Proposed All-New Apple Silicon Mac Line-Up

[Boil]

Here it is, the entire plans for Apple transitioning to their own silicon...! ;^p

(...as seen within my fever dreams...)

Back to the 2 X 2 product matrix!


MacBook - starting at US$999.00

12 P cores / 4 E cores / 24 GPU cores - Monolithic SoC design
LPDDR5 RAM Unified Memory Architecture - 16GB / 32GB / 64GB
NVMe SSD (single NAND blade) - 512GB / 1TB / 2TB
Two USB4 / TB4 ports
14" display / 2560x1600 / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers / Nano texture option


Mac mini - starting at US$699.00

12 P cores / 4 E cores / 24 GPU cores - Monolithic SoC design
LPDDR5 RAM Unified Memory Architecture - 16GB / 32GB / 64GB
NVMe SSD (single NAND blade) - 512GB / 1TB / 2TB
Four USB4 / TB4 ports
One Gigabit Ethernet port
One HDMI 2.1 port


MacBook Pro - starting at US$2,499.00

24 P cores / 4 E cores / 48 GPU cores - CPU & GPU Chiplets on interposer / System in Package (SiP) design
HBM3 Unified Memory Architecture - 32GB / 64GB / 128GB
NVMe RAID 0 (dual NAND blades) - 1TB / 2TB / 4TB
Four USB4 / TB4 ports
16" display / 3072x1920 / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers / Nano texture option


Mac Pro Cube - starting at US$5,999.00

48 P cores / 4 E cores / 96 GPU cores - CPU & GPU Chiplets on interposer / System in Package (SiP) design
HBM3 Unified Memory Architecture - 128GB / 256GB / 512GB
NVMe RAID 0 (dual NAND blades) 4TB / 8TB / 16TB
Eight USB4 / TB4 ports
Two 10Gb Ethernet ports
One HDMI 2.1 port
Three MPX-C slots (for use with asst. MPX-C expansion modules)


Apple MPX-C Expansion Modules - starting at US$499.00

NVMe RAID Storage Module (Quad NAND blades)
GPGPU Module
FPGA Module
Neural Engine Module


Apple Cinema Displays - starting at US$799.00

24" Apple Cinema Display / TB3 / 4096x2304 / 4K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers / Nano texture option

27" Apple Cinema Display / TB3 / 5120x2880 / 5K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers / Nano texture option

32" Apple Cinema Display / TB3 / 6016x3384 / 6K / 120Hz ProMotion / Mini-LED / 4K webcam / FaceID / stereo speakers / Nano texture option


New Apple Accessories:

Apple Low-Profile Mechanical Keyboard US$199.00

Apple Magic Mouse 3D - US$149.00

Apple Sidecar Dock - US$199.00


Sidecar Dock provides stable / adjustable / rotatable easel stand for iPad Pro, charges iPad Pro & connects to two USB4 (TB3) ports & a Gigabit Ethernet port via Smart Connector

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

Entire line-up is all-new unified design, sorry Cheesegrater 2.0 & XDR Display...! ;^p

Discuss...! ;^p

 

[duanepatrick]

Any reference links for this?

 

[throAU]

Discuss...! ;^p

I think it's somebody's fantasy list and has no basis in reality.

Any reference links for this?

I'm guessing nope.

 

[Boil]

Oh yeah...

Apple also branches / forks off Blender, licenses Octane, revives Phenomenon, somehow acquires Allegorithmic / Substance Suite from Adobe, and integrates it all with Final Cut Pro X & Logic X into the be-all end-all Digital Content Creation software suite...!!!

US$1,999.00 - single seat, standard Octane Studio render license for two GPUs / GPGPUs

US$499.00 annual maintenance
[automerge]1594792160[/automerge]

Any reference links for this?

I think it's somebody's fantasy list and has no basis in reality.

I'm guessing nope.

Just trying to have a good time here folks, get some discussion stirring on what a complete Apple Silicon Mac line-up might look like...!

And a bit of hopeful thinking in regards to Apple taking more of a personal interest in 3D on the Mac & expanding their Pro software line-up...!

Lighten Up!!! ;^p

 

[peanutbridges]

Right, because silicon scales so linearly that no one else has simply added them all together before. Only Apple can of course without any penalty.

You can't just glue cores together and expect no repercussions. Nor are processors all that cheap especially given sub-5nm. Are they going monolithic design ala Intel's mesh ring design or chiplets ala AMD? Are cores all that much useful if all that can be done on accelerators reducing overall complexity and sunk costs and resources and time?

 

[endlessike]

40 core imac. why.

 

[leman]

What does “LPDDR5 tile memory mean”? You don’t actually mean using LPDDR5 for tile memory, do you? That would make one hell of a slow GPU. Tile memory is implemented using ultra-fast on-chip SRAM, neither do you need xGB of it.

 

[Boil]

Right, because silicon scales so linearly that no one else has simply added them all together before. Only Apple can of course without any penalty.

You can't just glue cores together and expect no repercussions. Nor are processors all that cheap especially given sub-5nm. Are they going monolithic design ala Intel's mesh ring design or chiplets ala AMD? Are cores all that much useful if all that can be done on accelerators reducing overall complexity and sunk costs and resources and time?

Who knows man, who knows...!

Apple knows...! ;^p

Lighten up, join the discussion... Tell us what you think high-end Apple Silicon might look like.

40 core imac. why.

The current Apple Silicon does not have SMT (multi-threading). the current iMac Pro goes up to a 18-core CPU, so 36 threads; my two iMac Pro APU offerings are 32 core & 48 core...
[automerge]1594794264[/automerge]

What does “LPDDR5 tile memory mean”? You don’t actually mean using LPDDR5 for tile memory, do you? That would make one hell of a slow GPU. Tile memory is implemented using ultra-fast on-chip SRAM, neither do you need xGB of it.

See, these are the things we learn while having a discussion about stuff...!

So, correct to SRAM for Tile Memory, but what quantities would you expect?

And thoughts about HBM2e for Tile Memory in the Mac Pro line-up (MacBook Pro / iMac Pro / Mac Pro Cube / Mac Pro Rackmount)...?

 

[leman]

So, correct to SRAM for Tile Memory, but what quantities would you expect?

Tile memory is basically a large on-chip register/cache file, with performance to match. Current Apple GPUs have 32KB per tile (I assume this is for each GPU core). This gives you up to 128 bytes of data per pixel using 16x16 tiles or 32 bytes for 32x32 tiles - absolutely sufficient for general use.

Maybe new GPUs will increase this amount - but I kind of doubt it. Register memory is very expensive and 32KB is plenty.

 

[peanutbridges]

I'll keep it simple: this is just low-level consumer speculation. If you're imagining a new future potential, the best you do is just keep stacking core count? Apple isn't primarily a chip maker, I doubt they will have ample time to redesign a bunch of petty count configurations that may or may not scale up to n cores with efficient use of space and energy. What's the difference between 32 and 48 cores if the system barely scales beyond four in 99% of workloads? It's wasted time and effort and money to make something big for no gain.

 

[Boil]

I'll keep it simple: this is just low-level consumer speculation. If you're imagining a new future potential, the best you do is just keep stacking core count? Apple isn't primarily a chip maker, I doubt they will have ample time to redesign a bunch of petty count configurations that may or may not scale up to n cores with efficient use of space and energy. What's the difference between 32 and 48 cores if the system barely scales beyond four in 99% of workloads? It's wasted time and effort and money to make something big for no gain.

Yes, this is totally low level speculation. I am not an engineer of any sort. I do not have any juicy rumors and/or leaks. I am just spitballing here. Just trying to think what might be comparable hardware to the outgoing systems? What would you see as a possible Apple Silicon line-up, on the high-end / Mac Pro line-up of products?!?

As for Apple being a chip maker, it seems that is what they have been doing the last ten years with the A-series. Now they want to replace ALL of their Mac computers with their own Apple Silicon. I am sure they have been working on more than just the A-series (and other assorted chips that go along with the A-series) APUs all this time...

BTW, my APUs listed above, there are eight variations total, across eight different Mac models; two laptops, three all-in-ones, three desktops, & one rackmount beast...

 

[ian87w]

Apple being Apple, I'd say the base storage will start at 256GB for the Macbooks. RAM will match current offerings, starting at 8GB.

Even the iMacs might come with 256GB base if they switch entirely to SSD. Current iMacs uses old-n-slow (but cheap) 5400rpm hard drive for base config. If the pricing remains the same and Apple uses completely SSD (no more mechanical drives or fusion drive), then base storage might be down to 256GB. 512GB+ SSDs are not that cheap yet.

For consumers, basically they need to think about their RAM and storage need for the life of the computer, as I'm sure everything will be soldered and/or not user accessible. Apple already made it super hard to just upgrade the RAM of the latest mac mini.

 

[Danny82]

Liking it when you are having a fever lol..!!! I assume all are mini led @ 120hz and those with 4k webcam comes with FaceID.. then 16" mbp for me lol..!!!

 

[Boil]

Tile memory is basically a large on-chip register/cache file, with performance to match. Current Apple GPUs have 32KB per tile (I assume this is for each GPU core). This gives you up to 128 bytes of data per pixel using 16x16 tiles or 32 bytes for 32x32 tiles - absolutely sufficient for general use.

Maybe new GPUs will increase this amount - but I kind of doubt it. Register memory is very expensive and 32KB is plenty.

Now you are making my head hurt. Mongo no like the maths...!

So what does that mean in relation to my envisioned GPU core counts & Tile Memory amounts?

And is there any benefit for having the Pro line-up using HBM2e for Tile Memory (over the previously mentioned SRAM), or am I just chasing Spec Buzz Acronyms?!?

Thanks!!!

Apple being Apple, I'd say the base storage will start at 256GB for the Macbooks. RAM will match current offerings, starting at 8GB.

Even the iMacs might come with 256GB base if they switch entirely to SSD. Current iMacs uses old-n-slow (but cheap) 5400rpm hard drive for base config. If the pricing remains the same and Apple uses completely SSD (no more mechanical drives or fusion drive), then base storage might be down to 256GB. 512GB+ SSDs are not that cheap yet.

For consumers, basically they need to think about their RAM and storage need for the life of the computer, as I'm sure everything will be soldered and/or not user accessible. Apple already made it super hard to just upgrade the RAM of the latest mac mini.

Went with a minimum of 512GB because that is what the Mac mini DTK has. Adjusting specs to also reflect my personal belief that 16GB of system RAM (yeah, yeah; Unified Memory Architecture) should be the minimum these days. Future proofing!

Liking it when you are having a fever lol..!!! I assume all are mini led @ 120hz and those with 4k webcam comes with FaceID.. then 16" mbp for me lol..!!!

That is the kind of enthusiasm I like seeing here! Adjustments to the specs made!

LOL, I just killed off the Apple XDR Display & regulated the "Big Chungus" Mac Pro to a rackmount...! ;^p

Remember, all-new Apple Silicon Macs deserve all-new chassis designs!

 

[leman]

Now you are making my head hurt. Mongo no like the maths...!

So what does that mean in relation to my envisioned GPU core counts & Tile Memory amounts?

Let me put it like this: it’s like you discussing your dream car and mentioning that it would have twenty batteries. So it’s more like: why?

And is there any benefit for having the Pro line-up using HBM2e for Tile Memory (over the previously mentioned SRAM), or am I just chasing Spec Buzz Acronyms?!?

Not much benefit unless you really want to drag the performance down. Internal chip caches are usually 10-20 times faster than any external DRAM.

And yes, you are absolutely chasing acronyms

 

[Boil]

Let me put it like this: it’s like you discussing your dream car and mentioning that it would have twenty batteries. So it’s more like: why?

Not much benefit unless you really want to drag the performance down. Internal chip caches are usually 10-20 times faster than any external DRAM.

And yes, you are absolutely chasing acronyms

So SRAM across the board it is, and I guess such a scheme would carry over to GPGPUs tied into the UMA...?

B... bu... but how much SRAM, Senpai...?!? ;^p

RE: chasing acronyms, been in the Ryzen / Vega / Navi playground too long, HBM / HBM2 / HBM2e are the ht ones for AMD GPUs... ;^p
[automerge]1594810275[/automerge]
The irony of my envisioned new Mac Pro Cube is that it could actually be what the Trashcan Mac Pro was trying to be...

If expansion slot is populated, it would have a CPU, a GPU, & a GPGPU (APU is first two combined, & the GPGPU is still the GPGPU). These two items could share a core heatsink (I would have one of the thicc fans from the current Mac Pro up top exhausting, maybe a second underneath pulling air in?) or they could have individual heatsinks sharing portions of the vertical wind tunnel area.

The USB4 (TB3) ports would be the only other avenue of expansion.

So, maybe the lower power, more efficient APUs Apple might now be crafting were what the Trashcan Mac Pro needed (along with third-party TB2 'stuff'); and now a new Mac Pro Cube can realize that dream...!

After all, the OG NeXT Cube sitting somewhere in the Mothership with the digital consciousness of Steve Jobs needs an total update to new Apple Silicon hardware..!!! ;^p

I am enjoying my thread.

 

[Boil]

Alrighty folks! I have updated my Hypothetical Specifications to reflect the thoughts from some that there are too many cores.

TOO MANY CORES...!!!

Something had to be done!

Still totally unsure about proper amounts of SRAM to assign throughout the systems...

But I still have eight products.

Four Consumer; one laptop, one desktop, & two all-in-ones

MacBook
Mac mini
24" iMac
27" iMac

Four Pro; one laptop, one desktop, one all-in-one, & one rackmount

MacBook Pro
Mac Pro Cube
iMac Pro
Mac Pro Rack

And between ALL eight products, there are only four APUs...!

Dang, I think I need to invoice Apple...!!! ;^p
[automerge]1594812631[/automerge]
The MacBook, Mac mini, & 24" iMac all share a "low-end" APU:

12 P cores / 4 E cores / 24 GPU cores / 1GB SRAM Tile Memory


The 27" iMac & the MacBook Pro share a mid-range APU:

16 P cores / 4 E cores / 32 GPU cores / 2GB SRAM Tile Memory


The iMac Pro & the Mac Pro Cube share a high-end APU:

24 P cores / 4 E cores / 48 GPU cores / 2GB SRAM Tile Memory


The Mac Pro Rack has an "extreme" APU all to itself:

32 P cores / 4 E cores / 56 GPU cores / 4GB SRAM Tile Memory

I like to call it "APUzilla"...! ;^p

 

[leman]

So SRAM across the board it is, and I guess such a scheme would carry over to GPGPUs tied into the UMA...?

B... bu... but how much SRAM, Senpai...?!? ;^p

I don’t think that it matters really. What I can tell you for sure that you won’t have 4GB cache on those APUs. That would be pointless and wasteful. Tile memory will still be 32-64KB, with maybe few meg total cache per GPU core (if at all), and probably 32MB+ SLC cache.

 

[Boil]

I don’t think that it matters really. What I can tell you for sure that you won’t have 4GB cache on those APUs. That would be pointless and wasteful. Tile memory will still be 32-64KB, with maybe few meg total cache per GPU core (if at all), and probably 32MB+ SLC cache.

I just changed the SRAM amounts across the board:

24 GPU cores / 256MB SRAM Tile Memory

32 GPU cores / 512MB SRAM Tile Memory

48 GPU cores / 768MB SRAM Tile Memory

56 GPU cores / 1GB SRAM Tile Memory


But you are saying a few meg per GPU core; so, 2MB per core:

24 CPU cores / 48MB SRAM Tile Memory

32 GPU cores / 64MB SRAM Tile Memory

48 GPU cores / 96MB SRAM Tile Memory

56 GPU cores / 112MB SRAM Tile Memory


???


I guess I am having trouble wrapping my head around numbers that do not match up with the numbers I am familiar with in the Ryzen / Radeon (Vega / Vega II / Navi) hardware world...!

 

[leman]

Just drop any mention of time memory. For performance, number of cores is much more important...

 

[Boil]

Just drop any mention of time memory. For performance, number of cores is much more important...

But Apple has "Tile Memory" in their block diagrams for the new Apple Silicon GPUs, in their Metal talks, something about the GPU cores not needing to go out to the UMA pool as much when Tile Memory is sitting on the GPU cores...?

 

[leman]

But Apple has "Tile Memory" in their block diagrams for the new Apple Silicon GPUs, in their Metal talks, something about the GPU cores not needing to go out to the UMA pool as much when Tile Memory is sitting on the GPU cores...?

It is sufficient that tile memory exist, it’s size is secondary. Tile memory is just special kind of cache, increasing it past a certain limit won’t have any benefit. The key question here is: how much per-pixel data a program might want to use? Current Apple GPU allow you to use up to 128 bytes per pixel, that’s 32 single prevision numbers. Which is quite a lot. Maybe they will bump the size in the next revision, maybe they won’t, who knows. But increasing this number too much would make as much sense as making a family car with luggage compartment of a size of a tennis court - it would be useless, too large, too awkward and way too expensive.

 

[JSRinUK]

I'm sure I've no idea what 90% of all that means, but I'm sure I need it all! 😁

 

[Boil]

It is sufficient that tile memory exist, it’s size is secondary. Tile memory is just special kind of cache, increasing it past a certain limit won’t have any benefit. The key question here is: how much per-pixel data a program might want to use? Current Apple GPU allow you to use up to 128 bytes per pixel, that’s 32 single prevision numbers. Which is quite a lot. Maybe they will bump the size in the next revision, maybe they won’t, who knows. But increasing this number too much would make as much sense as making a family car with luggage compartment of a size of a tennis court - it would be useless, too large, too awkward and way too expensive.

Yeah, I went thru & wiped them all...

I really think the line-up is looking GREAT...!!!

 

[dlewis23]

What's the difference between 32 and 48 cores if the system barely scales beyond four in 99% of workloads? It's wasted time and effort and money to make something big for no gain.

This is potently (And really likely) the really cool part of Apple moving to ARM that wont get discussed very much. Because of what ARM is, having CPU's with much higher core counts will become normal. Your thinking of software today where its super common to run into just 2 cores 4 threads or, 4 cores 4 threads. Combine that with operating systems that haven't made it super easy to scale across a bunch of cores. There is no incentive for the developer to write the software to take more advantage of more cores.

Now with moving to ARM where 32 cores could very well be common place with high and low power cores the OS has to spread the load out very different than it ever had to before when on x86 and likely will be designed to deal with many more cores. The software you use will go along with that.

In the server world this has already happened a lot because having a server with 32 cores 64 threads or more is very easy to come by.

It's not common place to have 32 cores in consumer hardware today so the software often doesn't take advantage of it even when its there. Switching to ARM and having the potential to have CPU's with a ton of cores will fix your concern.