Discussion Apple Silicon SoC thread

[BorisTheBlade82]

I'm not sure you're comparing apples to apples in the single thread efficiency score. How do the chips compare in multi-threaded? Comparing the power consumption, especially single thread, between different architectures (especially Apple silicon) can be tricky because they report things in different ways. There's also significantly more IO support on the x86 side compared to M1 which will show the largest impact on idle and single threaded power consumption if you are taking package power on x86 as your measurement.

Well, it was not me who measured M1, but rather Andrei and @sallymander. Please read #42ff. for further details. But how I see it, it is pretty much "apples to apples". And who does really care if the advantage is 400% or 500%?

 

[blckgrffn]

Whereas,

The accuracy of previous rumors about the dies were almost spot on,

and

given the rumors implying that Apple will lash 2 and 4 M1 Max dies together

Let it be stated on this day

With full awareness of the unlikely nature

and acceptance of the additional costs to be passed on to me and then some

I claim desire of a Jade 2C Chop for Mac Minis and 16 inch laptops.

Even a cut down double M1 Pro 12 P + 4 E + 28 GPU and I’d be happier than a 5-year-old in a bounce house.

I mean, I think the thing is these dreams have prices.

Double CPU Mac Pro is going to start where? $5k? Quad? I mean, Apple shareholder here and all that, love the margin, but these are specific machines for specific needs.

M1 Pro Mini is going to start at what, $1500 and work its way up from there? Might as well.

 

[Eug]

M1 Pro Mini is going to start at what, $1500 and work its way up from there? Might as well.

I am hoping for US$1299 retail... but for 8-core (6+2) CPU and 14-core GPU Mac mini with updated case design, 512 GB SSD, and 16 GB RAM. However, that would mean $1499 for 10-core (8+2).

 

[blckgrffn]

I am hoping for US$1299 retail for 8-core (6+2) CPU and 14-core GPU Mac mini with updated case design.

256GB of non-expandable storage?

I know, I know, it actually has ports that support real options but ugh I hate dongles/external stuff like that.

I feel like the "optimal" Mini Pro is going to somehow cost north of $2k.

I think my favorite part of the new Macbooks is Apple finally getting tired of putting the screws to Macbook Pro owners and giving them F keys and real useful ports back. Hahaha, they might as well have been singing that "It's OK, it's OK, you're welcome!" song while announcing that.

 

[guidryp]

I am hoping for US$1299 retail... but for 8-core (6+2) CPU and 14-core GPU Mac mini with updated case design, 512 GB SSD, and 16 GB RAM. However, that would mean $1499 for 10-core (8+2).

Why does it need an updated case? Right now a current M1 Mini 4core/16GB/512GB is $1099 USD.

I am betting they charge more than $200 to move to the M1 Pro chip.

 

[The Hardcard]

I mean, I think the thing is these dreams have prices.

Double CPU Mac Pro is going to start where? $5k? Quad? I mean, Apple shareholder here and all that, love the margin, but these are specific machines for specific needs.

M1 Pro Mini is going to start at what, $1500 and work its way up from there? Might as well.

I don’t see a second M1 Pro as being double the full machine price, but yes it would be a lot higher. A laptop that could unplugged approach Threadripper performance? Worth it for me.

 

[blckgrffn]

I don’t see a second M1 Pro as being double the full machine price, but yes it would be a lot higher. A laptop that could unplugged approach Threadripper performance? Worth it for me.

I am only adding $1500 from a Macbook 16" With a Max, I brain farted on that. I can't imagine them bothering with the M1 Pro in desktop "Pro" chassis. M1 Max or bust (times 2-4x, based on the rumors I've observed here).

The rest of that is just the premium because they can. You want it, pay for it. They'll likely be supply constrained anyway, and these are essentially the "Server" type margins that are awesome.

 

[Eug]

256GB of non-expandable storage?

I know, I know, it actually has ports that support real options but ugh I hate dongles/external stuff like that.

I feel like the "optimal" Mini Pro is going to somehow cost north of $2k.

I edited my post above as you were posting yours.

$1299 retail for 16 GB and 512 GB, with base M1 Pro. I don't think they will offer a 256 GB version.

Why does it need an updated case?

Well, it's ancient, and due for a redesign. It hasn't been updated since 2010. Right now it's just mostly empty space, and it's not needed because I'm 100% sure they won't have anything beyond M1 Max.

From a practical point of view though, some people have complained about the WiFi and Bluetooth strength in that all metal case, so a redesign may be in order to correct that.

Right now a current M1 Mini 4core/16GB/512GB is $1099 USD.

I am betting they charge more than $200 to move to the M1 Pro chip.

The M1 Mac mini was $100 cheaper than the corresponding Intel model at launch IIRC.

They increased the pricing on the 14" MacBook Pro by $200 vs. the corresponding 13" Intel model but that's because it's a brand new design with new mini-LED screen and bigger screen. However, AFAIK, they only increased the 16" model by $100, which makes sense since it's the same size screen.

So, given all that background, I would expect the 16/512 base model M1 Pro 8-core to be the same US$1299 retail as the current Intel Mac mini, or at worst $1399, unless they radically redesign it with some sort of new awesome features we aren't predicting.

 

[Hitman928]

Well, it was not me who measured M1, but rather Andrei and @sallymander. Please read #42ff. for further details. But how I see it, it is pretty much "apples to apples". And who does really care if the advantage is 400% or 500%?

The problem is, assuming the power measurements are even perfectly equatable, is that you are basically penalizing the x86 cores for supporting more functionality in their IO which increases package power for 0% performance improvement. You could argue that the M1 is just better for not supporting these things, but that doesn't really get at the heart of the argument of the x86 designs versus M1. While certainly there is overlap, a Cezanne die and an M1 die have different design targets. It would be like comparing a 5950x and an M1 and saying, what is the point of the M1 when the 5950x is so much faster in MT and can have greater efficiency than the M1 (by your own numbers). If we look at this comparison, it seems like Apple is actually behind, not only drastically in absolute performance but it can't even win in efficiency. Now, I don't think anyone would think this is a fair comparison, so why is an essentially stripped down M1 (comparatively) which is sacrificing certain features compared to the competition to improve efficiency a valid comparison to show that M1 is hundreds of percent away from any x86 competitor? Neither comparison makes much sense.

I'm not arguing against gathering the data and discussing it, but there needs to be some nuance in that discussion. The M1 line from Apple are amazing chips and honestly, if I didn't hate the Mac software eco system, one of the new macbooks would probably be my next laptop. But we need to not get carried away and pretend like Apple is using alien technology here. If we take the 5900HS and M1 numbers from your multi-threaded chart and assume that cinebench scales similarly to what we've seen in Geekbench, then we would get ~1.6x performance improvement for 2x (according to Apple) power consumption increase. We'd then be comparing an 8p/2e core M1 against an 8c/16t core Cezanne which is a pretty good comparison. That would be good for around a roughly 30% MT efficiency lead for the 8p/2e M1 over the 8c/16t Cezanne. Certainly an appreciable lead, but nothing crazy, especially when the node difference is taken into account.

Even in ST, where I expect Apple to continue to be the efficiency leader, a move for AMD to N5 would be a very different story than it is today. With the new node density, you should be able to get 20%+ IPC increase over Cezanne. That would allow the theoretical (I'll call it) Zen 4 to drop the frequency 20+% without losing any performance which would be a large power savings and then you have the additional N5 power savings to add as well. I think this theoretical Zen 4 core on the same process node as M1 could just about match it in ST efficiency. This last paragraph though is just speculation on my part, we'll have to see what AMD has in store with the real Zen 4.

 

[Gideon]

I'm now far less convinced that any additional DDR controllers will be forthcoming though. For a high-end iMac, the 2C maxing out at 128GB of LPDDR5 is probably fine, and depending on the design, they might even be able to jam the 4C in there. On the other hand, a Mac Pro with the 4C would max out at 256GB, which is a lot less than the current Intel Mac Pro. Perhaps this will go in the rumored smaller Mac Pro, which can be smaller because it sheds the DIMM slots and most of its PCIe slots. And maybe the big Mac Pro gets an Icelake refresh to keep that market segment satisfied for a couple years until Apple does their next round of pro chips. I guess we'll know in another 8 months.

I seriously doubt they'll add any DIMM slot s now that CXL is a thing.

IMO it's far more likely they just offer CXL add-on-cards for those that need more than 256 GB of ram. Like the one Samsung announced:

The third type is perhaps the one we’re most interested in today: memory buffers. Using CXL.memory, a memory buffer can be installed over a CXL link and the attached memory can be directly pooled with the system memory. This allows for either increased memory bandwidth, or increased memory expansion, to the order of thousands of gigabytes.

I mean, sure the latency will be somewhat worse, but still totally usable. Workstation users needing more than 256 GB at absolute lowest latency are quite few and far between anyway (and certainly not the usual mac pro customers).

And depending on the LPDDR5 modules - Couldn't they also not fit 512GB (with the most premium modules) for people that need a bit more than 256 and absolutely require the best latency and all the memory bandwidth ... for an astronomical price of course.

 

[guidryp]

Well, it's ancient, and due for a redesign. It hasn't been updated since 2010. Right now it's just mostly empty space, and it's not needed because I'm 100% sure they won't have anything beyond M1 Max.

Ancient is Meaningless. Either there is a problem to be solved or there isn't . It's really not significantly more empty than when it had an Intel CPU. I know some where calling for a smaller mini, but note that the Macbook Pro 16 got larger/heavier, than the Intel model it replaces. I think the current Mini is a lot more elegant the Intel NUCs. I really hope they don't turn the Mini into a NUC with a power brick.

From a practical point of view though, some people have complained about the WiFi and Bluetooth strength in that all metal case, so a redesign may be in order to correct that.

So it needs an antenna redesign.

The M1 Mac mini was $100 cheaper than the corresponding Intel model at launch IIRC.

They increased the pricing on the 14" MacBook Pro by $200 vs. the corresponding 13" Intel model but that's because it's a brand new design with new mini-LED screen and bigger screen. However, AFAIK, they only increased the 16" model by $100, which makes sense since it's the same size screen.

So, given all that background, I would expect the 16/512 base model M1 Pro 8-core to be the same US$1299 retail as the current Intel Mac mini, or at worst $1399, unless they radically redesign it with some sort of new awesome features we aren't predicting.

Notably though, there is now an M1 Mini to compare with and Apple will want space between them. Also the M1 Pro 16" is only $100 more than a model with discrete GPU (which will increase costs quite a bit), and the next mini tier has no discrete graphics, and there is lot more costs with having both parts that they skipped in the Mini.

But yeah $1299 to $1399 is reasonable, I just lean a bit more toward the latter, unless they cut the similarly configed M1 $100, then I would agree on $1299 being more likely.

 

[Doug S]

I can almost guarantee you that there will not be an additional I/O chip.

After studying the die shot, and noting the complete lack of I/O on the bottom edge, I'm thinking that there's still more to this floor plan that we haven't seen yet. In other words, M1 Max is itself a chop, and Jade 2C and 4C will use a slightly longer version. Apple should be able to comfortably fit three PCIe Gen5 x16 interfaces for CXL interconnect links along the bottom edge of the die. Since Apple stuck with a conventional organic substrate and (slightly crazy) LPDDR memory, there is no issue with scaling up the package to accommodate 2 SoC dies and 8 memory packages, or even 4 SoC dies and 16 memory packages.

I'm now far less convinced that any additional DDR controllers will be forthcoming though. For a high-end iMac, the 2C maxing out at 128GB of LPDDR5 is probably fine, and depending on the design, they might even be able to jam the 4C in there. On the other hand, a Mac Pro with the 4C would max out at 256GB, which is a lot less than the current Intel Mac Pro. Perhaps this will go in the rumored smaller Mac Pro, which can be smaller because it sheds the DIMM slots and most of its PCIe slots. And maybe the big Mac Pro gets an Icelake refresh to keep that market segment satisfied for a couple years until Apple does their next round of pro chips. I guess we'll know in another 8 months.

One other thing to note about this strategy is that although there will be reduced bandwidth / increased latency between dies, Apple probably won't have to reduce clock speeds compared to the M1 Max at all. TDP will be around 340W, but whatever, that's still less than a single GA102 in an NVIDIA RTX 3090.

Yeah you might be right that there's another chop to this we haven't seen yet, and the bottom section could integrate on chip fabric and a DDR5 controller.

Sure its possible they might figure the Ice Lake Mac Pro will handle the people who want lots of RAM, or as others have suggested connecting it via CXL. But why? If doing DDR5 controllers and DIMM slots required some sort of exceptional engineering effort I could understand the idea they might wait until the M2 generation to add them, or take a shortcut by using CXL, but considering what they've done already adding DDR5 slots is child's play.

So could possibly be a reason why they would not, unless one believes they'll have 256 GB as the top end and those who need more DRAM are simply SOL.

 

[Doug S]

I seriously doubt they'll add any DIMM slot s now that CXL is a thing.

IMO it's far more likely they just offer CXL add-on-cards for those that need more than 256 GB of ram. Like the one Samsung announced:


I mean, sure the latency will be somewhat worse, but still totally usable. Workstation users needing more than 256 GB at absolute lowest latency are quite few and far between anyway (and certainly not the usual mac pro customers).

And depending on the LPDDR5 modules - Couldn't they also not fit 512GB (with the most premium modules) for people that need a bit more than 256 and absolutely require the best latency and all the memory bandwidth ... for an astronomical price of course.

OK explain why they would use CXL? What's the benefit to Apple versus adding DDR5 controllers? They need to add capability to the M1 Max (either a not yet revealed even bigger die that adds more on the bottom, or an I/O chip) and slots to the board either way.

There has to be a reason why they would do this, and "I believe Apple hates DIMM slots and will do anything to avoid them" isn't a reasonable answer as far as I'm concerned.

 

[Abwx]

There was a cinebench R23 test floating around :

These Apple M1 Cinebench R23 scores are quite impressive for a 10 W processor

Looks like ditching Intel for in-hose ARM-based chips was quite the power play from Apple. The new M1 processors appear to trade blows with AMD's / Intel's 15 W processors. According to the latest Cinebench R23 tests, the M1 can be slightly faster than the AMD Renoir-U APUs in single-core tests...

www.notebookcheck.net

Cinebench R23: CPU-Ergebnisse mit AMD und Intel

Maxon hat Cinebench R23 veröffentlicht. Wie schnell welcher Prozessor im CPU-Benchmark ist, kann die Community im Schwarm ermitteln.

www.computerbase.de

7500pts at 30W while a Zen 3 5980HS does 10600 pts@35W, so clearly efficency is not that good given that it s 5nm vs 7nm.

 

[Hitman928]

@Abwx The M1 uses ~15W for ~7500 pts.

 

[guidryp]

I can almost guarantee you that there will not be an additional I/O chip.

After studying the die shot, and noting the complete lack of I/O on the bottom edge, I'm thinking that there's still more to this floor plan that we haven't seen yet. In other words, M1 Max is itself a chop, and Jade 2C and 4C will use a slightly longer version. Apple should be able to comfortably fit three PCIe Gen5 x16 interfaces for CXL interconnect links along the bottom edge of the die. Since Apple stuck with a conventional organic substrate and (slightly crazy) LPDDR memory, there is no issue with scaling up the package to accommodate 2 SoC dies and 8 memory packages, or even 4 SoC dies and 16 memory packages.

I'm now far less convinced that any additional DDR controllers will be forthcoming though. For a high-end iMac, the 2C maxing out at 128GB of LPDDR5 is probably fine, and depending on the design, they might even be able to jam the 4C in there. On the other hand, a Mac Pro with the 4C would max out at 256GB, which is a lot less than the current Intel Mac Pro. Perhaps this will go in the rumored smaller Mac Pro, which can be smaller because it sheds the DIMM slots and most of its PCIe slots. And maybe the big Mac Pro gets an Icelake refresh to keep that market segment satisfied for a couple years until Apple does their next round of pro chips. I guess we'll know in another 8 months.

One other thing to note about this strategy is that although there will be reduced bandwidth / increased latency between dies, Apple probably won't have to reduce clock speeds compared to the M1 Max at all. TDP will be around 340W, but whatever, that's still less than a single GA102 in an NVIDIA RTX 3090.

The M1 Max already has 8 DDR channels. Typically these can be configured for DDR Memory slots, or LPDDR chips. Why would they need more than that?

If they do 4 x M1 Max, they will have 32 DDR channels, and that's freaking insane. One thing they don't need more of is DDR channels.

 

[Mopetar]

I wonder if they'd just go to GDDR or some other type of memory that offers more bandwidth for the desktop parts. I'm not sure if they'd want to solder the memory to the board instead of offering user upgradable memory, but if they do go the route of just sticking it on the board there's no reason not to go with something more exotic because they can just pass any extra cost on to their customers.

Of course part of the reason that we didn't see desktop updates is that they're sticking with user upgradable memory and they don't want to release anything before the memory is more widely available. If they're going with some kind of MCM design they might go the same route as AMD and design an IO die that acts as a central hub because it gets hard to route everything through separate chips and to scale that up beyond a certain point.

Whatever they end up doing it will probably be pretty beastly.

 

[Eug]

Hmmm... I didn't realize, but the SSDs on these laptops are up to 7.4 GB/s. Wow.

And interestingly the dot pitch is exactly 0.1 mm, or 100 pixels per cm (254 ppi). That should make calculations easy.

The high impedance headphone compatibility is cool too.

And it can charge by either MagSafe or USB-C, which is great for those running powered USB-C dongles or USB-C displays.

They've basically thrown everything in there but the kitchen sink (except for FaceID).

 

[tempestglen]

There was a cinebench R23 test floating around :

These Apple M1 Cinebench R23 scores are quite impressive for a 10 W processor

Looks like ditching Intel for in-hose ARM-based chips was quite the power play from Apple. The new M1 processors appear to trade blows with AMD's / Intel's 15 W processors. According to the latest Cinebench R23 tests, the M1 can be slightly faster than the AMD Renoir-U APUs in single-core tests...

www.notebookcheck.net

Cinebench R23: CPU-Ergebnisse mit AMD und Intel

Maxon hat Cinebench R23 veröffentlicht. Wie schnell welcher Prozessor im CPU-Benchmark ist, kann die Community im Schwarm ermitteln.

www.computerbase.de

7500pts at 30W while a Zen 3 5980HS does 10600 pts@35W, so clearly efficency is not that good given that it s 5nm vs 7nm.

Wrong! M1 cpu consumes only 12-15w at full load while 5980HS is far higher than 35w at full load.

 

[Hitman928]

Wrong! M1 cpu consumes only 12-15w at full load while 5980HS is far higher than 35w at full load.

M1 can use just over 20W at full load, but Cinebench doesn't push the cores that hard.

 

[Eug]

We are inching closer to 13000 and 1.7X M1 my friends.

1783 / 12693



It looks like the real multiplier for M1 Pro/Max is probably going to be somewhere around 1.6X to 1.7X M1 for this bench.

It seems all the embargoed reviewers are doing their benchmarking. I wonder if they'll actually be allowed to post benchmarks this time. Usually Apple says they can't post synthetic benchmarks in their reviews and instead only allows them to say stuff like how smooth the video scrubbing is and stuff like that.

Also, it seems that Geekbench has been updated, since it's now showing as 3.2 GHz instead of the 24 MHz (!) it was reporting before.

 

[Mopetar]

What's the point in not allowing benchmarks to be posted if they're public on some third party website? Everyone finds out what they are anyways and there's nothing stopping non-reviewers from posting about them.

 

[Eug]

What's the point in not allowing benchmarks to be posted if they're public on some third party website? Everyone finds out what they are anyways and there's nothing stopping non-reviewers from posting about them.

I dunno. Ask Apple. However, that's traditionally what Apple has done. Then the benchmarks show up in non-embargoed reviews a week after that.

Apple typically sends review units to people like iJustine or Marques Brownlee or The Verge, but does not send review units to places like AnandTech or Linus Tech Tips.

iJustine will gush about the colours or whatever and other embargoed reviewers will talk about their butter-smooth timelines in Final Cut.

 

[eek2121]

I seriously doubt they'll add any DIMM slot s now that CXL is a thing.

IMO it's far more likely they just offer CXL add-on-cards for those that need more than 256 GB of ram. Like the one Samsung announced:


I mean, sure the latency will be somewhat worse, but still totally usable. Workstation users needing more than 256 GB at absolute lowest latency are quite few and far between anyway (and certainly not the usual mac pro customers).

And depending on the LPDDR5 modules - Couldn't they also not fit 512GB (with the most premium modules) for people that need a bit more than 256 and absolutely require the best latency and all the memory bandwidth ... for an astronomical price of course.

I will be shocked if they allow expandable memory at all. I suspect Apple may kill off expandability altogether. The Mac Pro will just have absurd specs (20-40 cores, tons of RAM, lots of storage). That is typically the way Apple rolls. We will see.

Hmmm... I didn't realize, but the SSDs on these laptops are up to 7.4 GB/s. Wow.

And interestingly the dot pitch is exactly 0.1 mm, or 100 pixels per cm (254 ppi). That should make calculations easy.

The high impedance headphone compatibility is cool too.

And it can charge by either MagSafe or USB-C, which is great for those running powered USB-C dongles or USB-C displays.

They've basically thrown everything in there but the kitchen sink (except for FaceID).

You can obtain SSDs this fast, but they aren't cheap. I have one that is just a tad slower. It isn't my laptop, however, because the laptop only supports PCIE 3.0.

 

[DRC_40]

Until this hardware is in someone’s capable hands I don’t believe any of Apples claims. I’m probably wrong but I don’t trust Apple. Consummate puffing the goods.