Should have moved to AMD....
AMD now have the ST and MT lead in desktop. Apple Sillicon is already behind. Now the A16 barely had any GPU/CPU pref improvements.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP All Models Mac Pro based Apple Sillicon will be a mistake and A16 confirmed it for me.
- Thread starter exoticSpice
- Start date
- Sort by reaction score
Should have moved to AMD....
AMD now have the ST and MT lead in desktop. Apple Sillicon is already behind. Now the A16 barely had any GPU/CPU pref improvements.
Seems a bit premature as we have no actual information on the Apple silicon Mac Pro capabilities.
It does matter the GPU and CPU and RAM will all be slow than AMD and Nvidia and expensive to boot sadly.
Look for laptops AS is great but for a workstation nah.
| Generation | Pref/Clk (Gbench/GHz) | % y-over-y | Node | Size (mm^2) | Pref/Clk/mm2 |
| A11 | 393 | -- | 10nm | 88 | 4.5 |
| A12 | 450 | 14.5% | N7 | 83 | 5.4 |
| A13 | 504 | 12% | N7 Enhanced (N7P) | 98 | 5.14 |
| A14 | 528 | 5% | N5 | 88 | 6 |
| A15 | 545 | 3% | N5P | 108 | 5.04 |
| A16 | 543 | -1% | N4 (N5 family) | ? | probably close to same ( ~5) |
| A17 (projection) | likely higher | likely higher. extremely likely positive. | N3 highly likely | smaller | so higher due to smaller die. |
Geekbench isn't measuring everything that is going on. As much as a bunch of hyper Apple fans want to sping the notion that the logic designers at Apple are all super geniuses and everyone else just has "second tier" folks on their teams as the pure source of Apple's previous leads is just smoke. When Apple ran out of node increase uplift and only did minor micro-architectural changes the increases slowed.
The node increase between A14 and A13 was a big lift. Both because Apple got to apply transitors to other functional units besides the CPU cores and gave them a window to do some substantive tweaks to implementation (significantly bigger caches , etc.) . Likewise going from 10nm to N7 got a large assist from the node change.
A15 most of the budget was spent elsewhere than the CPU complexes. ( GPU , camera/video/ProRes , etc. )
A16 same thing. Only the node change is small. There is very little performance or power uplift with N4 (or N5P). It is mainly smaller. There are likely some optimizations that eek out a bit more power savings when not single threaded drag racing for short bursts of speed. With an always on screen how the SoC performs when not 'hot rodding' is equally has important for a mobile phone with limited battery.
Even iff Apple doesn't do a medium-large microarchtectural changes with A17 and just lean on doing a very good first implement with N3's new unique features , it is extremely likley that the y-over-y trend is going to reverse. Full node moves previously gave significant uplift. Not good reason to think it won't here again.
AMD has no competitive solution for hte Mac laptop line up at all. That 70% of what Apple sells. With the reuse of the laptop SoC in iMac , Mini , Mac Studio ..... that pragmatically closer to 97% of what Apple sells. AMD isn't a viable Mac ecosystem solution.
A significant part of AMD's progress here is process node and chip packaging where for 2022-2023 they are going to be more even with Apple for a while.
Apple is going to get off of "N5-family" before AMD does. Zen4 is only going to get to N4 around end of 2023 - early 2024. Apple will likely be "gone" from N4 by then. Zen5 isn't going to get to N3 until end of 2024 - early 2025. Apple will be on 2nd or 3rd generation optimizations for N3-family. Apple is not likely to be falling far behind.
Probably won't have single threaded drag racing crown, but in all-around performance , on real apps ... not that far behind.
If bound the core count for AMD/Intel to less than 40 then Apple is more completive. The highest end of the MT numbers is just as much about number of cores as it is about quality of cores. Apple is highly unlikely to throw more cores at the SoC than macOS can handle. Apple isn't going to compete in the > 64 space. And there is little good rational reason to try to press close to 64 threads sooner rather than later to give themselves more long term headroom.
I suspect that the next Mac Pro is N3 based. ( the general M2 architecture abstractions ported down to N3 to control complexity. Whether Apple calls that M2 or M3 isn't material). If so that is a signficant contribution to the delay. A17 is so incremental because they are more busy putting effort into a major shift that is difficult to do. (N3 is not "easy" to work with. Requires big changes to design tools to effectively use to its best . New design rules. etc.)
If the next Mac Pro is 'stuck' on N4 , then yes Apple has got 'issues' even beyond perhaps having a "one-slot-wonder' I/O limitations. Where is Apple is way, way , way behind at this point is in general I/O performance , not CPU performance. They haven't even entered the 'race' on ECC performance. Geekbench ST obsession is a major distraction from the most critical issues in the Mac Pro space.
LPDDR5/5X is not that more expensive than GDDR6/7. And if using a wider memory bus it isn't that much slower in aggregate bandwidth either.
Where Apple is going to get into trouble is when AMD/Nvidia level up on GPUs N5 (+N6) / N5 to level up on much bigger cache increases and more specialized function units that Apple has leverged earlier to offset some of their design trade-offs ( avoid high power hungry GDDRx and just be wider and hold more data closer to the cores).
Mid range Apple will likely hold their ground better.
The A16 isn't going to have any of the significant I/O elements that the Mac Pro is going to need. And certainly Geekbench isn't going to measure it. So you really don't even have a workstation measurement metric here that is worthy of evaluating the product area at all.
I stand corrected todays Geekbench result is good for the A16.
If Apple managed this while decreasing power usage its good considering they did not focus on pref but on efficiency.
If Apple managed this while decreasing power usage its good considering they did not focus on pref but on efficiency.
A 0.4% in GB score difference and it dramatically moves in to the "is good" category is demonstrative of the mania of reducing everything down to a mindless number can get. GB as a measurement tool has no where near that kind of measurement accuracy. The 'error bars' on GB is bigger than that even on same fundamental architecture comparisons. Across major architectural differences, it is larger still.
That new post I was referring to the MT score, which is a good improvement.
Yes the ST did not improve that much, as I said if Apple's A17 does not have real meaningful IPC improvements then its time to say then Apple has slowed down in the CPU space
I haven't gone through the entire history of Apple CPUs - but in general Apple tends to follow a tick/tock cycle like the rest of the industry.
Generally Apple/Intel/AMD/Samsung/Qualcomm do not see massive gains year over year. They follow a cycle of minor gains and major gains. So nothing weird is going on here.
Apple being able to ship (almost) year over year is already a pretty big upgrade over Intel who's had trouble doing that. We might actually get yearly Mac updates again.
Seems there are trade offs when Apple went with an older concept of full integration. There is no room for after the fact modularity unless it is external. Any thoughts one what, other than power and ports , would separate the Pro from other systems? It seems getting additional items such as audio and video processing should do better than being eternally attached by Thunderbolt or USB-C 3.2 gen 2. What's your take on what should set the Pro model apart?
I'd prefer a Pro model with the same expansion options we have now.
I don't think it's a "full integration" problem. Different chips are expensive to produce. We're yet to see how much money Apple wants to throw at a unique Mac Pro architecture.
Ok A14 was Apples tick and tock was A15. The A16 should have been the tock but really it's a really poor improvement in IPC and also the GPU did not improve at all in A16.
NVidia is expected to announce a 150% performance improvement for the 40-series over the 30-series, which had a similar improvement over the 20-series. For perspective on how badly the wheels are falling off at Cupertino.
The problem there is if Apple throws too much money, because they are not going to 'eat' the costs. They will pass them on. So if Apple throws 'crazy' large money at the Mac Pro the entry price point of the Mac Pro would follow that path. If the Mac Pro entry price jumped another $3K from $5,999 to $8,999, then the product would be on a pricing death spiral.
Lots of folks want "everything and the kitchen sink" as features right up until you tell them they have a pay a large amount for all of that. Then many of them disappear as actual real, paying customers.
Two whole years into the transition and to all visible appearances Apple has spend about $0.00 on dGPU driver support. No megabucks spending there or they are jumping through serious hoops to hide it.
For the Mac Pro 2019 , Apple spent a substantial amount of money on Apple specific custom stuff. T2 addition, which makes the MP more like the most other Macs in the line up. MPX to integrate Thunderbolt in similar fashion it was implemented in the rest of the Mac line up ( and make power wires go away like the rest of the Mac line up). The other expensive stuff they used was largely off-the-shelf. So yes expensive, but Apple was really just selecting not really 'throwing" ( charged higher prices at least as much as investing gobs of R&D dollars) Intel Xeon W . Dual input provisioning Plex PCI-e switch. AMD "Pro" GPUs with Infinity Fabric.
P cores , E core , GPU cores , Thunderbolt controllers , UltraFusion , Memory controllers , etc .... those are likely going to be investments in lower differentiation from the rest of the Mac line-up ; not some kind of huge fork.
It is far more likely about how big of a "sidecar addition' is Apple going to put on the shared SoC infrastructure architecture, than some "how much money for major differences in architecture.". Bolting on a substantive PCI-e v4 controller complex would cost more money , but not really going to be mimicking Intel Xeon W and AMD Threadripper architecture strategy.
This sequence was lots more like the phase after Intel's Tick/Tock stopped. A14 ( Intel 14nm ) , A15 ( minor uarch tweak and 14nm++ ) , A16 ( minor uarch tweak and 12nm+ (a shrink but smallish one) )
Similar issues over in Watch SoC space.
"... The CPU within Apple's S8 SiP carries the same T8301 identifier as the CPU in the S6 and S7 chips that debuted in the Apple Watch Series 6 and Apple Watch Series 7. This explains why Apple has only compared its S-series chips in recent years to the S5 chip or earlier. ..."
https://www.macrumors.com/2022/09/12/apple-watch-s8-chip-features-same-cpu-as-s6-and-s7/
Apple has made substantive adjustments to the non CPU-core elements to the SoC , but the CPU cores are not a blocking factor.
For the Mac Pro it is very similar situation, cranking up the Geekbench CPU scores 14% isn't going to help if all that is provisioned by the SoC is four x1 PCI-e v4 lanes. Apple doesn't have CPU core performance problem , they have an overall aggregate general utility I/O bandwidth problem. A big one. If Apple rolls out some relatively under provisioned one-slot-wonder then it will largely be a fail. Better Geekbench scores isn't going to save that.
If Apple lets themselves get caught in the 'trap' the iPhone is in ( new A-series every 12 months ) then probably going to see a steady stream of this more incremental stuff coming out. There are major factors that are going to spin out of Apple's control. First, the breadth of their competitors is going to go way up. Making just a better iPhone and Watch SoC then others is easier than making a better iPhone, Watch , MBA , Mac Studio , Mac Pro SoC than others. Throw the better cellular modem in there also and kind of have drifing into the "Don't fight a land war in Aaia in Risk" situation. Apple doing a much broader set of SoC than they have over most of last 10 years is going to be a o more dificult task. Lots more "walk and chew gum at the same time" required.
Second, TSMC ( and other bleeding edge fab producers) aren't going to roll out big node shrikage/shifts every 12 months. The cycle is likely going to get longer. Apple has plucked off all the low hanging fruit in IPC tricks on general code execution. Some other vendors are going to see some bigger 'catch ups' because they were riding further behind ( slower uptick to bleeding edge nodes and off more problematical fab families). E.g., Nvidia going from Samsung 8nm to TSMC N4 . There is a huge uplift there that will be hard to repeat with a N4 -> N3 move.
No, M1 was the Tick. M2 is the Tock.
Apple's 3nm chips will be the next Tick.
Trying to map that back to the A series is difficult. A series is optimized for low power smartphone devices and has diverged somewhat from the M series architecture.
Ignoring the feature-by-feature questions on a Mac Pro... It's too early to make this judgement. If the new Mac Pro is on a 3nm process, they'll likely hold out against Nvidia. If the Mac Pro is still on a 5nm process - things are going to be ugly for them.
No one is quite sure what process M2 Max/Ultra/Extreme/Ludicrous will be on. It's possible it's on a smaller process, or maybe it uses the same 5 nm as the base M2.
If Apple moves to 3 nm, they'll have a very similar improvement. The A16 did not move to 3nm, but it's very possible the reason why is because Apple is holding 3nm production back for Macs instead of iPhones.
What Nvidia is doing is pumping up the watts. The rtx 3090 is 350 watts and the rtx4090 is rumoured to be 450 watts.
All in all it will be a 70-80% increase at the same TDP but considering it's been 2 years since rtx 3000 has been released its not surprising for me.
People have been saying Cupertino has been off it's wheels since it's inception but meh got bored of the negative people.
I think the GPU power of the AS Mac Pro is the real question I have, what's the best offering Apple can make in that area? It seems like the GPU is weakest part of the whole AS lineup. An NV 3090 is about a $1200 card and falling these days, do we need to drop $10K in the apple store to get even close to that compute power? I should likely get a Mac Studio.. and was hoping to hear something at the iphone14 launch about the AS MP, I'm just too curious about what the AS MP offering will actually be on the gpu side.. the path apple takes.
Do we have any idea of an upcoming event where we possibly would hear anything about the AS MP?
Do we have any idea of an upcoming event where we possibly would hear anything about the AS MP?
Ok, you guys who have the prototype, I don't want you to breach your nondisclosure agreement, but if you could just post:
= it's fast, somewhat upgradable and expandable (storage is upgradable and it'll accommodate a few PCIE cards) and overall pretty good;
👏 = it's amazingly fast, upgradable and expandable (storage, multiple fast PCIE lanes, dGPU, the whole shebang) and I already put down my deposit;
= it's better than the Studio Ultra (you can upgrade the SSD's) but that's about it;
= it's closed box (not upgradable, not expandable) basically a faster version of the Studio Ultra;
🤮 = it's a total mistake that will likely bankrupt Apple.
We await your hints.
= it's fast, somewhat upgradable and expandable (storage is upgradable and it'll accommodate a few PCIE cards) and overall pretty good;👏 = it's amazingly fast, upgradable and expandable (storage, multiple fast PCIE lanes, dGPU, the whole shebang) and I already put down my deposit;
= it's better than the Studio Ultra (you can upgrade the SSD's) but that's about it; = it's closed box (not upgradable, not expandable) basically a faster version of the Studio Ultra;🤮 = it's a total mistake that will likely bankrupt Apple.
We await your hints.
Any new Mac pro with any M chip is not a true Mac pro.
Mac pro stands and always has for upgradability.
CPU Socket Upgradable.
GPU via Pcie Upgradable.
Sound card via Pcie Upgradable.
Storage options via Pcie Upgradable.
Memory to user's needs. Upgradable.
More than 1 operating system can be booted ( windows ) Upgradable.
M chip Mac pro
Fixed CPU any failure means new board.
Fixed Memory again any failure means new board.
Soldered Storage ( external available via thunder bolt ) internal storage fails new Board.
Fixed GPU any failures with that and again new board.
No external Egpu support.
No other operating system support
Possible 1 16X Pcie slot, but what hardware will it support? sound card possible if driver supported.
All this requires you to have apple care, because any failure will be costly with new Board to original config. but after 3 years its risky as expensive to replace via your own pocket to original config. Do you really want or need to buy a new Mac pro every 3 years just to have warranty alone and an upgrade path.
If any of the above M chip spec ends up being true with non upgradable path for CPU, GPU, Memory it will not be a true Mac pro.
What the new Mac pro should be
AMD thread ripper CPU upgradable.
DDR 5 memory Upgradable.
AMD 7000 series Upgradable to 8000 series with in 3 years.
Storage upgradable via TB 5 Pcie card offering 80 GB transfer speeds.
Pcie Slots for sound and storage cards.
Different operating system options for those who need it. Mac OSX isn't best at everything.
Now this is what i call a real Mac pro, versatility and upgradable.
Plus Apple and Nvidia should stop behaving like children and offer a Nvidia upgrade path. Put your toys back in the pram and offer what people would want in there Mac pro, a true professional option.
Of course my own personal thoughts.
Mac pro stands and always has for upgradability.
CPU Socket Upgradable.
GPU via Pcie Upgradable.
Sound card via Pcie Upgradable.
Storage options via Pcie Upgradable.
Memory to user's needs. Upgradable.
More than 1 operating system can be booted ( windows ) Upgradable.
M chip Mac pro
Fixed CPU any failure means new board.
Fixed Memory again any failure means new board.
Soldered Storage ( external available via thunder bolt ) internal storage fails new Board.
Fixed GPU any failures with that and again new board.
No external Egpu support.
No other operating system support
Possible 1 16X Pcie slot, but what hardware will it support? sound card possible if driver supported.
All this requires you to have apple care, because any failure will be costly with new Board to original config. but after 3 years its risky as expensive to replace via your own pocket to original config. Do you really want or need to buy a new Mac pro every 3 years just to have warranty alone and an upgrade path.
If any of the above M chip spec ends up being true with non upgradable path for CPU, GPU, Memory it will not be a true Mac pro.
What the new Mac pro should be
AMD thread ripper CPU upgradable.
DDR 5 memory Upgradable.
AMD 7000 series Upgradable to 8000 series with in 3 years.
Storage upgradable via TB 5 Pcie card offering 80 GB transfer speeds.
Pcie Slots for sound and storage cards.
Different operating system options for those who need it. Mac OSX isn't best at everything.
Now this is what i call a real Mac pro, versatility and upgradable.
Plus Apple and Nvidia should stop behaving like children and offer a Nvidia upgrade path. Put your toys back in the pram and offer what people would want in there Mac pro, a true professional option.
Of course my own personal thoughts.
None of this will give the next Mac Pro the expected ability to handle GPU workloads with say half a terabyte of unified memory.
You're living in the past.
If you want to run different operating systems, get a machine for it. Or virtualise it so you can run both at the same time.
Linux runs amazingly well on M series in a VM. Like... better than native on my Ryzen desktop in some ways, because the metal drivers make it so smooth.
Same with Windows 11 ARM.
Rebooting is stupid if the VM performance is there, and it is.
Last edited:
First obstacle: a Mac Pro that can be configured at time of purchase to compete with 2X 4090 (or corresponding Pro NVIDIA cards) and 64 core Threadripper. Second obstacle, upgrades of AS modules.
The Mac Pro you describe is for the enthusiast who rather buy CPU, GPU, SSD and RAM somewhere else than Apple. A corporate pro would configure, buy at Apple, use for three years (or until unusable) and then repeat that cycle. Who do you think Apple wants as customer?
Apple wants to control everything including upgrade paths so you end up buying a new mac pro every 3 years so you have warranty and the latest speed bumps. Original Mac pro's could be upgraded and thats what pro's want, they dont want to have to reload everything every 3 years if a simple upgrade is possible. GPU or CPU or sound card which ever they need.
Apple might want the buy new every 3 years customer, but that is not an option for many as it would be to expensive and a waste of some of the hardware they might well be happy with performance wise that they already own and paid for.
The thing is by the time apple get any where near the performance of 2 X 4090 gpu's in 3 years time Nvidia will be already sporting the 5090 or higher. same with AMD the route of OSX only machines with new OSX version's every year is getting tiresome. its like the iron curtain falling around the Apple ecosphere to tie you into Apple hardware only.
If this is the way Apple are going i will not want to be behind any iron curtain of apple hardware apple OSX only.