TSMC starts production of A15 for iPhone 13. Macbook Pros will probably get SoCs based on A15.

[CWallace]

For Apple is no complication...they are just derivations...you cant have the same architecture as you have in an 13" ipad pro/iphone as you do in an mac pro.

Can't you?

Bloomberg says the ASi Mac Pro will have up to 40 CPU cores and up to 128 GPU cores. If you take four "M1X", each with 10 CPU cores and 32 GPU cores, and interconnect them, you have 40 CPU cores and 128 GPU cores. Take two, and you have the other Mac Pro configuration Bloomberg noted: 20 CPU cores and 64 GPU cores. And Apple has evidently been working on just such an interconnect for multiple M-series SoCs.

If Apple feels that ECC memory is a requirement for the ASi Mac Pro, they can add that support into the memory controller and use ECC DIMMs. If they feel the ASi Mac Pro needs 8 USB4/TB4 ports, they can add the additional controllers to support it.

This is really no different than how Intel does things. A W-2100 Xeon is the same as a Skylake-X Core i9. The differences are in the supporting chipsets. In the case of Apple Silicon, the "supporting chipset" is part of the main die.

 

[Lemon Olive]

Anything is possible, but to me, this sounds like complication where the goal of Apple Silicon as a family appears to be one of simplification by using a common foundation and tweaking it for the final product it is in.

Especially since...this is all that is needed.

This leman character is just smart enough to be his own enemy. He's completely confused on what Apple is doing, and using really poor logic to make very unlikely outcomes sound possible.

 

[09872738]

Especially since...this is all that is needed.

This leman character is just smart enough to be his own enemy. He's completely confused on what Apple is doing, and using really poor logic to make very unlikely outcomes sound possible.

Sorry to burst your bubble: @leman is way more often correct than wrong, generally pretty knowledgeable and tends to thoroughly back up his claims. Clearly he is no newbie when it comes to scientific working.

None of this applies to you. Compare @leman s thumbs up score with yours. Should tell you something.
His posts are actually both logical and correct. You confuse leaks and rumors with facts, which they may or may not be. Hence rumors do not qualify as proof of any sort. That is exactly what @leman argues, and he is very correct here. You, Sir, on the other hand, are not.

 

[leman]

This leman character is just smart enough to be his own enemy. He's completely confused on what Apple is doing, and using really poor logic to make very unlikely outcomes sound possible.

All I am saying that Apple might have a more thorough plan and better execution that you are prepared to give them credit for Anyway, it’s silly to get emotional over speculations. I am ok with being wrong just as much as I am ok with being right. We will see what they are up to very soon anyway.

@09872738 aww, thanks mate! That’s awfully nice of you

 

[09872738]

@09872738 aww, thanks mate! That’s awfully nice of you

My apologies, I am aware you are capable of speaking for yourself. And: My pleasure, not nice, just true

 

[leman]

My apologies, I am aware you are capable of speaking for yourself. And: My pleasure, not nice, just true

No, no, no apologees needed! I was not being sarcastic or anything like that. It’s just really nice to be appreciated once in a while.

 

[Krevnik]

Bloomberg says the ASi Mac Pro will have up to 40 CPU cores and up to 128 GPU cores. If you take four "M1X", each with 10 CPU cores and 32 GPU cores, and interconnect them, you have 40 CPU cores and 128 GPU cores. Take two, and you have the other Mac Pro configuration Bloomberg noted: 20 CPU cores and 64 GPU cores. And Apple has evidently been working on just such an interconnect for multiple M-series SoCs.

So, this is where language gets tricky, since what you say to me reads as multiple dies on a package with an interconnect between them. That could be right, but it could also be more about making a design that's easier to tile to 2x1 and 2x2 layouts that would be manufactured as a single die. Bloomberg is vague enough here that it's hard to take anything specific away from it.

That said, I suspect Bloomberg has at least a nugget of truth in this report, but some of that is because this sort of scaling for the Mac Pro already made a lot of sense to me a year ago.

Why not though? I mean, if all Apple was after is multi-core performance, things would be easy. They already have the best performing core under 5 watts. All they need to do is slap several dozens of these together and all the Xeons and EPYCs can "nervously smoke in the corner" (as we used to say in my home country).

But Apple doesn't just want to have the best multi-core processors. They want to have the best processors, period. And that is much more complicated.

True, but the flip side is that Apple also has been the type to offer as much speed as they can manage across the lineup, and they will want stuff to trickle down pretty quickly. But that to me just makes something like "Jade C-Die" more credible, and it makes sense to adopt Avalanche as part of that work if production on both the MBP chips and A15 are happening so close together.

Honestly, if they just wanted to ship an "M1X" with minimum changes needed to hit the performance marks to transition, it feels like they would have started production much earlier to me.

 

[thekev]

Which is why Apple has been using special Xeons with higher turbo boost.

Did I miss something or was this only applicable to the last mac pro? Apple's SKUs have appeared on Intel's list for every model I have checked. I think there may have been a weird exception around 2008 or something but that's about it, as far as I am aware.

 

[CWallace]

Honestly, if they just wanted to ship an "M1X" with minimum changes needed to hit the performance marks to transition, it feels like they would have started production much earlier to me.

I wonder if it is a case of lack of 5nm wafer capacity at TSMC?

In late 2020, there were reports that Apple had booked all of TSMC's 5nm production through the end of 2020 and into 2021 for the A14 and M1. And DigiTimes Asia reported last month that TSMC is boosting their 5nm production for the second half of this year.

If TSMC was constrained on wafer capacity for 2H20 and 1H21, I could see Apple concentrating on A14 and M1 production because the bulk of their products (iPhone 12, iPad Air, "Tranche One" Macs) used those chips and Apple did not want to run larger dies ("M1X") that each wafer could hold less of. And perhaps quality yields were lower, meaning even more "wafer waste".

Between better yield quality and greater wafer capacity, perhaps now is when Apple can "afford" to run larger dies because TSMC now has enough wafer capacity to handleA14/M1/M1X as well as starting the production ramp of A15 for the next iPhone.

 

[CWallace]

Did I miss something or was this only applicable to the last mac pro? Apple's SKUs have appeared on Intel's list for every model I have checked. I think there may have been a weird exception around 2008 or something but that's about it, as far as I am aware.

Intel did manufacture three "B" model Xeon W's for use in the iMac Pro: the W-2140B (8 cores), the W-2150B (10 cores) and the W-2170B (14 cores). Clock-speeds were slightly lower in most cases and it was presumed this was to reduce TDP compared to the standard model.

To my knowledge, the 2019 Mac Pro uses standard W-3100 series Xeon W's. And the previous Mac Pros also used standard Xeon parts for their respective generations/architectures.

 

[Krevnik]

I wonder if it is a case of lack of 5nm wafer capacity at TSMC?

So long as TSMC is hitting expectations of the capacity Apple bought, then I don’t think Apple’s plans would necessarily have to change much unless they decided to repurpose capacity they bought to handle higher than expected demand of the M1 machines. Last year, they said it would take 2 years to transition, so they were already thinking in terms of multiple silicon cycles.

Keep in mind that Apple is still projecting double-digit growth next quarter, and is bullish on iPad and Mac in particular. Even though they also warn of component shortages. That makes me think that the constraint is likely not TSMC failing to hit those initial expectations. What little data is out there suggests that TSMC’s 5nm node is scaling up pretty quick.

 

[thekev]

Intel did manufacture three "B" model Xeon W's for use in the iMac Pro: the W-2140B (8 cores), the W-2150B (10 cores) and the W-2170B (14 cores). Clock-speeds were slightly lower in most cases and it was presumed this was to reduce TDP compared to the standard model.

To my knowledge, the 2019 Mac Pro uses standard W-3100 series Xeon W's. And the previous Mac Pros also used standard Xeon parts for their respective generations/architectures.

Oh right imac pro...

@leman mentioned higher turbo boosts though, and it was the first time I spotted anything about that on the forums.

 

[leman]

Honestly, if they just wanted to ship an "M1X" with minimum changes needed to hit the performance marks to transition, it feels like they would have started production much earlier to me.

This is also what I have been thinking and why my confidence in the M1X shrinks by the hour. And regarding the TSMC production constraints: it seems to me like it would make sense to prepare an improved design for when the production issues are fixed rather than spend half a year sitting on an old old. Apple was always super-aggressive about their roadmap, and I would think that they double their efforts for the Apple Silicon platform.

Did I miss something or was this only applicable to the last mac pro? Apple's SKUs have appeared on Intel's list for every model I have checked. I think there may have been a weird exception around 2008 or something but that's about it, as far as I am aware.

What I mean is that Xeon W "magically" appeared in Intel's roadmap in 2017, few months after the iMac Pro was announced. I am quite certain that Apple (who was unhappy about lower turbo of server Xeons) played a significant role in the introduction of these SKUs. And I don't think there is anything weird with Intel putting them on a public list, it was a good workstation product after all.

 

[leman]

So, this is where language gets tricky, since what you say to me reads as multiple dies on a package with an interconnect between them. That could be right, but it could also be more about making a design that's easier to tile to 2x1 and 2x2 layouts that would be manufactured as a single die. Bloomberg is vague enough here that it's hard to take anything specific away from it.

Both are entirely possible, but hey, this stuff is funky

US20190319626A1 - Systems and methods for implementing a scalable system - Google Patents

Multi-chip systems and structures for modular scaling are described. In some embodiments an interfacing bar is utilized to couple adjacent chips. For example, a communication bar may utilized to coupled logic chips, and memory bar may be utilized to couple multiple memory chips to a logic chip.

patents.google.com

US20210125967A1 - Fully Interconnected Heterogeneous Multi-layer Reconstructed Silicon Device - Google Patents

Reconstructed 3DIC structures and methods of manufacture are described. In an embodiment, one or more dies in each package level of a 3DIC are both functional chips and/or stitching devices for two or more dies in an adjacent package level. Thus, each die can function as a communication bridge...

patents.google.com

And check out this on-package RAM patent... I know it's just abstract schematics, but did they really draw 30+ RAM modules on one of the diagrams?

US10685948B1 - Double side mounted large MCM package with memory channel length reduction - Google Patents

Double side mounted package structures and memory modules incorporating such double side mounted package structures are described in which memory packages are mounted on both sides of a module substrate. A routing substrate is mounted to a bottom side of the module substrate to provide general...

patents.google.com

 

[Andropov]

None of the changes mentioned by Bloomberg indicate a new chip architecture, and they all are consistent with the already reported information by Bloomberg, that M1 variants exist and were planned for Pro products in late 2021.

However the changes for the MacBook Air do in fact indicate a new chip architecture.

Also, common sense indicates that the M2 is not debuting ahead of the A15, nor it is debuting as a Pro product. This simply is not how Apple does things.

If the next Mac processor truly has 2 efficiency cores only, that does seem to point to a different architecture. 'Pro' users would likely have more background tasks than consumers, not less, so having less cores dedicated to them wouldn't make a lot of sense. Specially for MacBooks. More performance cores simply doesn't make up for that.

People have this idea that any position is as valid as any other and that simply is not true. ALL available information to this point indicates 1 direction here with extreme weight in its favor. If you want to argue that the possibility that has less than a 10% chance of being true is more likely, I can't take you seriously.

You're right! Not all opinions are born equal, some are more informed than others. The problem with using that fact in an argument is, well, everyone thinks their opinion is the one that is superior.

This is the ONLY thing that has supported the possibility of new chip architecture, and yet is not guaranteed at all. Apple can (and more than likely is) simply adding fewer efficiency cores alongside twice as many performance cores in the newly designed Pro product that needs power more than it needs efficiency, and also likely has updated battery design to account for it.

Of all your points, you have only offered 1 that is based off any actual information that is available. You also haven't tackled any of the other points that overwhelmingly point away from your conclusion.

The Metal API for raytracing is also a good point. The API was introduced last year, an there's no hardware support yet. The WWDC seems like a good moment to talk about the new hardware support for this API. They could skip that now, and talk about that in september, but then the iPhones would be realeased with no games using the hardware raytracing. It's not the kind of feature you can 'casually' add to an app. And then the M2 MacBook Airs would have it, but not the MacBook Pros... which would be weird. Raytracing can be a very useful tool in many 'Pro' workflows.

Also, common sense indicates that the M2 is not debuting ahead of the A15, nor it is debuting as a Pro product. This simply is not how Apple does things.

Why? We don't know how Apple does things with Apple Silicon processors (for Mac). We have ONE data point. We can't extrapolate anything from that.

 

[EntropyQ3]

If the next Mac processor truly has 2 efficiency cores only, that does seem to point to a different architecture. 'Pro' users would likely have more background tasks than consumers, not less, so having less cores dedicated to them wouldn't make a lot of sense. Specially for MacBooks. More performance cores simply doesn't make up for that.

Or, the stuff that goes on in the background IS the heavy lifting! Dithering around in the user interface isn't that demanding after all. My point is that Apple should have a fair idea how their computers are actually used through collected data, and could let that influence the configuration of the SoC for reasons that are not obvious if you lack said information.

The Metal API for raytracing is also a good point. The API was introduced last year, an there's no hardware support yet. The WWDC seems like a good moment to talk about the new hardware support for this API. They could skip that now, and talk about that in september, but then the iPhones would be realeased with no games using the hardware raytracing. It's not the kind of feature you can 'casually' add to an app. And then the M2 MacBook Airs would have it, but not the MacBook Pros... which would be weird. Raytracing can be a very useful tool in many 'Pro' workflows.

For Apples iOS devices (their by far main business), and consumer Macs, raytracing is a horribly inefficient way to adress reflection corner cases and to produce somewhat better (arguable) ambient occlusion in games. Dedicating silicon area to such ends makes no sense at all for pretty much any of their devices or users.
It could make sense for for machines that actually do the content creation side of rendering. Which could be a justification to have a separate GPU solution for their highest end workstations, if that is a market Apple wants to compete in. That's possible, but it seems like a difficult market for Apple to adress directly - submitting work to a rendering box/farm that returns the finished results to the Mac seems to be a quite competitive, and a lot more flexible.

Why? We don't know how Apple does things with Apple Silicon processors (for Mac). We have ONE data point. We can't extrapolate anything from that.

This is very true.
Johny Srouji did say a "family" of SoCs for the Mac for instance, but whether he intended that over time or at any one instant is opaque.
Apple is already covering their high-end iPad, and all consumer oriented Macs with a single SoC. How many more are they likely to produce for the remaining, much smaller, user base? How will they adress niche use cases? At all? We'll simply have to wait and see. Until Apple has shown their hand, it's a speculation free-for-all.

 

[leman]

Dithering around in the user interface isn't that demanding after all.

You'd be surprised Updating and drawing the UI is a prime example of a demanding burst task that has to be completed as quickly as possible. You can't delegate this to an efficiency core as ever added millisecond means lag and bad usr experience.

For Apples iOS devices (their by far main business), and consumer Macs, raytracing is a horribly inefficient way to adress reflection corner cases and to produce somewhat better (arguable) ambient occlusion in games. Dedicating silicon area to such ends makes no sense at all for pretty much any of their devices or users.

It is horribly inefficient when you look how it has been done so far by Nvidia and AMD. But it doesn't mean it has to be as inefficient on Apple hardware. Apple has large LLC caches, access to IMG RT IP and it is possible that they could build an RT solution that works quite well while retaining low power consumption.

 

[Andropov]

Or, the stuff that goes on in the background IS the heavy lifting! Dithering around in the user interface isn't that demanding after all. My point is that Apple should have a fair idea how their computers are actually used through collected data, and could let that influence the configuration of the SoC for reasons that are not obvious if you lack said information.

No, but other non-system tasks would be better suited for the efficiency cores. Take, for example, Adobe Lightroom building previews for a batch of imported RAW photos. It would make no sense to do that on the performance cores. After all, no one would want to burn through their battery in an hour to build a set of previews that are only used for speeding up the interface. Having less efficiency cores to do things like that doesn't seem right.

I agree that Apple has the data at hand, so they surely have put more thought into this than any of us could. But I would be very surprised if they find out that having less efficiency cores in the 'Pro' Macs is actually beneficial.

For Apples iOS devices (their by far main business), and consumer Macs, raytracing is a horribly inefficient way to adress reflection corner cases and to produce somewhat better (arguable) ambient occlusion in games. Dedicating silicon area to such ends makes no sense at all for pretty much any of their devices or users.
It could make sense for for machines that actually do the content creation side of rendering. Which could be a justification to have a separate GPU solution for their highest end workstations, if that is a market Apple wants to compete in. That's possible, but it seems like a difficult market for Apple to adress directly - submitting work to a rendering box/farm that returns the finished results to the Mac seems to be a quite competitive, and a lot more flexible.

You may be right. I don't know enough about raytracing to have an informed opinion on how efficient or inefficient it has to be, specially on TBDR GPUs. But the API does look like hardware acceleration could be dropped in with little effort required to support them.

About the rendering farms, I think there's a sizeable amount of creators that are not big enough to afford rendering farms but who could make good use of raytracing features for renders. I bet at least the Mac Pro and possibly the top end MacBook Pro will have hardware raytracing support eventually.

We'll see! Looks like a very interesting WWDC ahead.

 

[EntropyQ3]

You'd be surprised Updating and drawing the UI is a prime example of a demanding burst task that has to be completed as quickly as possible. You can't delegate this to an efficiency core as ever added millisecond means lag and bad usr experience.

My point was that a task that starts to take seconds or more and uses all the performance cores is not interactive (unless you sit around watching the paint dry). The mental association of background tasks with low power cores is unfortunate and for computers I’d contend, erroneous. Gaming is pretty much the one exception I can see.

It is horribly inefficient when you look how it has been done so far by Nvidia and AMD. But it doesn't mean it has to be as inefficient on Apple hardware. Apple has large LLC caches, access to IMG RT IP and it is possible that they could build an RT solution that works quite well while retaining low power consumption.

Fair enough.
I don’t know how a modern take on Caustics old stuff that IMG bought would perform. (I checked a couple of years ago where the Caustics folks went because I was curious to see if Apple had recruited them but at the time they seemed to have scattered.)
The problem with dedicated RT silicon is not only theoretical efficiency and cost/benefit but also real world utilization. If the overwhelming majority of users won’t use software that takes advantage of the resource, the die area is simply wasted. And if it’s something that has been a hallmark of Apples SoCs it’s efficiency. You could alternatively use that area to make a beefier GPU in general, increase last level cache size, add a core et cetera that would benefit just about everyone. (Or make less outwardly obvious enhancements to buffers, bla bla bla…)
My position is that focussing on efficiency is a Very Good Thing Indeed.

Edit:fixed quote

 

[EntropyQ3]

Sorry guys, I’m trying to post from my phone and it’s a nightmare when you’re a bear with a very small brain. 😀 I’ll edit once I’m home.

It will be very interesting indeed to see where Apple goes with this, but I fear that whatever Apple reveals next is going to be confined to the laptops, leaving a lot of questions unanswered.

 

[CWallace]

No, but other non-system tasks would be better suited for the efficiency cores. Take, for example, Adobe Lightroom building previews for a batch of imported RAW photos. It would make no sense to do that on the performance cores. After all, no one would want to burn through their battery in an hour to build a set of previews that are only used for speeding up the interface. Having less efficiency cores to do things like that doesn't seem right.

I imagine the performance cores do not operate at only one clock speed - full - but can scale their performance as necessary to the task being asked. I expect that they use more power at lower clock speeds than the efficiency cores do, but I doubt highly that having 2 vs 4 is going to halve the battery life of the machine, otherwise Apple never would have made the trade-off.

Even if it shipped with 10 performance cores and no efficiency cores, the "M1X" 16" MacBook Pro is going to comfortably exceed the 11 hour rated battery life of the Intel model because "M1X" will draw so much less power than the Core CPU and there is the power savings of not having a dGPU. We might very well see a 15-20 hour battery life rating like on the MacBook Air thanks to the larger battery pack and in real-world settings, we might see 100% or more battery life compared to the Intel model even with "only" two efficiency cores.

 

[leman]

My point was that a task that starts to take seconds or more and uses all the performance cores is not interactive (unless you sit around watching the paint dry). The mental association of background tasks with low power cores is unfortunate and for computers I’d contend, erroneous. Gaming is pretty much the one exception I can see.

I think the question is what exactly you define as “background task”. I am closely following Apples QoS terminology where a ba ground task is a low-priority service task that can take a very long time and does not need to be completed quickly (or in some cases, at all). Non-interactive work that needs to be done ASAP is not a background task.

Fair enough.
I don’t know how a modern take on Caustics old stuff that IMG bought would perform. (I checked a couple of years ago where the Caustics folks went because I was curious to see if Apple had recruited them but at the time they seemed to have scattered.)
The problem with dedicated RT silicon is not only theoretical efficiency and cost/benefit but also real world utilization. If the overwhelming majority of users won’t use software that takes advantage of the resource, the die area is simply wasted. And if it’s something that has been a hallmark of Apples SoCs it’s efficiency. You could alternatively use that area to make a beefier GPU in general, increase last level cache size, add a core et cetera that would benefit just about everyone. (Or make less outwardly obvious enhancements to buffers, bla bla bla…)
My position is that focussing on efficiency is a Very Good Thing Indeed.

I think they will probably add special instructions for bounding volume intersection/traversal, so that will just become a part of the usual compute work.

 

[Andropov]

I imagine the performance cores do not operate at only one clock speed - full - but can scale their performance as necessary to the task being asked. I expect that they use more power at lower clock speeds than the efficiency cores do, but I doubt highly that having 2 vs 4 is going to halve the battery life of the machine, otherwise Apple never would have made the trade-off.

Even if it shipped with 10 performance cores and no efficiency cores, the "M1X" 16" MacBook Pro is going to comfortably exceed the 11 hour rated battery life of the Intel model because "M1X" will draw so much less power than the Core CPU and there is the power savings of not having a dGPU. We might very well see a 15-20 hour battery life rating like on the MacBook Air thanks to the larger battery pack and in real-world settings, we might see 100% or more battery life compared to the Intel model even with "only" two efficiency cores.

I agree that Apple wouldn't make that tradeoff. That's why I think the new 8+2 SoC has a new core design

As of now, Apple seems to be confining tasks with a low QoS to run in the efficiency cores only. If the new Pro Macs have only 2 efficiency cores identical to the ones in the M1, three obvious possibilities appear:

• Apple relaxes the requirement that all low-QoS tasks must run in the efficiency cores, switching to the performance cores if needed. This will result in worse battery life and performance per watt, even with performance cores keeping a low clock frequency. I don't know how much worse that would be, but Apple uses efficiency cores for a reason.
• Apple continues enforcing the requirement that all low-QoS tasks must run in the efficiency cores. That would leave the new Pro Macs taking longer to complete some background tasks than their M1 siblings.
• Apple presents a new core design, where 2 of the new efficiency cores are as fast as 4 of the old efficiency cores.

My bet is on the third option. I guess option 1 is possible, but why would Apple do that? The efficiency cores don't take a lot of die space. Why wouldn't they be able to squeeze two additional efficiency cores to match the M1? Icestorm cores are tiny!

And I agree that, whichever route Apple takes, the battery will last much longer than the current Intel model, no doubt.

 

[EntropyQ3]

I think the question is what exactly you define as “background task”. I am closely following Apples QoS terminology where a ba ground task is a low-priority service task that can take a very long time and does not need to be completed quickly (or in some cases, at all). Non-interactive work that needs to be done ASAP is not a background task.

Ah. The stringence of that definition makes a huge difference. You pretty much need a programming background to appreciate it though. 🙂

I think they will probably add special instructions for bounding volume intersection/traversal, so that will just become a part of the usual compute work.

Could well be. We’ll see.
For better or worse, Apple is in the process of unifying their architecture across iOS and MacOS devices. And they are their own eco-system software wise, they have little reason to chase the latest buzzwords used to push upgrades on the PC side. I’d assume that would mean that their GPU architecture will evolve according primarily to the needs of their phones. I largely see that as positive, it means that their customers can enjoy very decent performance in for instance light fanless laptops with very long battery life.
Also, it would seem that it would encourage them to be frugal with bandwidth and RAM, helped by for instance further improvements to compression and upscaling, which would also clearly benefit the MacOS (unified memory) products.
That doesn’t rule out that some Mac GPUs could contain enhancements that would be more relevant for that niche, particularly if the functionality was accessed through the same software interface. But my crystal ball is completely opaque when it comes to whether Apple feels that is a good path to take, and I can’t really judge the cost/effort/benefit.

 

[dieselm]

This will probably be a similar situation to the iPad Pros (2020) and iPad Airs. The Air's single core performance beat the Pros, while the multi-core and GPU performance beat the Airs.

Doesn't hold back the Pro buyers at all. They need the 75%+ greater multi-core performance and will pay for it.

In a fall M2 Air vs M1X Pro situation, they'll have an unbeatable "highest performance" low(er)-cost laptop in the market. At the same time, Macbook Pro buyers will have have an unbeatable 75%+ higher multi-core, w/2-4X GPU performance with 32/64GB memory option at twice the price, with the M1X.