Claim - New 12 CPU Core M1 SoC for iMac Pro

[arvinsim]

Mark Gurman leaked the entire M1 roadmap back in May 2021, though at the time he only knew the codenames, core counts and memory capacities:

• JadeC-Chop, which became M1 Pro;
• JadeC-Die, which became M1 Max;
• Jade2C-Die, which became M1 Ultra;
• Jade4C-Die, which will be two M1 Ultras in a "chiplet" or some other package.

Apple Readies MacBook Pro, MacBook Air Revamps

Apple Inc. is preparing to release several new Mac laptops and desktops with faster processors, new designs and improved connectivity to external devices, accelerating the company’s effort to replace Intel Corp. chips and leapfrog rival PC makers.

www.bloomberg.com

I guess we can safely assume that Jade4C-Die is the Mac Pro.

The question now is if it supports expandability.

 

[DeepIn2U]

I guess we can safely assume that Jade4C-Die is IN Mac Pro.

The question now is if it supports expandability.

is 'IN' but Apple SVP of hardware Ternus already stated the last chip in the M1 lineup is the M1 Ultra. Does this mean a 2022 Mac Pro gets 2 M1 Ultra's put together with their unique interconnect FabricFusion or whatever it's called?

More importantly with the Mac Studio are all TB4/USB-C4 connections have their very own TB4 controller or is bandwidth been split?

 

[arvinsim]

More importantly with the Mac Studio are all TB4/USB-C4 connections have their very own TB4 controller or is bandwidth been split?

That got me thinking. Maybe Apple is banking on TB4 to be the expandability option in the Apple Silicon Mac Pro.

 

[CWallace]

Apple SVP of hardware Ternus already stated the last chip in the M1 lineup is the M1 Ultra. Does this mean a 2022 Mac Pro gets 2 M1 Ultra's put together with their unique interconnect FabricFusion or whatever it's called?

That is my belief.

 

[crazy dave]

is 'IN' but Apple SVP of hardware Ternus already stated the last chip in the M1 lineup is the M1 Ultra. Does this mean a 2022 Mac Pro gets 2 M1 Ultra's put together with their unique interconnect FabricFusion or whatever it's called?

That is my belief.

Unlikely. As you yourself noted, Ternius said that the M1 Ultra is the last M1 chip and it uses the interconnect for 2 M1 Maxes. Putting two M1 Ultras together for a new M1 chip would make that statement false. Further, the current M1 Max cannot be used in a x4 configuration (irq supports only 2 dies and there is only one connector per die) and is probably unsuitable for a workstation.

If there’s to be a special workstation chip with Firestorms and Icestorms in 1x, 2x, 4x configurations then it’s based on a die we haven’t seen yet and the base die won’t be called M1 something.

 

[CWallace]

Unlikely. As you yourself noted, Ternius said that the M1 Ultra is the last M1 chip and it uses the interconnect for 2 M1 Maxes. Putting two M1 Ultras together for a new M1 chip would make that statement false. Further, the current M1 Max cannot be used in a x4 configuration and is probably unsuitable for a workstation.

Apple could combine four M1 Max into a chiplet via a four-way hub using the built-in connectors. That would not contradict Ternius' statement and it would align with Gurman's "Jade4C-Die" codename just as "Jade2C-Die" turned out to be M1 Ultra.

 

[crazy dave]

Apple could combine four M1 Max into a chiplet via a four-way hub using the built-in connectors. That would not contradict Ternius' statement and it would align with Gurman's "Jade4C-Die" codename just as "Jade2C-Die" turned out to be M1 Ultra.

If that logic held then the M1 Max is the last M1 chip, not the M1 Ultra, as the M1 Ultra is simply two M1 Max chiplets connected directly to each other. You can't have it both ways. And, again, the M1 Max does not appear to be set up for a x4 chiplet configuration - only a x2 (the interrupt controller needs to know how many dies there are and it is only setup to be aware of two dies total). Further, Apple was fairly dismissive of IO dies in their presentation and it's not clear that a full IO die could allow the 4 GPUs to operate as a single logical GPU as the two GPUs do in the M1 Ultra. I imagine that it's hard enough with the Ultrafusion bridge. Maybe they won't want that anyway. Then there are the issues with PCIe and RAM. Maybe he was being cute and they've figured a way around all this, but it doesn't seem likely.

 

[CWallace]

If that logic held then the M1 Max is the last M1 chip, not the M1 Ultra, as the M1 Ultra is simply two M1 Max chiplets connected directly to each other.

Well then maybe Apple could not get "Jade4C-Die" to work and had to scrap it. William Ma has claimed that Apple will have a version of M3 with 40 CPU cores that will release in late 2023. And now the pundits are claiming Mac Pro will not ship until late 2023, so maybe we will have to wait for this 40 core M3 in 2023.

If true, maybe Apple "rushed" Apple Studio out yesterday because they are not in a position to announce an Apple silicon Mac Pro at WWDC in June when everyone is expecting them to.

Perhaps Tim will use the keynote to talk up M1 Ultra and Mac Studio and note that "Mac Pro continues to be a product in our line-up" just as he did with the Mac mini in 2017 prior to the new model dropping the following year.

 

[JPack]

Well then maybe Apple could not get "Jade4C-Die" to work and had to scrap it. And since M2 is already taped-out, they can't use it, either.

That means they have to use M3 for the Mac Pro and William Ma has claimed that Apple will have a version of M3 with 40 CPU cores that will release in late 2023. And now everyone says Mac Pro will not ship until late 2023, so maybe it has to wait for this 40 core M3.

So maybe Apple "rushed" Apple Studio out yesterday because they are not in a position to announce an Apple silicon Mac Pro at WWDC in June when everyone is expecting them to.

Instead, Tim will use the keynote to talk up M1 Ultra and Mac Studio and note that "Mac Pro continues to be a product in our line-up" just as he did with the Mac mini in 2017 prior to the new model dropping the following year. Maybe Apple will announce an update with Alder Lake Xeon W CPUs at WWDC since M1 Ultra crushes the current Skylake Xeons in a number of tasks or as with Mac mini, they'll just run silent and everyone will speculate until the hard leaks detailing the M3 Apple silicon Mac Pro arrive in the weeks before WWDC 2023.

If Mac Pro weren't close being ready, Apple wouldn't mention it on stage. After the AirPower fiasco, I seriously doubt Apple would even hint at Mac Pro if they weren't confident Jade 4C is on track.

All these theories rely on looking at the M1 Max die. That's like looking at the base M1 and predicting a dual-die configuration wouldn't be possible. In all likelihood, we haven't seen all M1 derivatives yet.

 

[DeepIn2U]

Well then maybe Apple could not get "Jade4C-Die" to work and had to scrap it. William Ma has claimed that Apple will have a version of M3 with 40 CPU cores that will release in late 2023. And now the pundits are claiming Mac Pro will not ship until late 2023, so maybe we will have to wait for this 40 core M3 in 2023.

If true, maybe Apple "rushed" Apple Studio out yesterday because they are not in a position to announce an Apple silicon Mac Pro at WWDC in June when everyone is expecting them to.

Perhaps Tim will use the keynote to talk up M1 Ultra and Mac Studio and note that "Mac Pro continues to be a product in our line-up" just as he did with the Mac mini in 2017 prior to the new model dropping the following year.

Interesting thoughts.

It’s possible Apple rushed on their 2yr transition statement being too bold to do it in the same time that Jobs once promised with Intel transition. Apple did give the iMac Pro and it definitely was a stop-gap for them publicly admitted until the Mac Pro was ready. Then again Ternus could’ve just been cheeky, just because.

This Studio really feels like a tribute to the Cube.

 

[jdb8167]

If Mac Pro weren't close being ready, Apple wouldn't mention it on stage. After the AirPower fiasco, I seriously doubt Apple would even hint at Mac Pro if they weren't confident Jade 4C is on track.

All these theories rely on looking at the M1 Max die. That's like looking at the base M1 and predicting a dual-die configuration wouldn't be possible. In all likelihood, we haven't seen all M1 derivatives yet.

This is certainly true. Either Apple is going to leverage the M1 Ultra for the Mac Pro or they are going to go with a new SoC/CPU. John Ternus was very clear, that the M1 Ultra was the last M1 they were making and that the Mac Pro was the last Mac to update for the ASi transition.

Either a multi M1 Ultra Mac Pro or a new chip. I think it is possible that Apple is planning a Pro CPU that removes the GPU and media engines and moves them to a separate chip/board. Make the GPU/Media engine a PCIe 5/CXL plug-in boards for modularity. Add as many as you need. Also with slotted DDR5 RAM DIMMS. This might be wishful thinking on my part but the fact is that Apple is still planning a Intel Mac Pro replacement and they have to solve some of these problems to reach the goal of replacing the Intel Mac Pro.

 

[crazy dave]

Well then maybe Apple could not get "Jade4C-Die" to work and had to scrap it. William Ma has claimed that Apple will have a version of M3 with 40 CPU cores that will release in late 2023. And now the pundits are claiming Mac Pro will not ship until late 2023, so maybe we will have to wait for this 40 core M3 in 2023.

If true, maybe Apple "rushed" Apple Studio out yesterday because they are not in a position to announce an Apple silicon Mac Pro at WWDC in June when everyone is expecting them to.

Perhaps Tim will use the keynote to talk up M1 Ultra and Mac Studio and note that "Mac Pro continues to be a product in our line-up" just as he did with the Mac mini in 2017 prior to the new model dropping the following year.

I'll be honest, I have no idea anymore what the Mac Pro will look like at this point. The only thing that seems unlikely is that it will be based on the M1 Max. Ternius (I think) did seem to have a cheshire cat grin on his face with the Mac Pro and the future. So they probably have something up their sleeve it just isn't clear what that is ... at least to me.

If Mac Pro weren't close being ready, Apple wouldn't mention it on stage. After the AirPower fiasco, I seriously doubt Apple would even hint at Mac Pro if they weren't confident Jade 4C is on track.

All these theories rely on looking at the M1 Max die. That's like looking at the base M1 and predicting a dual-die configuration wouldn't be possible. In all likelihood, we haven't seen all M1 derivatives yet.

I dunno they did it with the last Mac Pro and currently all I'll say is that Jade 4C isn't the M1 Max/Ultra. If it exists, it is based on something new.

This is certainly true. Either Apple is going to leverage the M1 Ultra for the Mac Pro or they are going to go with a new SoC/CPU. John Ternus was very clear, that the M1 Ultra was the last M1 they were making and that the Mac Pro was the last Mac to update for the ASi transition.

Either a multi M1 Ultra Mac Pro or a new chip. I think it is possible that Apple is planning a Pro CPU that removes the GPU and media engines and moves them to a separate chip/board. Make the GPU/Media engine a PCIe 5/CXL plug-in boards for modularity. Add as many as you need. Also with slotted DDR5 RAM DIMMS. This might be wishful thinking on my part but the fact is that Apple is still planning a Intel Mac Pro replacement and they have to solve some of these problems to reach the goal of replacing the Intel Mac Pro.

The M1 Ultra in a x2 or M1 Max in x4 seems unlikely for all the reasons listed above. Something new cannot be ruled out.

 

[JPack]

This is certainly true. Either Apple is going to leverage the M1 Ultra for the Mac Pro or they are going to go with a new SoC/CPU. John Ternus was very clear, that the M1 Ultra was the last M1 they were making and that the Mac Pro was the last Mac to update for the ASi transition.

Either a multi M1 Ultra Mac Pro or a new chip. I think it is possible that Apple is planning a Pro CPU that removes the GPU and media engines and moves them to a separate chip/board. Make the GPU/Media engine a PCIe 5/CXL plug-in boards for modularity. Add as many as you need. Also with slotted DDR5 RAM DIMMS. This might be wishful thinking on my part but the fact is that Apple is still planning a Intel Mac Pro replacement and they have to solve some of these problems to reach the goal of replacing the Intel Mac Pro.

As I mentioned elsewhere, the biggest question is, what is the definition of M1?

If Apple designs a chip with zero E-cores and only P-cores, but continues to use A14 Firestorm cores, is that considered an M1 family chip? If Apple updates the memory controller to drive user-replaceable DIMMs, is that still an M1 family chip? Or would Apple call it a W1 chip? We simply don't know.

His statement about the last chip in the M1 family doesn't mean much unless we know what is M1.

 

[oz_rkie]

Apple could combine four M1 Max into a chiplet via a four-way hub using the built-in connectors. That would not contradict Ternius' statement and it would align with Gurman's "Jade4C-Die" codename just as "Jade2C-Die" turned out to be M1 Ultra.

There is one interconnect per m1 max chip. Thus the max that can be connected via the interconnect is 2. How would they have a 4 way connection?

 

[jdb8167]

As I mentioned elsewhere, the biggest question is, what is the definition of M1?

Anything Apple wants the definition to be. They can come out with a chip they label P1 with the same FireStorm and IceStorm cores with a completely different layout that emphasizes CPU cores and moves the GPU cores off die and call it a completely different family.

 

[deconstruct60]

is 'IN' but Apple SVP of hardware Ternus already stated the last chip in the M1 lineup is the M1 Ultra. Does this mean a 2022 Mac Pro gets 2 M1 Ultra's put together with their unique interconnect FabricFusion or whatever it's called?

You are making a huge assumption that it actually is a 2022 Mac Pro. Also even if there is a Mac Pro 2022 that it gets a M1 package in it. Could be an M2/M3. So the Studio could be one of the last systems. Ternus said a later date. There was zero boundary set on when that date would be. ( last two MP only got "demos" 6 months in advance. Could have a 2022 "sneak peak" at a 2023 product. )

Same issue with the iMac Pro. If sliding into next year. M2/M3

If just use a M1 Ultra then the general purpose I/O on it will be relatively lame. Making something almost as limited as the Studio I/O already is would not be much of a differentiation .

Apple needs a substantively different package and/or set of dies to do a proper Mac Pro. If M1 family is done, then seems more likely that Mac Pro is waiting on generation 2 or 3 for a solution.

More importantly with the Mac Studio are all TB4/USB-C4 connections have their very own TB4 controller or is bandwidth been split?

Not all of the USB-C. The M1 Max Studio has two front USB-C. The M1 Ultra Studio has two front Thunderbolt. That second twin die has "extra" ones that the Max package doesn't have.

Where it is Thunderbolt then it is controllers per port. Pragmatically to get TB4 certification out of Intel have to do it that way. Otherwise doesn't really qualify (like a 3 or lower would ).

 

[quarkysg]

Anything Apple wants the definition to be. They can come out with a chip they label P1 with the same FireStorm and IceStorm cores with a completely different layout that emphasizes CPU cores and moves the GPU cores off die and call it a completely different family.

I would think the Mac Pro will still be Mx based SoCs and thus named as such.

I would also think that it will have ECC DIMM slots. It will take a hit for latency and power efficiency, but in turn will get tons of unified memory for the SoC. dGPU via PCIe will be out, IMHO, as it doesn't gel with their SoC UMA design.

 

[DeepIn2U]

You are making a huge assumption that it actually is a 2022 Mac Pro. Also even if there is a Mac Pro 2022 that it gets a M1 package in it. Could be an M2/M3. So the Studio could be one of the last systems. Ternus said a later date. There was zero boundary set on when that date would be. ( last two MP only got "demos" 6 months in advance. Could have a 2022 "sneak peak" at a 2023 product. )

Same issue with the iMac Pro. If sliding into next year. M2/M3

If just use a M1 Ultra then the general purpose I/O on it will be relatively lame. Making something almost as limited as the Studio I/O already is would not be much of a differentiation .

Apple needs a substantively different package and/or set of dies to do a proper Mac Pro. If M1 family is done, then seems more likely that Mac Pro is waiting on generation 2 or 3 for a solution.




Not all of the USB-C. The M1 Max Studio has two front USB-C. The M1 Ultra Studio has two front Thunderbolt. That second twin die has "extra" ones that the Max package doesn't have.

Where it is Thunderbolt then it is controllers per port. Pragmatically to get TB4 certification out of Intel have to do it that way. Otherwise doesn't really qualify (like a 3 or lower would ).

The timeline I was going by for Mac Pro being in 2022 release is due to Tim announcing 2 yr transition to Apple Silicon which was done on June 22, 2020.

 

[CWallace]

There is one interconnect per m1 max chip. Thus the max that can be connected via the interconnect is 2. How would they have a 4 way connection?

One option would be to have hub could have the same interconnect on each of its four sides. So the four Max connect to the four interconnects on the hub. The hub would then have logic necessary to allow allow the four Max to share everything between them.

Another option would be to have two M1 Ultra placed side-by-side with the interconnect extended between them via core logic.

 

[deconstruct60]

There is one interconnect per m1 max chip. Thus the max that can be connected via the interconnect is 2. How would they have a 4 way connection?

The interposer pragmatically completes the connection. What is being proposed in some how the interposer splits up the bandwidth between the four ( instead of one target would need to go three ways in/out for each die. Take 10,000 connections and conceptually divide by 4. ) It is quite doubt that is just 5,000 pads "in" pipe and 5000 pad "out" pipe. Probably multiple channels bound up for the aggregate bandwidth number. Conceptually could send some of those channels off into a different die, What they would not have though would be a 2TB/s connection between the individual die pairs anymore. So probably could not do a "present GPU cores as one single GPU" solution.

As soon as the fractional inter die connection bandwidth allocation drops below the memory bandwidth you don't have a 'fat tree' network anymore.


Pretty good chance that Jade4C got dropped. It isn't coming. The next generation will have a revised/updated interface that is incrementally faster. That could allow Apple perhaps to "make do" with a 'fused' GPU across two dies (with lower GPU bandwidth pressure; e.g., 16 instead of 32 GPU cores) and the last two in the set are presented separately ( perhaps more as "compute" GPUs). Probably would need some GPU/GPGPU changes to go along with this but some foundation is already there with the Metal support for Infinity Fabric 'back channel' memory communication.

What Apple has done here likely isn't going to scale out as well. They take up an entire edge of the die to hook just two dies together. If needed a shortest distance connection to two more dies there aren't two more edges of the die to use up for that interconnect. What Apple sacrificed here that the other GPU vendors didn't want to do is any other kind of external connection from the die ( e.g. x16 PCI-e v4/5 ) . That is a dual edge sword when need to talk to more than just one neighbor. Four dies is probably " A Bridge too Far" ( at least for generation 1 of the family).

 

[crazy dave]

One option would be to have hub could have the same interconnect on each of its four sides. So the four Max connect to the four interconnects on the hub. The hub would then have logic necessary to allow allow the four Max to share everything between them.

Another option would be to have two M1 Ultra placed side-by-side with the interconnect extended between them via core logic.

View attachment 1970773

View attachment 1970765

View attachment 1970771

Yes theoretically they could’ve done it this way with just a single connection. But as @deconstruct60 and I pointed out it would likely necessitate a loss of their ability to present a unified GPU to the system … if they care about that for the Mac Pro product stack (and of course the lacking interrupt controller).

 

[deconstruct60]

One option would be to have hub could have the same interconnect on each of its four sides. So the four Max connect to the four interconnects on the hub. The hub would then have logic necessary to allow allow the four Max to share everything between them.


View attachment 1970773

View attachment 1970765

View attachment 1970771

That I/O die hub is going to present as a NUMA interconnect. Going to be substantively harder to present as "One GPU". What AMD does with their I/O die is make all the cores accesses to memory slower equally. Apple can't really afford that with their GPU , display output processor(s) , and NPU bandwidth demands.

Apple uses an interposer so the "interfacing" die is layered beneath the M1 Max dies. It would have to be a pretty big interposer to lie underneath all four. Second problem is would have to fixed amount of bandwidth that would have to be divided up between the additional two dies to make it all completely seamless in all GPU latency contexts.

It is one thing to make it on uniformly addressable. It is another to smooth out all the latency hiccups that might spring up. Unless UltraFusion is only using 1/2 or 1/3 of its bandwidth to make two Max dies appear uniform, there is a scaling problem.

 

[deconstruct60]

The timeline I was going by for Mac Pro being in 2022 release is due to Tim announcing 2 yr transition to Apple Silicon which was done on June 22, 2020.

He never said that. He said "about two years". Apple is never going to put a promise out there for an exact month, on a exact date far off into the future. For 2013 Mac Pro it was. "Fall 2013" ... which turned out to be the literally the last day of Fall before Winter officially starts in December. In 2017, it was. "later" . In 2018, it was "not this year". They never said "December 2019" 1-2 years in advance.

Apple knows "crap happens". And sure as the sky is blue , Cook knew that "crap happens" back in June 2020 when presenting to a video prompter because everyone was on lockdown due to Covid.

Cook gave Apple enough wiggle room to finish in 2022 without pinning it to any specific month, let alone date.

If in 2020 Apple pinned the Mac Pro to a TSMC N3 die then it is in trouble of delivering in 2022. Probably sliding to 2023. If were pinning hopes on Jade4C that appears to have been a 'miss' (another reason Apple doesn't talk about future products so that can bury the early revision that doesn't work out according to early plans ). The MBP 14"/16" and Studio Display ... etc all have signs of rolling out on blown schedules.

Pretty good chance Apple has drifted off the script they created back in early 2019 for this rollout.

 

[deconstruct60]

I would also think that it will have ECC DIMM slots. It will take a hit for latency and power efficiency, but in turn will get tons of unified memory for the SoC. dGPU via PCIe will be out, IMHO, as it doesn't gel with their SoC UMA design.

RAM capacity isn't the main key to making Apple's GPU work so well. (they are not shunning it, but it isn't the main key) The latency hit would present a substantive problems. One of the major factors mentions for their UltraFusion interconnect was very efficient power utilization and lowest possible minimal latencies. They create an interconnect like that for the dies and then piss it out on DIMMs? That doesn't sound like anything they have done on M-series at all.

There are no high performance dGPUs out there that use DIMMs for memory. There are very good practical reasons for that. Relatively lower end iGPUs have used it, but this are at no where near Apple's performance levels.

Apple isn't just simply using LPDDR5 for their memory. They have pragmatically composed a "poor man's" HBM like solution out of it at lower costs and low enough power consumption to make them happy. Pretty doubtful they are going to give that up for the Mac Pro. The capacity bumps are somewhat a byproduct of going "super wide" on the memory channels. Need to be attached to more RAM dies concurrently and that also tends to make the capacity go up.

And as long as they are primarily hooked on LPDDR5/6 solutions then ECC is probably pragmatically out. ECC and LPDDR data bus doesn't work like DIMMs. Full fledged ECC would mean a bus bandwidth hit. Apple isn't going to take that. And they would have to be a substantively different memory controller to switch off of LPDDR to DIMMs. (again unlikely when the Ultra is literally the minimal effort to use a different die. )

In Apple's presentation about the M1 Ultra they mentioned that the. 128GB RAM was "enormous". They aren't really going to say that it is bad, but they are normalizing it again VRAM capacities and probably looking at what RAM sizes people use ( or perhaps mistakenly about RAM sizes people buy from Apple ). Not only dGPUs are out, but likely large triple or quad digit GB RAM capacities are out also. When get too large then error tracking is more of an issue and run into conflicts with their other design parameters.

 

[quarkysg]

They create an interconnect like that for the dies and then piss it out on DIMMs? That doesn't sound like anything they have done on M-series at all.

Actually, my thinking is that the UltraFusion fabric is mainly used for cache coherency between the two M1 Max dies.

Connection to the memory modules are still handled by each die's LPDDR5 memory controllers, where each M1 Max have 4 of those. With 8 memory controllers, that's where they get the 800GB/s bandwidth. The UltraFusion fabric does not handle the memory modules at all.

My thinking is that for the Mac Pro, they will likely glue 4 of those M2 Max SoCs together (with the same design as M1 Max) with their next gen UltraFusion fabric and take a hit on latency, bandwidth and power efficiency. So maybe 600GB/s per M2 Max die. With four of them, they will get 2.4TB/s of memory bandwidth, assuming all memory slots are filled. They likely will try to hide latency with larger SLC.

Imagine 1.5TB or more of high thruput memory that's available for use to the CPU, GPU, NE, and the other IP cores.

Apple isn't just simply using LPDDR5 for their memory. They have pragmatically composed a "poor man's" HBM like solution out of it at lower costs and low enough power consumption to make them happy. Pretty doubtful they are going to give that up for the Mac Pro. The capacity bumps are somewhat a byproduct of going "super wide" on the memory channels. Need to be attached to more RAM dies concurrently and that also tends to make the capacity go up.

Which is why I think they will continue to use LPDDR5. It is a practical solution for them. Maybe they are planning to use LPDDR5X for the Mac Pro for all we know, for even more bandwidth.

And as long as they are primarily hooked on LPDDR5/6 solutions then ECC is probably pragmatically out. ECC and LPDDR data bus doesn't work like DIMMs. Full fledged ECC would mean a bus bandwidth hit. Apple isn't going to take that. And they would have to be a substantively different memory controller to switch off of LPDDR to DIMMs. (again unlikely when the Ultra is literally the minimal effort to use a different die. )

I remember reading somewhere that LPDDR5 support link ECC, and LPDDR5 mandates on-chip ECC.