No dedicated video RAM on ARM Macs

[Boil]

That said, I would stop at 16 CPU cores. At higher clock speeds those extra cores are a liability not worth their weight. You can always get some bang for your buck increasing the GPU core count, but boosting the CPU core count is a game of diminishing returns.

Let's assume the GPU cores are now clocked up relative to what TSMC's 5nm process can easily allow and each consume 1W. And in taking the Firestorm cores from 3GHz to 3.5GHz, they now consume an egregious 4W (this might be too generous, but I don't feel like math right now).

64W - 16 Firestorm Cores
64W - 64 Apple GPU Cores
20W - 4x 16GB HBM2 (64GB)
---
148W

The APU might still melt. The CPU cores are hot and very close together and surrounded by GPU cores and HBM stacks that are also hot and very close together. But it will maybe be OK. Its biggest problem is the slew of 5GHz x86 processors that will run circles around it. Unified high speed memory will help with that a little.

I would think the Mac Pro line-up would want more than 16 cores, current Mac Pro goes up to 28 cores & has hyperthreading. iMac Pro goes up to 18 cores & also has hyperthreading.

Many have said the current Apple arm64-based APUs do not have hyperthreading, and if that stays the norm, then a higher P core count is needed to make up for the lack of hyperthreading?

As far as cooling a 148W APU, two thing:

1 - Larger die (Threadripper-sized package) so not so 'crowded'
2 - Large heat sink like in current Mac Pro should cool any of the wattage I outlined earlier with no problem

 

[Pressure]

Apple was already clear at WWDC that the Mac SOCs are a new series which we have not seen yet - a new family.

There is no reason to overthink it. The only new thing from ImgTec are the A-series GPUs, AXE, AMX and AXT.

It scales from 16 FP32 FLOPs/Clock up to 2048 FP32 FLOPs/Clock.

 

[Kpjoslee]

There is no reason to overthink it. The only new thing from ImgTec are the A-series GPUs, AXE, AMX and AXT.

It scales from 16 FP32 FLOPs/Clock up to 2048 FP32 FLOPs/Clock.

Apple stopped using Imgtec (or custom variant) GPU since A11.

 

[leman]

Many have said the current Apple arm64-based APUs do not have hyperthreading, and if that stays the norm, then a higher P core count is needed to make up for the lack of hyperthreading?

I am wondering if Apple Silicon even needs hyper threading. With its string decode performance and large reorder buffer, it seems it can do a decent job utilizing the execution units without SMT.

 

[Boil]

I am wondering if Apple Silicon even needs hyper threading. With its string decode performance and large reorder buffer, it seems it can do a decent job utilizing the execution units without SMT.

Yeah, not saying we need SMT, just saying we might need more cores to offset not having SMT? For a Mac Pro line-up perspective?

 

[Pressure]

Apple stopped using Imgtec (or custom variant) GPU since A11.

No, they are still based on ImgTec.

They are also still licensing the technology. I don’t think it even is possible to make a tile-based deferred renderer without licensing from ImgTec.

 

[leman]

Yeah, not saying we need SMT, just saying we might need more cores to offset not having SMT? For a Mac Pro line-up perspective?

I'm curious about this as well. SMT in the end is a hack to increase the hardware utilization (you "fake" presence of multiple CPUs so that you can feed all the execution units by combining instructions from different threads). But if Apple CPUs already have high hardware utilization, they don't need SMT. In a nutshell, if SMT on an Intel CPU increases the performance bey 30%, but Apple already offers 30% higher single-threaded performance, they won't need SMT. Fewer faster cores are always better than more slower cores. But one would need to do a lot of CPU execution profiling to see how good Apple Silicon is at utilizing it's hardware.

No, they are still based on ImgTec.

They are also still licensing the technology. I don’t think it even is possible to make a tile-based deferred renderer without licensing from ImgTec.

Apple GPUs are definitively historically based on PowerVR technology, but they have been so heavily customized and tweaked, that it is a whole new design by now. For example, the shader core is custom Apple design from what I hear. They are also a good deal faster. So, no, whatever GPUs Imagination has available now bears little relevance to what Apple is currently doing.

 

[Tech198]

I always hated the shared memory approach, however if it's integrated you can bet it will defiantly be used more.. How do we know it uses dedicated only when Apple reckons it should? .. ok.. i'm speculating..

 

[Pressure]

Apple GPUs are definitively historically based on PowerVR technology, but they have been so heavily customized and tweaked, that it is a whole new design by now. For example, the shader core is custom Apple design from what I hear. They are also a good deal faster. So, no, whatever GPUs Imagination has available now bears little relevance to what Apple is currently doing.

Yeah, I didn't mean to imply they have taken any of those designs without customising the basic GPU Core for their needs.

My only point being that it is practically impossible to make a TBDR without violating ImgTec patents. Hence the continuing licensing of ImgTec technology by Apple.

 

[leman]

Yeah, I didn't mean to imply they have taken any of those designs without customising the basic GPU Core for their needs.

I was referring more to your previous post, where you listed Imagination's GPUs. Just wanted to make it clear that in the end these are different families.

 

[abhibeckert]

This is where my speculation starts.

You're forgetting the part where Apple seems to have the best CPU and GPU design team in the world.

The A12, both the CPU and GPU, is about on par with an i7 desktop processor and a GTX 1080. And it does that with something like a 5W TDP. The i7 and GTX 1080 should be paired with a 600 watt power supply.

Sure, an i9 and GTX 2080 will perform a bit better than an A12, but not much better. Imagine what Apple's chip design team can do with a 600W TDP. Or heck even just a 60W TDP (which would be less than an Intel MacBook Pro).

Honestly, I don't think we should bother speculating, I don't think we're qualified to predict what they're capable of. I'm happy to wait and see what they ship towards the end of this year, and over the next 5 years.

 

[diamond.g]

You're forgetting the part where Apple seems to have the best CPU and GPU design team in the world.

The A12, both the CPU and GPU, is about on par with an i7 desktop processor and a GTX 1080. And it does that with something like a 5W TDP. The i7 and GTX 1080 should be paired with a 600 watt power supply.

Sure, an i9 and GTX 2080 will perform a bit better than an A12, but not much better. Imagine what Apple's chip design team can do with a 600W TDP. Or heck even just a 60W TDP (which would be less than an Intel MacBook Pro).

Honestly, I don't think we should bother speculating, I don't think we're qualified to predict what they're capable of. I'm happy to wait and see what they ship towards the end of this year, and over the next 5 years.

GFXBench 5.0 Aztec Ruins High Offscreen results disagrees with your assertion that the A12 GPU is on par with a GTX1080.

 

[leman]

The A12, both the CPU and GPU, is about on par with an i7 desktop processor and a GTX 1080. And it does that with something like a 5W TDP. The i7 and GTX 1080 should be paired with a 600 watt power supply.

The A12Z is a surprisingly fast GPU, but its definitely far of being a match for a GTX 1080.

Honestly, I don't think we should bother speculating, I don't think we're qualified to predict what they're capable of. I'm happy to wait and see what they ship towards the end of this year, and over the next 5 years.

Well, I think that speculating on these topics is a lot of fun.

 

[chengengaun]

Just want to say what a treasure trove of information and intelligent speculation here (I meant that as a compliment). Thanks all and keep it coming, I am reading.

 

[Krevnik]

Die space is just a matter of cost and wafer yields. The reticle limit for TSMC 7nm node is ~830mm².

A12X/Z	122mm²
A13	98mm²
Ryzen 4800H	156mm²
Radeon 5700 XT (Navi)	251mm²
Radeon VII (Vega20)	331mm²
NVidia A100 (Ampere)	826mm²

I already mentioned the wafer yields being the cost of chasing this, so thanks for agreeing, I guess.

But you are partly making my own point for me. Desktop GPU dies are enormous compared to CPUs and getting a good binned GPU die with everything intact is not exactly cheap.

Its not unreasonable to expect that for Apple to compete with a 5700 XT while also competing with say, the i7 8700 or Ryzen 3600, they need to be bigger than the 5700 XT itself to make room for the CPU and other custom components on the die. And yields aren’t exactly linear with die size. So an SoC, because it shares die space with other components, will always have that disadvantage. And when cost/yields are part of the equation, that’s important.

I’m not even attempting to claim it’s not possible. I’m saying an SoC has to include components a dGPU doesn’t have to, which creates problems and will mean that it’s unlikely that we will see an SoC that beats good dedicated GPUs. It’s not like it hasn’t been tried repeatedly. But it doesn’t have to, when there’s no reason to not have the PCIe lanes on the machines that need it.

I would think Apple would do increasingly larger SoCs depending on the resources needed. Mac Pro line-up would have Threadripper-sized SoCs. Whether monolithic or chiplets, who knows (well, Apple has an idea)?

Thats obvious. But that’s not my argument. My argument was against the statement that there’s no reason an SoC can’t hold up to a dGPU. My point is that an SoC is always going to be fighting extra hurdles because those dies have to also share space with the CPU and other components.

But you’re feeding my point. Those large dies are going to be needed for the CPU cores and I/O the Mac Pro will need. That’s less room to add hardware that can hold up to a W5700 or Vega II.

 

[awesomedeluxe]

I would think the Mac Pro line-up would want more than 16 cores, current Mac Pro goes up to 28 cores & has hyperthreading. iMac Pro goes up to 18 cores & also has hyperthreading.

Many have said the current Apple arm64-based APUs do not have hyperthreading, and if that stays the norm, then a higher P core count is needed to make up for the lack of hyperthreading?

As far as cooling a 148W APU, two thing:

1 - Larger die (Threadripper-sized package) so not so 'crowded'
2 - Large heat sink like in current Mac Pro should cool any of the wattage I outlined earlier with no problem

I certainly take your point about reaching parity with current Mac Pro designs. I think you may be confused on the difference between a die and a package, which is certainly understandable. Cores go on dies and dies go on packages. Apple's A13 is just one die, which means near-zero latency. The biggest threadripper has multiple dies, not one "big die," separated out on one package where each die contains eight cores. This is partially for cooling, but it's also because as your die gets larger it's more likely to accrue manufacturing defects. It looks like this:

So what you are actually suggesting here is something leman suggested before; instead of making a mega-die APU, Apple might separate the CPU and GPU into different dies on the same package that share the same memory (also on the same package). There may even be two or four dies for CPU cores. That may well be the solution Apple goes with, but it's a solution to a different problem than I was trying to solve.

Note also that the reason a monster APU is possible at all is because 5nm is way denser than 7nm. That lets you fit more cores in a smaller space, solving the manufacturing problem I described above. But that leaves things more crowded and more thermal constrained, making cooling a bigger issue.

 

[cool11]

Is it possible, apple to offer less power/performance in their new arm macbook pros?
I can't imagine it.

So, what is realistic to expect?

 

[Zdigital2015]

I'd have to disagree that modular memory by itself makes a device less reliable. Memory does not commonly go bad and if it does you can easily swap a new stick of RAM at any time to fix an issue (I think from that point of view it can actually help with the reliability and longevity of the machine as a whole).

With soldered ram it makes it extremely difficult (and from the point of an average user pretty much impossible) to fix bad memory. I agree that soldered ram allows for lower power consumption and other benefits which are very important for portable machines.

Another clear benefit (from Apple's point of view) is that having soldered RAM has allowed Apple to effectively lock you in to paying extra for RAM up front if you need it or think you will need it.

Apple has massive profit margins on those RAM upgrades based on the ridiculous prices they charge in Macs today (even the iMacs that you can get standard SODIMMs for). So it clearly benefits Apple if users must pay them upfront for any memory upgrade (instead of being able to use SODIMMs like in previous times).

LPDDR4/X is not available as an SO-DIMM for any manufacturer, so this is a moot point, given where Apple intends to steer their mobile offerings. Apple bowed to pressure to offer 32GB and 64 GB options and that forced them to use regular DDR4. Apple Silicon Macs won’t be making that compromise.

Whether Apple continues using DDR4 SO-DIMMs is up in the air until an AS Mac mini or iMac ships. I suspect we’ll be treated to DDR5 for these two Macs as DDR4 DRAM is getting pretty long in the tooth. I also think we’ll probably get SO-DIMMs for these models. Time will tell though.

 

[awesomedeluxe]

Is it possible, apple to offer less power/performance in their new arm macbook pros?
I can't imagine it.

So, what is realistic to expect?

I don't think it's a problem until you get to the 16".

At that point it's not clear if Apple's chips scale up enough to beat 45W offerings from AMD and Intel. Single core performance will probably lag behind comparable Rocket Lake and Zen 3 offerings. Apple will be a year ahead in manufacturing tech and probably use higher-speed memory (LPDDR5 or HBM2). A lot will depend on whether Apple can engineer a good graphics solution, and it helps that they've kept the bar low by sticking to AMD parts instead of using nVidia's more competitive offerings.

At a minimum, a "unified" 16" would have 8 3GHz Firestorm cores, 32 Apple GPU cores, and up to 32GB HBM2. This machine would certainly perform well, and would be better than a like-sized Dell XPS in some areas but not others.

 

[Zackmd1]

Just throwing this out there for speculation....

Could Apple focus on mobile designs (small packages with 12-20 cores each) and simply add multiple processors together for something like a Mac Pro? So for instance the M14x used in the top spec 16" MacBook would be used without modifications in a Mac Pro. The Mac Pro just would include 4-8 separate M14x processors combined together through the logic board to result in a "monster" processor without needing to actually be a single SOC? That would allow them to focus efforts on smaller more efficient designs and just combine them together when more power is needed without as much concern for efficiency like in desktops. Maybe you could even have add-in PCIE cards for the Mac Pro that contain multiple additional M14x processors to further increase performance?

 

[leman]

At that point it's not clear if Apple's chips scale up enough to beat 45W offerings from AMD and Intel. Single core performance will probably lag behind comparable Rocket Lake and Zen 3 offerings.

Why do you think that? A13 currently beats both AMD and Intel in single-threaded performance at the same clock by at least 40%. If they can boost their new chips to 3.5 Ghz at the same IPC, the single-threaded performance will be excellent.

 

[Boil]

Could Apple focus on mobile designs (small packages with 12-20 cores each) and simply add multiple processors together for something like a Mac Pro? So for instance the M14x used in the top spec 16" MacBook would be used without modifications in a Mac Pro. The Mac Pro just would include 4-8 separate M14x processors combined together through the logic board to result in a "monster" processor without needing to actually be a single SOC? That would allow them to focus efforts on smaller more efficient designs and just combine them together when more power is needed without as much concern for efficiency like in desktops. Maybe you could even have add-in PCIE cards for the Mac Pro that contain multiple additional M14x processors to further increase performance?

I had a very similar thought on the run-up to the 2019 Mac Pro reveal. I was pitching a new Cube; four A-series APUs in the base model, with a backplane for three more daughtercards (each with four A-series APUs), for a total of 16 APUs in a fully loaded system. A mini Beowolf cluster of sorts...

But now I am of the LARGE APUzilla build (monolithic die or multiple chiplets in a package or whatever)

In the below, I would also wonder at having an expansion slot for a 'stripped-down' Apple APU, with the GPU cores being the focus, so a Cube user could expand GPU abilities?

Mac Pro Cube

64 P cores / 4 E cores / 96 GPU cores / 128GB HBM2e UMA / Six USB4 ports / Two 10Gb Ethernet ports

Apple 58" SuperUltraWide 8k4k Display (for the longest of timelines)

Apple Low-Profile Mechanical Keyboard

Apple Magic Mouse 3D

US$9,999.00

;^p

 

[leman]

Just throwing this out there for speculation....

Could Apple focus on mobile designs (small packages with 12-20 cores each) and simply add multiple processors together for something like a Mac Pro? So for instance the M14x used in the top spec 16" MacBook would be used without modifications in a Mac Pro. The Mac Pro just would include 4-8 separate M14x processors combined together through the logic board to result in a "monster" processor without needing to actually be a single SOC? That would allow them to focus efforts on smaller more efficient designs and just combine them together when more power is needed without as much concern for efficiency like in desktops. Maybe you could even have add-in PCIE cards for the Mac Pro that contain multiple additional M14x processors to further increase performance?

That is not necessarily easier to do. You still need to coordinate all these modules somehow. Besides, multi-GPU setups are not trivial to utilize efficiently. The kind of configurations are ok for certain tasks (like servers), but they are not necessarily the best choice for hybrid Pro workflows, where you want both strong single-threaded and unbalanced multi-threaded performance.

For Mac-Pro levels of chips, I believe that a System-On-Package might make the most sense (like described here: #116 )

 

[Boil]

That is not necessarily easier to do. You still need to coordinate all these modules somehow. Besides, multi-GPU setups are not trivial to utilize efficiently. The kind of configurations are ok for certain tasks (like servers), but they are not necessarily the best choice for hybrid Pro workflows, where you want both strong single-threaded and unbalanced multi-threaded performance.

For Mac-Pro levels of chips, I believe that a System-On-Package might make the most sense (like described here: #116 )

The Zen 2 Ryzen & Threadripper CPUs are chiplets on a package, bundled on a common substrate, CPU chiplets and an I/O chiplet.

There was the Kaby Lake-G, which was an Intel monolithic CPU, an AMD Vega monolithic GPU, & (one?) HBM2 stack; all on the same package.

AMD is doing chiplets, and there are rumors for Zen 4 to have chiplets & stacked memory on the package. I was always of the mind that getting a GPU in the mix would make a powerful APU.

Maybe Apple could do something similar, I dunno, I am not an engineer of any sort, I just read stuff on the internet & act like I might know a bit about the BS I type into these forums! ;^p

 

[awesomedeluxe]

Just throwing this out there for speculation....

Could Apple focus on mobile designs (small packages with 12-20 cores each) and simply add multiple processors together for something like a Mac Pro? So for instance the M14x used in the top spec 16" MacBook would be used without modifications in a Mac Pro. The Mac Pro just would include 4-8 separate M14x processors combined together through the logic board to result in a "monster" processor without needing to actually be a single SOC? That would allow them to focus efforts on smaller more efficient designs and just combine them together when more power is needed without as much concern for efficiency like in desktops. Maybe you could even have add-in PCIE cards for the Mac Pro that contain multiple additional M14x processors to further increase performance?

My question with this approach is what happens to the GPU. If the MBP is using an APU where the on-die GPU cores are its primary GPU, it doesn't really make sense to put a bunch of these on a package for a Mac Pro. If you have an 8 Firecore part, sure, put four of those dies on your Mac Pro package. But if you have an 8 Firecore, 4 Lightning Core, 32 Apple GPU core part I don't think putting a bunch of them on a package for a Mac Pro is a very good idea.

I think it's much easier to design APUs that can be reused throughout the iPad and MBP line. The primary graphics array for the iPad just become the low-power graphics for notebooks, and that's OK. I think the Mac Pro will need a substantially different approach from the MBPs, different enough that there may not be much shared between them.