzzled at some of the comments that the Studio won't get an update. Why wouldn't they offer an M2 Studio Ultra with fewer CPU and GPU cores using binning, like they seem to be doing already. An M2 Ultra with 32-48 GPU cores would probably fit in the lineup, with lower memory limits than the Pro might have.

 

[sam_dean]

I'm a bit puzzled at some of the comments that the Studio won't get an update. Why wouldn't they offer an M2 Studio Ultra with fewer CPU and GPU cores using binning, like they seem to be doing already. An M2 Ultra with 32-48 GPU cores would probably fit in the lineup, with lower memory limits than the Pro might have.

I think the premises on this thread is stated in a vacuum absent of other supply chain factors that many of us are unaware off.

Without it it makes it sound weird.

 

[vladi]

I'm a bit puzzled at some of the comments that the Studio won't get an update. Why wouldn't they offer an M2 Studio Ultra with fewer CPU and GPU cores using binning, like they seem to be doing already. An M2 Ultra with 32-48 GPU cores would probably fit in the lineup, with lower memory limits than the Pro might have.

What's funny is that Intel's current laptop flagship CPU 1380HX is close to 25% faster than Studio with 20 core option.

 

[Joe Dohn]

What's funny is that Intel's current laptop flagship CPU 1380HX is close to 25% faster than Studio with 20 core option.

B-but my Appel!... Mah Silicon!

 

[ADGrant]

What's funny is that Intel's current laptop flagship CPU 1380HX is close to 25% faster than Studio with 20 core option.

Why is that funny? It's a newer design with more cores. It also consumes a significant amount of power for a laptop CPU.

It is a pity though that Apple's last Intel Mac was offered with 10th gen CPUs and not the 12th or 13th gen.

 

[scottrichardson]

If the M2 Ultra supports it, Apple could make use of the extensive cooling in the Mac Pro case to significantly increase its clock speed.

At 4.2 GHz, that should give GB SC/MC speeds of 2,300 and 35,000, exceeding the performance of both of the above. That would be notable because, at least for the past several years, there's not been a desktop processor that was no. 1 in both categories. [The workstation processors with the highest MC speeds have been slow for SC.]

4.5 GHz should give SC/MC speeds of 2,500 and 37,000.

My back-of-the-envelope math:
Current GHz/SC/MC for M2 Max in 16" MBP: 3.7/2,050/15,300
At 4.5 GHz: SC = 2,050 x 4.5/3.7 = 2,500; MC = 15,300 x 4.5/3.7 x 2* = 37,000
*Multiplying by 2 b/c it's the Ultra.

They might also be able to increase the GPU clocks, which are currently quite low.

These changes would of course give up some efficiency, but that's less of an issue for the Mac Pro than for any other Apple product—you'd still be left with CPU and GPU TDP's significantly below what the Mac Pro was designed for.

As sam_dean mentioned at the top of this page, the power delivery would need to scale up exponentially for the faster clock speeds. The power requirements are not linear, so to get that extra 0.5Ghz would require a solid whack of power, but I agree, still within the constraints of a Mac Pro tower enclosure.

I had suggested that we might see Apple have two flavours of the M2 Ultra:

M2 Ultra and M2 Ultra X.

Similar to how they released A10X and A12Z chips for iPad Pros. The M2 Ultra X could be a higher clocked variant with a few extra cores for both the CPU and GPU, without taking it as far as the M2 Extreme that has been put on the backburner due to the complexities of the four-way ultra-fusion.

 

[leman]

What's funny is that Intel's current laptop flagship CPU 1380HX is close to 25% faster than Studio with 20 core option.

Where did you get that from? 13900HX gets 20k in GB5 multicore, M1 Ultra gets 23K. And the Intel chip needs 3x more power to achieve this result. In single Core, sure Intel is about as fast as M2 Pro. While using 5x more power. It is funny in retrospect.

Anyway, the HX series is hardly a “mobile” chip. These are undervolted desktop CPUs. They even use a different socket compared to the usual laptop CPUs. We‘ll see a couple extremely expensive limited quantity flagship gaming bricks using these and that’s about it. Most laptops people will actually buy will have a 13700H or friends in them.

 

[leman]

It is a pity though that Apple's last Intel Mac was offered with 10th gen CPUs and not the 12th or 13th gen.

No Mac chassis was designed for the insane power consumption of those chips.

 

[theorist9]

As sam_dean mentioned at the top of this page, the power delivery would need to scale up exponentially for the faster clock speeds. The power requirements are not linear, so to get that extra 0.5Ghz would require a solid whack of power, but I agree, still within the constraints of a Mac Pro tower enclosure.

I had suggested that we might see Apple have two flavours of the M2 Ultra:

M2 Ultra and M2 Ultra X.

Similar to how they released A10X and A12Z chips for iPad Pros. The M2 Ultra X could be a higher clocked variant with a few extra cores for both the CPU and GPU, without taking it as far as the M2 Extreme that has been put on the backburner due to the complexities of the four-way ultra-fusion.

The usual rule of thumb is that the power scales as the square of the clock speed, rather than exponentially (and it should be acknowledged sam_dean didn't say exponentially either). We can use this to roughly predict the power draw at higher clocks.

According to https://www.notebookcheck.net/Apple-M2-Max-Processor-Benchmarks-and-Specs.682771.0.html:

"When fully loading the CPU and GPU cores, the chip uses up to 89 Watt and the CPU part is limited to 25 Watt."

Thus we can estimate that, for the CPU, 25 W@3.7 GHZ for the Max would extrapolate to the following for higher-clocked M2 Ultras (I multipled by two to go from the Max to the Ultra):
25 W x (4.2/3.7)^2 x 2 = 64 W @ 4.2 GHz
25 W x (4.5/3.7)^2 x 2 = 74 W @ 4.5 GHz

So if it really is quadratic (increases as the square of the frequency), the difference between 50W for a 3.7 GHz M2 Ultra, vs. about 65W – 75 W for higher-clocked variants, would seem to be trivial for a Mac Pro.

Even if the increase is stronger than quadratic, we're still probably under 100W even at the higher clock speed. That is significantly lower than the power budget of even the least powerful CPU that goes into the current Mac Pro case, the 3.5 GHz 8-Core Xeon W (W-3223), whose nominal TDP is 160W.

So, for an M2 Ultra in the current Mac Pro case, the limit on scaling to higher clocks should depend solely on the limits of the M2 chip itself.

 

[leman]

The usual rule of thumb is that the power scales as the square of the clock speed, rather than exponentially (and it should be acknowledged sam_dean didn't say exponentially either). We can use this to roughly predict the power draw at higher clocks.

According to https://www.notebookcheck.net/Apple-M2-Max-Processor-Benchmarks-and-Specs.682771.0.html:

"When fully loading the CPU and GPU cores, the chip uses up to 89 Watt and the CPU part is limited to 25 Watt."

Thus we can estimate that, for the CPU, 25 W@3.7 GHZ for the Max would extrapolate to the following for higher-clocked M2 Ultras (I multipled by two to go from the Max to the Ultra):
25 W x (4.2/3.7)^2 x 2 = 64 W @ 4.2 GHz
25 W x (4.5/3.7)^2 x 2 = 74 W @ 4.5 GHz

So if it really is quadratic (increases as the square of the frequency), the difference between 50W for a 3.7 GHz M2 Ultra, vs. about 65W – 75 W for higher-clocked variants, would seem to be trivial for a Mac Pro.

Even if the increase is stronger than quadratic, we're still probably under 100W even at the higher clock speed. That is significantly lower than the power budget of even the least powerful CPU that goes into the current Mac Pro case, the 3.5 GHz 8-Core Xeon W (W-3223), whose nominal TDP is 160W.

So, for an M2 Ultra in the current Mac Pro case, the limit on scaling to higher clocks should depend solely on the limits of the M2 chip itself.

Your math is way off. The CPU doesn’t consume 25 watts running at 3.7 GHz, it will consume close to 40-45 watts (and it won’t run full 3.7ghz either). The review you quite describes a power virus scenario on both CPU and GPU, where CPU will be severely power-limited. And second, it’s not obvious as all that this design is even able to reach 4.0ghz. Power consumption is not the only limitation of clock frequency.



Edit: I know nothing about these things but I found this nice explanation on Quora:

Basically the internal operation of all of the larger functional elements (registers etc) within a cpu depends on smaller circuit elements called flip-flops and latches Flip-flop (electronics) that react to and potentially change state once per clock cycle. They are designed to be stable within a model-dependent range of frequencies. If you raise (or lower) the clock rate outside of that range, the latches can't work. Their states at the end of each clock cycle become a random crapshoot, which means your finely tuned cpu becomes a drooling idiot.

Why do computer chips malfunction when the driving clock is above a certain frequency?

Answer (1 of 7): They give the wrong answer or crash because when you speed up the clock rate eventually a zero to a one transition isn't going to get a chance to get up to the "one" level in time, so the receiving gate will see it as a "zero", so the wrong info gets received and things go all wo...

www.quora.com

 

[scottrichardson]

Your math is way off. The CPU doesn’t consume 25 watts running at 3.7 GHz, it will consume close to 40-45 watts (and it won’t run full 3.7ghz either). The review you quite describes a power virus scenario on both CPU and GPU, where CPU will be severely power-limited. And second, it’s not obvious as all that this design is even able to reach 4.0ghz. Power consumption is not the only limitation of clock frequency.

Can you talk more on the other limitations of clock frequency, in reference to the current Apple silicon architecture? I am genuinely interested.

 

[leman]

Can you talk more on the other limitations of clock frequency, in reference to the current Apple silicon architecture? I am genuinely interested.

Added a quote and a link to my previous post. I am not an electrical engineer and I don’t understand any of this stuff, but it makes sense to me that a system is designed for a specific range of speeds and will only work there. Designing for different ranges can have different advantages and disadvantages. I think we are too used to x86 chips which can operate at very high frequencies and often assume that this is true for all microprocessors. The heat is the limiting factor is what we learn on the internet. But this doesn’t seem to make much sense even with the x86 CPUs. Current chips can easily draw hundreds of watt of power and yet a single core only does like 20-25 watts when operated in single core mode. If it was just about heat why don’t they clock it higher? Clearly there must be another limiting factor.


Hopefully some of the actually knowledgeable people will see it and can comment with more insight.

 

[theorist9]

Your math is way off. The CPU doesn’t consume 25 watts running at 3.7 GHz, it will consume close to 40-45 watts (and it won’t run full 3.7ghz either). The review you quite describes a power virus scenario on both CPU and GPU, where CPU will be severely power-limited.

Actually, it's your critique that's way off. My math is fine. Instead, you're saying the value I used for the M2 CPU's TDP in my equation is too low. If, so then (using the same math), we have:
45 W x (4.2/3.7)^2 x 2 = 116 W @ 4.2 GHz
45 W x (4.5/3.7)^2 x 2 = 133 W @ 4.5 GHz

...both of which are still well under the cooling capacity of the Mac Pro.

And second, it’s not obvious as all that this design is even able to reach 4.0ghz. Power consumption is not the only limitation of clock frequency.

That's just repeating what I already said:

So, for an M2 Ultra in the current Mac Pro case, the limit on scaling to higher clocks should depend solely on the limits of the M2 chip itself.

My first post on this likewise indicated this was conditional: "If the M2 Ultra supports it." [emphasis added here]

So you've not said anything to contradict my two main points (the first of which was implicit, but obvious):

1) The M2 currently runs at such a low clock speed that it *may* have added headroom for increased clocks that is not available to the top AMD and Intel chips, which are already running at such high clocks.

2) If the M2 can be adjusted to higher clocks, then the Mac Pro case will have no issues handling the TDP.

Hopefully some of the actually knowledgeable people will see it and can comment with more insight.

I already made use of that expertise. When I wrote my post, I was recalling this from former AMD chip designer cmaier:

That seems much more on-point than the Quora exerpt you attempted to use to critique my post.

 

[sam_dean]

Can you talk more on the other limitations of clock frequency, in reference to the current Apple silicon architecture? I am genuinely interested.

The reason why Apple has the efficiency edge is mainly due to Apple having an edge in terms of the process node (5nm, 4nm, 3nm) they use, the PDN (power delivery network) tech, and packaging.

Their ARM cores are actually more complex than the x86 competitors; significantly wider and with larger resources for out of order and speculation. Most people assume there is some kind of "magic" that makes ARM better that x86, but that is not the case. The ISA has little impact on overall power consumption given the same microarchitectural resources.

Apple uses their larger/more complex cores to their advantage, by running them at a slower clock rate. While allowing them to do more work per clock cycle. This allows them to operate on the frequency/power sweet spot for their process. One has to note that power consumption increases significantly (way higher than linear) the higher the frequency.

That's why overclocking or up clocking Apple chips is not done. Want to double performance linearly? Double the the dies.

So M1 Max is doubled to become M1 Ultra.

You may notice on the Apple Store webpage of any Mac they advertise how many CPU/GPU/Neural Engine cores. No clock speeds are mentioned. Clock speeds are only detailed in deep dive tech-heavy hardware reviews.

This is not unique to Apple. IIRC as early as AMD's Athlon chips the clockspeed of a chip became an unreliable measure of CPU performance.

People get unintionally emotional about this as this is a major change of how we think desktops and laptops should be measured. Below is the reason why.

Chip design decision is influenced by the business models of Apple, Intel, Nvidia, AMD, etc.

Apple is a system's vendor. Meaning that they sell the finished product, not just the processors/part. So they can use several parts from the vertical process to subsidize others. In this case, Apple can afford to make very good SoCs because they don't sell those chips elsewhere, meaning that they are not as pressured to make them "cheap" in terms of area for example. Since they're going to recoup the profit from elsewhere in the product.

In contrast; AMD and Intel sell their processors to OEMs, so they only get profit from the processor not the finished system. So they have to prioritize cost, by optimizing their designs for Area first and then focus on power. This is why both AMD and Intel use smaller cores, which allows them for smaller dies. But which have to be clocked faster in order to compete in performance, unfortunately that also increases power.

This is probably they key difference; Apple can afford the larger design that is more power efficient for the same performance. Whereas AMD/Intel have to aim for the smaller design that is less power efficient for the same performance.

 

[sam_dean]

Where did you get that from? 13900HX gets 20k in GB5 multicore, M1 Ultra gets 23K. And the Intel chip needs 3x more power to achieve this result. In single Core, sure Intel is about as fast as M2 Pro. While using 5x more power. It is funny in retrospect.

Anyway, the HX series is hardly a “mobile” chip. These are undervolted desktop CPUs. They even use a different socket compared to the usual laptop CPUs. We‘ll see a couple extremely expensive limited quantity flagship gaming bricks using these and that’s about it. Most laptops people will actually buy will have a 13700H or friends in them.

Odds are this is treated as a marketing expense if it doesn't hit its KPI.

 

[leman]

Actually, it's your critique that's way off.

Not intended as a critique. Simply pointing out that the values you were using were likely incorrect. Even a small difference in the input parameters will change the result a lot. For example, using the actual multicore frequency of P-clusters (3.2Ghz) and your formula I get 180 watts. But then there are E-cores. We could probably say that a hypothetical vertically scalable M2 Max could operate at 4.5 Ghz (40% increase in performance) at under 200 watts. Does that tell us anything about the upcoming Mac Pro? Probably not...

I would love to see powermetrics output for single- and multi-core loads of M2 Max, but I am not aware of anyone having done this yet. And even then people usually take Cinebench (because it's easy to set up), and it doesn't reach maximal power usage on Apple Silicon

View attachment 2155745

That seems much more on-point than the Quora exerpt you attempted to use to critique my post.

Oh, I fully agree with Cliff. In fact, that is something I have been reiterating (and hoping for) since before M1 Pro/Max were even released. If Apple is serious about desktop they will have to design a new class of CPUs with more raw performance, either through higher clocks or some other means. But we don't know how such chip might perform, what it's energy efficiency will be, or what the power targets Apple will choose. I am not sure whether speculating about the power usage of a hypothetical M2 Mac@4.5 Ghz will help us anticipate the power curve of such Apple chip.

That's why overclocking or up clocking Apple chips is not done. Want to double performance linearly? Double the the dies.

One way to look at is is that Apple deliberately sacrifices the wider operating frequency range for energy efficiency. One thing is find very interesting is that so far the peak single-threaded performance was tuned precisely to be competitive with best mobile x86 chips. I don't think this is a coincidence. Appel probably has a bunch of simulators running 24/7 tweaking micro parameters to hit certain targets.

 

[sam_dean]

One way to look at is is that Apple deliberately sacrifices the wider operating frequency range for energy efficiency. One thing is find very interesting is that so far the peak single-threaded performance was tuned precisely to be competitive with best mobile x86 chips. I don't think this is a coincidence. Appel probably has a bunch of simulators running 24/7 tweaking micro parameters to hit certain targets.

Because Apple's business model is a system vendor that controls end to end then they can easily do that at a competitive price relative to the wanted outcome of their customers.

With the future M2 Ultra 24-core CPU can anyone predict its performance? We have an idea what the M2 Max raw performance and power consumption are.

No doubt it will run loud and hot in a MBP 16" but not as loud and hot as that Intel Core i9 24-core CPU in any laptop.

Can be easily powered by a less than 240W charger vs a 330W charger.

 

[bobcomer]

No Mac chassis was designed for the insane power consumption of those chips.

Yeah? (NOT!)

Just MBP and Mac Mini's, iMac's and especially iMac Pros. i5's and i7's in the smaller machines and i9 in the iMac Pro's (and Mac Pro ---- of course)...

 

[leman]

Yeah? (NOT!)

Just MBP and Mac Mini's, iMac's and especially iMac Pros. i5's and i7's in the smaller machines and i9 in the iMac Pro's (and Mac Pro ---- of course)...

Come on. People were already complaining about heat, noise and battery life of Coffee Lake i7 CPUs, and Alder Lake easily has PL2 two times higher.

 

[bobcomer]

Come on. People were already complaining about heat, noise and battery life of Coffee Lake i7 CPUs, and Alder Lake easily has PL2 two times higher.

Come on, that doesn't mean Apple couldn't have done better cooling-wise, *and* there are intel chips that are lower power by design. You're thinking i9's too much. I know an i7 would fit easily in the MBP14. Since I'm running that hardware in a Lenovo Windows laptop (ultralight) and I never hear the fans, even when running multiple VM's.

 

[leman]

Come on, that doesn't mean Apple couldn't have done better cooling-wise, *and* there are intel chips that are lower power by design. You're thinking i9's too much. I know an i7 would fit easily in the MBP14. Since I'm running that hardware in a Lenovo Windows laptop (ultralight) and I never hear the fans, even when running multiple VM's.

Alder Lake i7-12700H has PL2 of 115watts while the 14" Pro is designed for 40-50 watts. Anyway, I thought we are taking about the old chassis? Wouldn't make much sense for Apple to put these Intel chips in once their own silicon was out. At least not from the performance perspective.

 

[deconstruct60]

No Mac chassis was designed for the insane power consumption of those chips.

I think you left off an adjective. "No laptop Mac chassis". The laptop chassis don't encompass the entirety of the Mac chassis designs.

The iMac Pro didn't have a completely symmetrical thermal coverage for CPU and GPU. The Vega GPUs Apple used were higher than the W-2100 and inside the usable envelope of the Gen11 or Gen12 CPU. If just swapped thermal constraints between CPU and GPU, the chassis would handle it. ( There are two fans so need to adjust the role switch with the fan thermal control profiles. ).
And of course not pull the goofy move of allowing the Intel CPU to ask for any power level it wanted (cough MPB 16" intel ). PL2 doesn't have to last long. As pointed out, Intel runs it downclocked as a laptop processor just fine while doing productive work.

More than a few folks would not really like the drop in GPU thermals. They'd need the most bleeding edge 7600/4060 class GPU not to backslide much. However,


As for the Mac Pro 2019 chassis. ... you are joking right? Could the Gen 11 /12 provision 8 PCI-e slots well. Not really. But thermally wise, "around 20" Intel cores is what it was designed for.

 

[deconstruct60]

It is a pity though that Apple's last Intel Mac was offered with 10th gen CPUs and not the 12th or 13th gen.

The last 'new' Intel Mac was always highly likely going to be a desktop. So far it is the iMac 27" . Probably going to stay that way.

That was mainly on Intel. The Gen10 "Comet Lake" desktop processors didn't come out until Q2 2020.

Core i7 10700K ( for 2020 iMac 27" )

Intel® Core™ i7-10700K Processor (16M Cache, up to 5.10 GHz) Product Specifications

Intel® Core™ i7-10700K Processor (16M Cache, up to 5.10 GHz) quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

Rocket Lake ( which really wasn't much of a good update ) didn't Appear until Q1 2021.

Product Specifications

quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

Apple 'needed' best Intel processor they could get in 2020. If there had been a good (decent clocks and thermals ) Ice Lake Gen 11 desktop processor in June 2020 , then Apple probably would have used it. It never came. Rocket Lake was a somewhat kludged backport that was more a 'stop gap filler' than a good target.

Very similar issue with iMac Pro. W-2200 was basically a mildly tweaked W-2100 rebadged and given a saner, more competitive price point.

Also same 'slow motion train wreck' with W-3200 , super duper late , way off initial thermal target , peak boost clock regressions (major reason for the backport stunt for mainstream desktop Gen11 ) . The other major workstation vendors (Dell/HP/Lenovo) skipped it also.


By 2017-2018 Apple could have had very reasonable expectations that Intel was going to completely bungle their processor roll out in 2020. ( probably being given highly NDA briefings about 2019-2020 products by then. And left the briefings thinking Intel was blowing lots of smoke. ) That would make for a very good inflection point. Certainly would help in Apple's own smoke blowing exercise in saying they had 'desktop killer' SoCs to offer.

Highly doubtful Apple would want to do the transition when Intel was on the rebound upswing or when AMD had solidly passed Intel on their desktop upswing. ( Apple didn't want to have to use AMD to close out the x86 path nor fend off Intel making a 'marketplace' argument to stay longer (or try extended dual macOS path). )