Proposed All-New Apple Silicon Mac Line-Up

[Danny82]

Macbook pro 16" base model bought..

After u finalise your range of dream product, may want to start a new thread with poll see who will buy what if it is true.. because I'm curious which model people are interested in haha.. just for fun..

Edit: thanks.. did not know can add poll to existing thread.. because I prefer the 14" form factor but really could not give up the power of 16".. so would like to see everyone opinions base on this dream setup..

 

[pappkristof]

The iMac is really missing from a 2x2 product matrix. I’d make it 3x2 with portable, all-in-one and desktop each having a pro variant.

 

[Boil]

Back to the 2 X 2 product matrix!

Mac mini - US$749.00 / US$1,249.00

12 P cores / 4 E cores / 24 GPU cores
LPDDR5 RAM Unified Memory Architecture - 16GB / 32GB
NVMe SSD (single NAND blade) - 512GB / 1TB
Four USB4 (TB3) ports
One Gigabit Ethernet port
One HDMI 2.1 port


Apple Cinema Displays - US$349.00 / US$499.00 / US$749.00

24" Apple Cinema Display / 4096x2304 / 4K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers

27" Apple Cinema Display / 5120x2880 / 5K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers

30" Apple Cinema Display / 6016x3384 / 6K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers


Mac mini can be docked into an Apple Cinema Display, thereby creating a modular iMac...!

The iMac is really missing from a 2x2 product matrix. I’d make it 3x2 with portable, all-in-one and desktop each having a pro variant.

The largest problem with iMacs is the display & computer are one. When the computer goes, the monitor is useless. Or if the monitor goes...

By making the concept of an iMac a modular one, either monitor or Mac mini can be upgraded or repurposed as needed!

 

[pappkristof]

Mac mini can be docked into an Apple Cinema Display, thereby creating a modular iMac...!

Please no! The iMac has a perfect place in Apple’s lineup.

 

[Boil]

Please no! The iMac has a perfect place in Apple’s lineup.

Sorry bruh, modular iMac is the future; already emailed Tim Cook, so it is pretty much a done deal...! ;^p

Besides, I am sure Jony Ive will make it look great!

 

[profcutter]

While believe me, going to NVME sticks for the new machines would be wonderful, I think Apple would see this as a step "backwards". Forcing folks to buy non-upgradable storage at the outset means more $$ for Apple, so I think we're stuck with soldered SSDs from now on. A new MBP with RAM and storage slots would be lovely, I just don't think it's going to happen. Hope I'm wrong.

 

[Boil]

While believe me, going to NVME sticks for the new machines would be wonderful, I think Apple would see this as a step "backwards". Forcing folks to buy non-upgradable storage at the outset means more $$ for Apple, so I think we're stuck with soldered SSDs from now on. A new MBP with RAM and storage slots would be lovely, I just don't think it's going to happen. Hope I'm wrong.

I reference NVMe as to the speeds, but you will note I also reference NAND blades, which indicate the Apple special sauce ones. For my Apple Silicon Mac line-up, everything is soldered, because Apple...! ;^p

 

[awesomedeluxe]

Actually, the MBP has additional 45W for the GPU so the total SoC draw should be 80-90W.

Unfortunately that's not how thermal design works. If you put the CPU and GPU inside the same space as the CPU was before, your thermal limit is not much changed from what the CPU had previously. And make no mistake, on the denser 5nm process, Boil's original APU design would absolutely be smaller than Intel's 14nm Coffee Lake part. It probably has a little more room to play since the rest of the system isn't generating as much heat, meaning it lives in a cooler house. But ultimately you need to be looking at W/area and not total W.

mOaR cOrEz...!!!

I like the new design a little better. You can do what you're trying to do if you are willing to completely separate the CPU and GPU and give up on the unified memory design. You also need to use chiplets at those core counts to prevent yield issues and low clocks - this video JMacHack shared is a great introduction. It would look like:

CPU: 4x 6 core firestorm dies @ 3.2GHz
32GB LPDDR5

GPU: 4x 12 core GPU dies
8GB HBM2E

This means you also need two high-speed silicon interposers, one for the CPU and one for the GPU, that these dies can sit on. I haven't seen any evidence that Apple has this technology but maybe we can wish it into existence.

This gives you a CPU ~48W and a GPU ~38W (including memory) which can all be cooled in the current MBP chassis.

 

[Boil]

I like the new design a little better. You can do what you're trying to do if you are willing to completely separate the CPU and GPU and give up on the unified memory design. You also need to use chiplets at those core counts to prevent yield issues and low clocks - this video JMacHack shared is a great introduction. It would look like:

CPU: 4x 6 core firestorm dies @ 3.2GHz
32GB LPDDR5

GPU: 4x 12 core GPU dies
8GB HBM2E

This means you also need two high-speed silicon interposers, one for the CPU and one for the GPU, that these dies can sit on. I haven't seen any evidence that Apple has this technology but maybe we can wish it into existence.

This gives you a CPU ~48W and a GPU ~38W (including memory) which can all be cooled in the current MBP chassis.

Considering it is Apple themselves that are stating they will be going with a Unified Memory architecture, I don't think it is something to "give up on"...

 

[Boil]

Added TB4 to the Mac Pro line-up.

Changed the Mac Pro (with PCIe slots / MPX modules) to the Mac Pro Cube, the addition of TB4 should allow PCIe expansion cards (audio DSPs & I/O, video DSPs & I/O) to run fine in a PCIe expansion chassis; or to have dedicated breakout boxes running off the TB4 with the asst. DSPs & I/O in the breakout box. One MPX-C slot for a 96 core GPGPU expansion card!

Yes! The return of the concept started with the Trashcan Mac Pro, but with the hardware to actually work this time...!?!

 

[awesomedeluxe]

Considering it is Apple themselves that are stating they will be going with a Unified Memory architecture, I don't think it is something to "give up on"...

I understand wanting to follow that expectation, but you gotta explain how you are cooling 24 CPU cores, 48 GPU cores, and 64GB of HBM2E on one package.

Even if you keep the CPUs at an underwhelming 2.9GHz, they're using around 1W a piece. And even if the GPU clocks are the same as the iPhone 12, they're using around .7W a piece. So that's 57.6W. Then you have up to 64GB of HBM2E... considering each 16GB stack is 5W, that's 20W. So 77.6W.

Separating things out into chiplets helps a little bit but not nearly enough. I'd guess you have 60W capacity maximum, so you need to redesign to fit in that threshold. Your CPU clocks are also really low, and single core performance is important. You need to give them more power too.

 

[Kostask]

I understand wanting to follow that expectation, but you gotta explain how you are cooling 24 CPU cores, 48 GPU cores, and 64GB of HBM2E on one package.

Even if you keep the CPUs at an underwhelming 2.9GHz, they're using around 1W a piece. And even if the GPU clocks are the same as the iPhone 12, they're using around .7W a piece. So that's 57.6W. Then you have up to 64GB of HBM2E... considering each 16GB stack is 5W, that's 20W. So 77.6W.

Separating things out into chiplets helps a little bit but not nearly enough. I'd guess you have 60W capacity maximum, so you need to redesign to fit in that threshold. Your CPU clocks are also really low, and single core performance is important. You need to give them more power too.

In saying “unified memory” it seems that there is some confusion. By unified memory, is Apple referring to the classical computer definition (as in all RAM isof tge same type and accessible by both GPU and CPU) as used by AMD for their APUs, or what is being talked about here, namely on SoC RAM?

I personally do not believe that Apple is talking about all the RAM being on SoC. Quite simply, there is no need for it. On the mobile devices, it is necessary due to space limitations. Those space limitations are not present on desktops/laptops. I can see some RAM being on chip (maybe 16GB?), withany other RAM being on the motherboard.

Also, people keep going on about HBM2e. It isn’t going to happen. There are reasons it isn’t commonly used, from supply constraints to heat generation. They still apply to Apple Silicon designs as they do to Intel based designs. There are places where the tradeoffs of HBM2e may be worth it; consumer computers and workstations are not those placed.

 

[iPadified]

Unfortunately that's not how thermal design works. If you put the CPU and GPU inside the same space as the CPU was before, your thermal limit is not much changed from what the CPU had previously.

This gives you a CPU ~48W and a GPU ~38W (including memory) which can all be cooled in the current MBP chassis.

You argument is that spreading CPU and GPU over space provides better cooling. I agree with that but your example also uses 86W combined TDP which I also suggested.

Extrapolating from the performance of a passively cooled 8W A12Z to a 45W actively cooled AS chip, the MPB16 will be very well off in terms of CPU and GPU.

 

[Pressure]

Apple Cinema Displays - US$349.00 / US$499.00 / US$749.00

24" Apple Cinema Display / 4096x2304 / 4K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers

27" Apple Cinema Display / 5120x2880 / 5K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers

30" Apple Cinema Display / 6016x3384 / 6K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers

There have never been an Apple display priced at $349, let alone $749. Mini-LED is new and expensive.

Try $1299, $2499 and $3799.

To reach 120Hz and with those sizes you also need to give the Thunderbolt implementation at least 8 PCIe lanes (v3) and that is just enough to drive a single 5K panel at 120Hz and 10-bit colour without sorting to compression.
[automerge]1595061329[/automerge]

Sorry bruh, modular iMac is the future; already emailed Tim Cook, so it is pretty much a done deal...! ;^p

Besides, I am sure Jony Ive will make it look great!

What year are we in? Jony quit last year.

 

[Waragainstsleep]

While believe me, going to NVME sticks for the new machines would be wonderful, I think Apple would see this as a step "backwards". Forcing folks to buy non-upgradable storage at the outset means more $$ for Apple, so I think we're stuck with soldered SSDs from now on. A new MBP with RAM and storage slots would be lovely, I just don't think it's going to happen. Hope I'm wrong.

I think we should all whinge to Greenpeace about the environmental impact of Apple making $1000+ machines disposable. Last time they went after Apple, they got some some results and soldering RAM and SSD is an absolute dick move on Apple's part. Also gluing glass to LCD panels sucks. They are supposed to be building the best products they can. If they want to make extra money, they could offer upgrades down the line. Something they have never done with most Macs.

 

[aeronatis]

I believe MacBook Air will have regular A14 with 4P & 4E cores just like the regular iPhone and iPad Air/Mini. The only difference from the iPhone/iPad line-up would be the amount of RAM being configurable.

14" and 16" MacBook Pro would have A14X (or whatever it is called) with the rumored 8P & 4E cores as well as the iPad Pros (or A14X for iPad Pro and A12Z for MacBook Pro), still the difference being the amount of RAM (and TDP obviously). This would possibly reduce (if not minimize) the performance gap on 14" and 16" MacBook Pro just like it is the case with iPad Pro 11" and 12.9".

As for iMac, we would get one last Intel & AMD Navi ones with current design and iMac Pro will probably be phased out. Mac Pro, on the other hand, will be the last to join Apple Silicon family as it needs more advanced chips and MPX modules which I don't think Apple is giving up on just one year after introducing.

 

[ksec]

I just think by the end of the transition in ~2022 there should be a $499 Mac Mini and $799 MacBook. Not when it is first released though.

Personally I really want to see a $399 Mac Mini and $699 MacBook which should mean Apple finally competing for much larger PC market and double Apple's active Mac user to 200M+. Especially in the business 500 sector, part of the reason why they are working on MDM themselves. But I guess those price point will be saved for Educational Discount.

As to your speculation of SoC Spec, they are far too aggressive. I wont be surprised if A14x only has 6 HP Core. With a lot more emphasis on GPU and NPU.

And those SRAM number dont make any sense. That is not how is works, in performance tuning, yield, die cost, etc...

 

[awesomedeluxe]

You argument is that spreading CPU and GPU over space provides better cooling. I agree with that but your example also uses 86W combined TDP which I also suggested.

I'm not sure what's misunderstood here. Do we agree that separating out the CPU and GPU allows you to use 80-90W, but that Boil's APU design would only have capacity between 45-55W?

In saying “unified memory” it seems that there is some confusion. By unified memory, is Apple referring to the classical computer definition (as in all RAM isof tge same type and accessible by both GPU and CPU) as used by AMD for their APUs, or what is being talked about here, namely on SoC RAM?

I personally do not believe that Apple is talking about all the RAM being on SoC. Quite simply, there is no need for it. On the mobile devices, it is necessary due to space limitations. Those space limitations are not present on desktops/laptops. I can see some RAM being on chip (maybe 16GB?), withany other RAM being on the motherboard.

There's no confusion about that. I know you can do a pinout to high speed memory rather than putting it on the same die or package, and that's still unified memory if it's what the GPU and CPU use. Saying there's no reason for the memory to be on package other than space limitations overlooks the needs of a high performance GPU. GPUs like to be close to their memory--seriously, look at any GPU design and you'll see it's generally brushing right up against its GDDR6/HBM2/etc. GPUs want fast access to fast memory, full stop. This is why dedicated VRAM exists.

Also, people keep going on about HBM2e. It isn’t going to happen. There are reasons it isn’t commonly used, from supply constraints to heat generation. They still apply to Apple Silicon designs as they do to Intel based designs. There are places where the tradeoffs of HBM2e may be worth it; consumer computers and workstations are not those placed.

Apple has repeatedly strong-armed AMD into supporting HBM2 so that they include it as an option in all of their pro level consumer computers and workstations.

HBM2E is literally the only high speed memory appropriate for both graphics and computing. Apple obviously knows about and likes the tech.

What you're suggesting is that the APU will use memory appropriate for a CPU, and bottleneck the GPU. That's fine for the MBP13 or Air which never had a GPU with dedicated VRAM to begin with. No disagreement there.

But it's not fine in a MBP16 where the performance expectation is a discrete graphics card with GDDR6 or HBM2E. You can't take an APU with a pinout to LPDDR5 and say, "yeah, that'll cut it, that's about as good as the Radeon 5600M with 8GB of HBM2E we offered last generation. Consumers are going to love Apple Silicon."

Supply issue is solvable. This is only necessary for the high end Macs which probably won't ship until 2021, and SK Hynix is ramping up production now. Heat issue is solvable. These are larger machines with more thermal capacity. They can cope if there's two stacks of HBM using 10W on the package.

 

[Boil]

Moved the Pro line-up (MacBook Pro & Mac Pro Cube) to HBM3 for the Unified Memory Architecture & went with a CPU / GPU chiplets & HBM3 stacks on interposer package kinda dealio...

Also added two more MPX-C slots to the cube; GPGPUs, FPGAs, Neural Engines, NVMe RAIDs; oh my...!

 

[dogslobber]

All these will have at least a 25% markup compared to the legacy Intel systems.

 

[Boil]

All these will have at least a 25% markup compared to the legacy Intel systems.

That is the beauty of this being my imaginary line-up, I get to set the imaginary pricing...!

(...which I have since updated, bumped them a bit since I conceded to the fact that Apple likes money...)

 

[Kostask]

I'm not sure what's misunderstood here. Do we agree that separating out the CPU and GPU allows you to use 80-90W, but that Boil's APU design would only have capacity between 45-55W?


There's no confusion about that. I know you can do a pinout to high speed memory rather than putting it on the same die or package, and that's still unified memory if it's what the GPU and CPU use. Saying there's no reason for the memory to be on package other than space limitations overlooks the needs of a high performance GPU. GPUs like to be close to their memory--seriously, look at any GPU design and you'll see it's generally brushing right up against its GDDR6/HBM2/etc. GPUs want fast access to fast memory, full stop. This is why dedicated VRAM exists.

Apple has repeatedly strong-armed AMD into supporting HBM2 so that they include it as an option in all of their pro level consumer computers and workstations.

HBM2E is literally the only high speed memory appropriate for both graphics and computing. Apple obviously knows about and likes the tech.

What you're suggesting is that the APU will use memory appropriate for a CPU, and bottleneck the GPU. That's fine for the MBP13 or Air which never had a GPU with dedicated VRAM to begin with. No disagreement there.

But it's not fine in a MBP16 where the performance expectation is a discrete graphics card with GDDR6 or HBM2E. You can't take an APU with a pinout to LPDDR5 and say, "yeah, that'll cut it, that's about as good as the Radeon 5600M with 8GB of HBM2E we offered last generation. Consumers are going to love Apple Silicon."

Supply issue is solvable. This is only necessary for the high end Macs which probably won't ship until 2021, and SK Hynix is ramping up production now. Heat issue is solvable. These are larger machines with more thermal capacity. They can cope if there's two stacks of HBM using 10W on the package.

That is why I said "some" RAM can be on SoC and the rest on the motherboard. The on SoC RAM can be used by the GPU cores, and the off SoC RAM can be used by the CPUs (mostly), but not limited to CPUs, just as the on SoC RAM will not be limited to the GPUs. If the 16GB of RAM is insufficient to supply the GPUs, then we are talking very, very high end Pro machines, and I think they will have dGPUs, anyway. If there is 8GB of the 16GB on the SoC used for the GPUs, it should be sufficient, in a non-pro environment. You could make the 8GB or even all 16GB of the on SoC RAM, VRAM.

It is very difficult to see 32GB, or 64GB of on SoC RAM. 16GB will be pushing it, already. However, Apple will most likely be offering more than 16GB of RAM as upgrades or BTO options. So how do you put that into a machine?

Apple will not offer HBM2e or HBM3 memory. This is because of supply constraints. They have not left Intel due to supply issues, to end up with supply issues with Samsung and SK Hynix. SK Hynix will have some glitches at startup, and can be expected. Like I said, this is not a consideration for Pro machines, as I think they will be using dGPUs. While there may not be glitches in the best case, you are now talking about 16GB or more of HBM RAM on every shipping machine, volumes which have never been seen before. Meanwhile, GDDR5 or GDDR6 can provide the bandwidth needed at a lower cost and far less risk.

 

[iPadified]

I'm not sure what's misunderstood here. Do we agree that separating out the CPU and GPU allows you to use 80-90W, but that Boil's APU design would only have capacity between 45-55W?

Yes, that sums it up. 45-55 W of AS Mac will probably be more than enough though.

 

[Boil]

That is why I said "some" RAM can be on SoC and the rest on the motherboard. The on SoC RAM can be used by the GPU cores, and the off SoC RAM can be used by the CPUs (mostly), but not limited to CPUs, just as the on SoC RAM will not be limited to the GPUs. If the 16GB of RAM is insufficient to supply the GPUs, then we are talking very, very high end Pro machines, and I think they will have dGPUs, anyway. If there is 8GB of the 16GB on the SoC used for the GPUs, it should be sufficient, in a non-pro environment. You could make the 8GB or even all 16GB of the on SoC RAM, VRAM.

It is very difficult to see 32GB, or 64GB of on SoC RAM. 16GB will be pushing it, already. However, Apple will most likely be offering more than 16GB of RAM as upgrades or BTO options. So how do you put that into a machine?

Apple will not offer HBM2e or HBM3 memory. This is because of supply constraints. They have not left Intel due to supply issues, to end up with supply issues with Samsung and SK Hynix. SK Hynix will have some glitches at startup, and can be expected. Like I said, this is not a consideration for Pro machines, as I think they will be using dGPUs. While there may not be glitches in the best case, you are now talking about 16GB or more of HBM RAM on every shipping machine, volumes which have never been seen before. Meanwhile, GDDR5 or GDDR6 can provide the bandwidth needed at a lower cost and far less risk.

Apple will most likely not being going the SoC route for the Mac Pro line-up. CPU & GPU chiplets, an I/O chiplet, & a bunch of stacks of HBM3; all gathered on an interposer / package...

Your talk of memory on the SoC & memory on the logic board is not compatible with Apple's desire for a Unified Memory Architecture...

Supply constraints... Might be possible that Apple is working with SK Hynix to secure vast quantities of HBM3 (which is supposed to be cheaper than previous HBM products, and run cooler)...

 

[Kostask]

Unified memory is memory that can be used by both CPU and GPU. Apple puts 16GB of GDDR5/6 on SoC, accessible to both GPU and CPU. Anything more, to take the system to 24GB, 32GGB, or more, is LPDDR5 and is off SoC, also accessible to both CPU and GPU.

My other point was, with 16GB of GDDR5/6 on SoC, there is far more than enough memory for the GPUs in a consumer level computer, so while the GPUs could get data from off SOC, there would be very few instances where they would actually need to do so. This fulfills the high bandwidth requirements of the GPUs, while keeping costs and heating down.

I still don't see where HBMx fits in a consumer grade Mac.