Rumored pro chip

[Falhófnir]

I always saw them using 3 classes of chips. The M1X would be both the 16 and 32 core variants. You’d just specify which variant you want in checkout but Apple would always display “up to 32 GPU cores” in all their marketing. Just how they do with the 7 vs 8 on M1. The M1Z or M1Pro (whatever) would be an even higher-end chip likely for the Mac Pro and also have a few variants of its own.

Though by then it may be on the next gen architecture so M2X or M3 who knows.

Totally could be for sure. That being said, you’d also think they would tag M2 as well. AFAIK Apple hasn’t really done SEO for rumored products before though. I could be wrong.

Edit: some say it’s been removed. Hmm ?

Yeah, M1, M1X, M1 Pro was pretty much my initial thoughts, it's just now I feel like if they want to charge more/less based on binning or activating/ deactivating cores (which they might to make maximum use of each wafer) they could end up with too many chip SKUs to fit into that pattern. Unless they use it like Intel and AMD use 'U' and 'H' - M1 indicates a 10-15W TDP, M1X a 25-45W TDP and M1 Pro a 60W+ TDP?

I'd say being removed makes it a bit suspicious, though it seems a bit unprofessional to reference either an unreleased product or a colloquial term for an unreleased product lol!

 

[cmaier]

Why would Apple troll people?

Sosumi - Wikipedia

en.wikipedia.org

https://en.wikipedia.org/wiki/Power_Macintosh_7100 (search for BHA)

Apple is *fantastic* at trolling.

 

[senttoschool]

M1X (16 core GPU) and M1Z (32 core GPU) has a nice ring to it, but I think it limits them in terms of adding more options down the line if future chip generations offer more options?

I really don't think they will do this.

First, there will be a lot more binning options on Macs. (4 core, 6 core, 8 core, 10 core, and all the way up to 32 cores according to Bloomberg). There are no SoC options for iPhones and iPads.

Second, there will be binning options for GPUs as well.

Lastly, these letters mean nothing to 99% of customers. But if you tell how many cores there are, at least they know that the more cores there are, the better usually.

So I think Apple will market their SoCs like:

M2 4-Core Processor. 8-Core GPU.
M2 10-Core Procesor. 32-Core GPU.

P for "Pro" line.

 

[GubbyMan]

I hope the next MacBook Pros will have something newer than the firestorm/icestorm. It would be slightly underwhelming to see an improved architecture for A15 in iPhone 13 and at the same time get performance focused MacBooks based on last year's A14.

 

[CWallace]

We could / should end up with six options:

• 8 CPU (4P+4E) and 7 GPUs ("Tonga")
• 8 CPU (4P+4E) and 8 GPUs ("Tonga")
• 10 CPU (8P+2E) and 16 GPUs ("JadeC-Chop")
• 10 CPU (8P+2E) and 32 GPUs ("JadeC")
• 20 CPU (16P+4E) and 64 GPUs ("Jade2C")
• 40 CPU (32P+8E) and 128 GPUs ("Jade 4C")

 

[CWallace]

I hope the next MacBook Pros will have something newer than the firestorm/icestorm. It would be slightly underwhelming to see an improved architecture for A15 in iPhone 13 and at the same time get performance focused MacBooks based on last year's A14.

It will only be underwhelming on workflows that primarily access a single Efficiency or Performance core.

If your workflows access more than four Performance cores or more than eight GPU cores, the A14-based "M1X" MacBook Pro will be significantly faster and an A15-based "M2".

 

[GubbyMan]

Since it looks like Jade2C and Jade4C will have on chip GPU, are we expecting a future Mac Pro to have everything on SoC, no dedicated "Lifuka" GPU?

 

[GubbyMan]

It will only be underwhelming on workflows that primarily access a single Efficiency or Performance core.

If your workflows access more than four Performance cores or more than eight GPU cores, the A14-based "M1X" MacBook Pro will be significantly faster and an A15-based "M2".

Yeah, that's why I wrote slightly underwhelming. It would still be a beast of a chip no doubt.

 

[CWallace]

Since it looks like Jade2C and Jade4C will have on chip GPU, are we expecting a future Mac Pro to have everything on SoC, no dedicated "Lifuka" GPU?

Unsure. The speculation was that it would be an independent GPU based on the October 2020 China Times report that "Apple will launch its first self-developed GPU under the research and development code name Lifuka next year".

It could be that "Jade C" will use a different on-die GPU than what we have seen in M1 (which is the same GPU as used in the A14) and this newer GPU design is "Lifuka". So it might very well either be more powerful than the GPUs in the M1 or it might support more features that "pro workflows" need/desire and that are found in nVidia and AMD GPUs, but not the GPUs in the A series SoCs.

It is also quite possible "Lifuka" was nothing more than the codename for the A14's GPU (as also used in the M1 and possibly being used in the "M1X") as Lifuka is the name of an island in the Kingdom of Tonga and "Tonga" was the claimed codename for what became the M1.

 

[AgentMcGeek]

It doesn't make sense to me that Apple would move its GPU out of the SoC. You'd loose all the advantages of Unified Memory and low latency. It's a Mac Pro, nothing's stopping them from making a huge chip, apart from cost / wafer limitations. Since they're in control of the whole design, they can adapt the motherboard and cooling solution.

 

[reallynotnick]

We could / should end up with six options:

• 8 CPU (4P+4E) and 7 GPUs ("Tonga")
• 8 CPU (4P+4E) and 8 GPUs ("Tonga")
• 10 CPU (8P+2E) and 16 GPUs ("JadeC-Chop")
• 10 CPU (8P+2E) and 32 GPUs ("JadeC")
• 20 CPU (16P+4E) and 64 GPUs ("Jade2C")
• 40 CPU (32P+8E) and 128 GPUs ("Jade 4C")

If you are going to count binned versions there likely will be more than 6 versions. Apple is already binning the M1, and larger chips are even more likely to get defects that could benefit from binning. So I'd expect to at least get something like 14 and 28 GPU core versions for JadeC-Chop and JadeC respectfully.

 

[cmaier]

It doesn't make sense to me that Apple would move its GPU out of the SoC. You'd loose all the advantages of Unified Memory and low latency. It's a Mac Pro, nothing's stopping them from making a huge chip, apart from cost / wafer limitations. Since they're in control of the whole design, they can adapt the motherboard and cooling solution.

No you wouldn’t. It can be in the same package but on a different die.

 

[AgentMcGeek]

Aaah yes. I assumed they meant out of package.

 

[CWallace]

If you are going to count binned versions there likely will be more than 6 versions. Apple is already binning the M1, and larger chips are even more likely to get defects that could benefit from binning. So I'd expect to at least get something like 14 and 28 GPU core versions for JadeC-Chop and JadeC respectfully.

True. Guess it depends on how many options Apple wants to offer with their product lineup, though it we get to a dozen (or more) configurations, it could make things messy.

Personally, I think Apple might want to simplify and that the 7 GPU option could be dropped with M2 as yields could be good enough to ensure enough "full" SoCs pass QA to meet supply and then we would have the 8/8, 10/16, 10/32, 20/64 and 40/128 options.

 

[Apple Knowledge Navigator]

It doesn't make sense to me that Apple would move its GPU out of the SoC. You'd loose all the advantages of Unified Memory and low latency. It's a Mac Pro, nothing's stopping them from making a huge chip, apart from cost / wafer limitations. Since they're in control of the whole design, they can adapt the motherboard and cooling solution.

I'd say it's much too early too call. Surely the increased speeds would offset the placement of the GPU?

 

[Sydde]

No you wouldn’t. It can be in the same package but on a different die.

Or PoP?

 

[leman]

I'd say it's much too early too call. Surely the increased speeds would offset the placement of the GPU?

You seem to assume that separating a GPU in its own, isolated component will automatically mean increased speed. Why? Traditional high-performance GPUs pair a large, powerful GPU with fast RAM. Maybe that’s what you are thinking about. But you don’t have to ship the GPU as a separate device to make it big, and nobody says that CPU memory cannot be fast. Sure, Apple could probably build a faster, larger dGPU that uses faster RAM. But they could also put a faster, larger GPU as well as faster RAM on the package. And that approach would retain all the inherent advantages of the unified memory model such as the ability to perform heterogeneous computing or the improved efficiency. In fact, it’s because of Apples integrated memory design that they don’t have to rely on super fast video RAM to achieve good performance.

 

[cmaier]

Or PoP?

Sure!

(ps: didn’t forget about you - am tied up on my day job)

 

[diamond.g]

Would it be safe to assume the memory controller would be with the cpu due not the GPU die?

You seem to assume that separating a GPU in its own, isolated component will automatically mean increased speed. Why? Traditional high-performance GPUs pair a large, powerful GPU with fast RAM. Maybe that’s what you are thinking about. But you don’t have to ship the GPU as a separate device to make it big, and nobody says that CPU memory cannot be fast. Sure, Apple could probably build a faster, larger dGPU that uses faster RAM. But they could also put a faster, larger GPU as well as faster RAM on the package. And that approach would retain all the inherent advantages of the unified memory model such as the ability to perform heterogeneous computing or the improved efficiency. In fact, it’s because of Apples integrated memory design that they don’t have to rely on super fast video RAM to achieve good performance.

 

[CWallace]

Would it be safe to assume the memory controller would be with the cpu due not the GPU die?

We're assuming the CPU, GPU and memory controller will all be on the same die and the memory will be off-die, but on-package (so same configuration as the M1).

 

[leman]

Would it be safe to assume the memory controller would be with the cpu due not the GPU die?

The memory controller would be where it makes most sense for it to be. It depends on how the memory hierarchy works, the fabric topology is etc.

 

[Jorbanead]

First, there will be a lot more binning options on Macs. (4 core, 6 core, 8 core, 10 core, and all the way up to 32 cores according to Bloomberg).

We have only heard of a 10-core, 20-core, and 40-core chip from Gurman. These are in addition to the current 8-core.

 

[reallynotnick]

True. Guess it depends on how many options Apple wants to offer with their product lineup, though it we get to a dozen (or more) configurations, it could make things messy.

Personally, I think Apple might want to simplify and that the 7 GPU option could be dropped with M2 as yields could be good enough to ensure enough "full" SoCs pass QA to meet supply and then we would have the 8/8, 10/16, 10/32, 20/64 and 40/128 options.

They always have multiple SKUs though, it's really not that messy. The 14" can come with 14 or 16 GPU cores and the 16" can come with 28 or 32 Cores. It also helps upsell people to the higher SKUs other than just RAM or storage changes.

The size of these chips are just going to be considerably bigger than the M1/M2, so while better yields might be enough to drop the 7 GPU core model I can't see that happening with a 32 core version and especially no way with 128 cores. Every vendor bins these large chips and it's just the reality of the process, otherwise you have to chuck silicon out.

 

[GubbyMan]

They always have multiple SKUs though, it's really not that messy. The 14" can come with 14 or 16 GPU cores and the 16" can come with 28 or 32 Cores. It also helps upsell people to the higher SKUs other than just RAM or storage changes.

The size of these chips are just going to be considerably bigger than the M1/M2, so while better yields might be enough to drop the 7 GPU core model I can't see that happening with a 32 core version and especially no way with 128 cores. Every vendor bins these large chips and it's just the reality of the process, otherwise you have to chuck silicon out.

I think there’s a possibility that Jade 2C and Jade 4C can be chiplet designs which would alleviate the need for binning by a little.

 

[AgentMcGeek]

Yes! I don’t think we’ve mentionned the possibility of chiplets before. That would make a lof of sense.