Discussion Apple Silicon SoC thread

[Eug]

M1
5 nm
Unified memory architecture - LP-DDR4
16 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 12 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache
(Apple claims the 4 high-effiency cores alone perform like a dual-core Intel MacBook Air)

8-core iGPU (but there is a 7-core variant, likely with one inactive core)
128 execution units
Up to 24576 concurrent threads
2.6 Teraflops
82 Gigatexels/s
41 gigapixels/s

16-core neural engine
Secure Enclave
USB 4

Products:
$999 ($899 edu) 13" MacBook Air (fanless) - 18 hour video playback battery life
$699 Mac mini (with fan)
$1299 ($1199 edu) 13" MacBook Pro (with fan) - 20 hour video playback battery life

Memory options 8 GB and 16 GB. No 32 GB option (unless you go Intel).

It should be noted that the M1 chip in these three Macs is the same (aside from GPU core number). Basically, Apple is taking the same approach which these chips as they do the iPhones and iPads. Just one SKU (excluding the X variants), which is the same across all iDevices (aside from maybe slight clock speed differences occasionally).

EDIT:



M1 Pro 8-core CPU (6+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 16-core GPU
M1 Max 10-core CPU (8+2), 24-core GPU
M1 Max 10-core CPU (8+2), 32-core GPU

M1 Pro and M1 Max discussion here:

Page 78 - Discussion - Apple Silicon SoC thread

Page 78 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M1 Ultra discussion here:

Page 109 - Discussion - Apple Silicon SoC thread

Page 109 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M2 discussion here:

Page 127 - Discussion - Apple Silicon SoC thread

Page 127 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Second Generation 5 nm
Unified memory architecture - LPDDR5, up to 24 GB and 100 GB/s
20 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 16 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache

10-core iGPU (but there is an 8-core variant)
3.6 Teraflops

16-core neural engine
Secure Enclave
USB 4

Hardware acceleration for 8K h.264, h.264, ProRes

M3 Family discussion here:

Page 215 - Discussion - Apple Silicon SoC thread

Page 215 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[JasonLD]

The M1 Ultra's GPU performance should put NVIDIA on notice. No, it doesn't beat a 3090 or even a 3080 in all workloads, however, the perf/watt is out of this world.

I don't expect Nvidia (nor AMD or Intel) is going to follow Apple's method of going very wide/lower freq to achieve top perf/watt since it won't be economically viable for them.
Next generation Nvidia GPUs using 5nm TSMC is going to close the perf/watt gap considerably though, since they are basically doing 1.5 generation jump on nodes.

 

[Doug S]

With only a couple graphics benchmarks, neither of which really touch on the Studio's target market (maybe Handbrake does, kinda, but transcoding isn't something have to sit and watch - you can set it loose on a bunch of files overnight and don't have to care how long it takes) that ArsTechnica "review" is pretty crap. Hard to form any conclusion about the M1 Ultra vs Nvidia based on that.

Hopefully we will get much better from Anandtech, though given the recent departure of their two best people who knows if we'll get it.

 

[vadimyuryev]

The M2 will Not reach 2000+ numbers, it's predicted to stay at 1800s

View attachment 58485

View attachment 58486

If the M2 is based on the A16 Bionic cores, then 2,000 point single-core scores are possible.
There is a chance that now that Apple's Macs are using Apple Silicon chips, they could flip the script and create the M2 chip first and then bring it down to the A16 design in terms of lower core counts and power usage. This could mean that the M2 could come before the A16 in September, therefore making it possible for the M2 to skip the A15 architecture.

Apple is not stitching adjacent dies for M1 Ultra. They're almost certainly using TSMC CoWoS-L or a similar variation of that technology proprietary to Apple. The relevant Apple patent would be US20210159180A1 as reported by ComputerWorld. TSMC's advanced packaging roadmap has also been covered previously by Anandtech. During the Apple presentation, Johny Srouji specifically said that they were connecting two M1 Max dies using a silicon interposer, and the accompanying graphics certainly appear to show two singulated M1 Max dies coming together and being joined by an interposer chiplet sitting below them. I don't think there's any reason to believe they are misrepresenting this.

I'd also pay attention to what Hector Martin is saying, because he's got a very good handle on M1 hardware. There probably is no chip forthcoming that uses four M1 Max dies, and no possibility of a "multi-socket" M1 Ultra. Seeing as the M1 Ultra isn't socketed, it would simply be a multi-processor configuration, which of course is a tried and true topology. However, the highest bandwidth interface the M1 Ultra has available for chip-to-chip communication would be PCI Express Gen4 x4—at just 16 GB/s aggregate bandwidth before you even account for encoding and protocol overhead. So that idea is a non-starter.

The die-to-die interface that the M1 Ultra uses does seem like it could provide sufficient bandwidth for four dies though. The LPDDR5-6400 memory interfaces on the M1 Max can theoretically provide peak bandwidth of 409.6 GB/s. Double that for a bidirectional link between two dies and you're at 819.2 GB/s. Triple that for a 4-die all-way link and you get 2.4576 TB/s. Apple claims that UltraFusion can provide 2.5 TB/s of interprocessor bandwidth. 🤔

This is certainly enough to make one ponder how Apple might package four M1 Max dies along with memory for a Mac Pro system, and I'm right there along with everyone else going down that rabbit-hole. But in the end, I don't think it's going to happen with this generation of chips.

I agree. I don't think the sandwich style is likely because according to Hector Martin, the IRQ controller built into the M1 Max only supports 2 dies. Apparently, each die is supposed to have an address to know where to send data for processing, or to know when to pause processing on a die. So I guess there are only 2 possible addresses in terms of the die.

This makes it more likely for the quad-chip to be based on the M2 Max, which could have an entirely new design with die-to-die interconnects on both sides.

 

[Ajay]

Waiting for a tear down on this new system..............

 

[vadimyuryev]

Waiting for a tear down on this new system..............

We'll do our best to get in -Vadim from Max Tech

 

[Eug]

Are there benchmarks for M1 Ultra yet?

Apple’s charts set the M1 Ultra up for an RTX 3090 fight it could never win

Better performance per watt doesn’t equal better overall performance.

www.theverge.com

Apple’s chart is, for lack of a better term, cropped. The company only shows the head to head for the areas where the M1 Ultra and the RTX 3090 are competitive against each other, and it’s true: in those circumstances, you’ll get more bang for your buck with the M1 Ultra than you would on an RTX 3090.

But what the chart doesn’t show is that while the M1 Ultra’s line more or less stops there, the RTX 3090 has a lot more power that it can draw on

Why should Nvidia care, even if the M2 based Mac Pro with its double the SoCs and more & better GPU cores is able to beat Nvidia's best? No matter how successful Mac Pro sales are by Apple's standards, it won't hurt Nvidia at all.

A smart company always pays attention to the lay of the land.

Would Ford care if Mercedes introduced a $150,000 car with better performance as well as better gas mileage?

Mac Studio is $1999 with M1 Max + 32 GB and $3999 with M1 Ultra + 64 GB RAM. That is the entire computer including sleek case, SSD, CPU, GPU, etc.

Car analogies always suck.

 

[Markfw]

I just have to say this... Until most business software uses a MAC and I can play my favorite games on a MAC, I could care less what performance apple has.

 

[eek2121]

With only a couple graphics benchmarks, neither of which really touch on the Studio's target market (maybe Handbrake does, kinda, but transcoding isn't something have to sit and watch - you can set it loose on a bunch of files overnight and don't have to care how long it takes) that ArsTechnica "review" is pretty crap. Hard to form any conclusion about the M1 Ultra vs Nvidia based on that.

Hopefully we will get much better from Anandtech, though given the recent departure of their two best people who knows if we'll get it.

I have seen enough on various review sites to know where it sits. The issue is that you are comparing the performance of a machine that uses well under 100W to a single GPU that uses 2-3 times that amount. The M1 ultra is somewhere around the 3070ti for graphics workloads, and around the 3060ti for compute. The 3060ti uses around 200W of sustained power, while the 3070ti uses around 300W of sustained power.

Don’t get me wrong, I would never buy a Mac for GPU performance. None of the stuff I use, be it professional tools or games, work on a Mac. However, it is encouraging to see someone paying attention to perf/watt. Perhaps NVIDIA and AMD will begin to as well.

 

[MadRat]

I just have to say this... Until most business software uses a MAC and I can play my favorite games on a MAC, I could care less what performance apple has.

With virtualization support in hardware today, what stops you from doing so?

 

[igor_kavinski]

I just have to say this... Until most business software uses a MAC and I can play my favorite games on a MAC, I could care less what performance apple has.

You have favorite games???

 

[Muadib]

You have favorite games???

I do, and that’s the main reason I’m hesitant to order the studio. All work and no play yada yada. Why does Apple continue to ignore the gaming market?

 

[TheStu]

I do, and that’s the main reason I’m hesitant to order the studio. All work and no play yada yada. Why does Apple continue to ignore the gaming market?

Because the gaming market ignores them (outside of mobile). It's a circle between the two; Apple could start to address it on their side by throwing a ton of money and resources at developing new APIs and funding game studios so that titles can be ported over, but you have to ask both Apple and the developers... how much of a market are they missing out on? How many people have a Mac, and NO other way of playing any games from these developers? If magically, tomorrow, all games from all devs were natively compatible on macOS, how many people would really choose a Mac vs a gaming PC? I would think about it, but only if Apple's GPUs can compete. What's Apple's impetus to spend all that time and money? What's the developers' drive to do the same?

 

[Ajay]

I do, and that’s the main reason I’m hesitant to order the studio. All work and no play yada yada. Why does Apple continue to ignore the gaming market?

They don’t ignore it - its just not as high a priority for them as it is for the Win/x86 market. It’s a profitable business line for Apple, but still niche compared to the Windows OS market. If the Apple Mac line commanded 30% of the market - the game developers themselves would be knocking on Apple's door for better game support, and they themselves would better optimize their games for the metal GFX API.

 

[you2]

I just have to say this... Until most business software uses a MAC and I can play my favorite games on a MAC, I could care less what performance apple has.

Can we run linux on them yet?

 

[igor_kavinski]

Apple can have my $2000 if they give me original M1 having 8 GB on-chip RAM in MBP with user upgradable DDR5 RAM up to 128 GB and replaceable NVMe SSD.

 

[jpiniero]

Apple can have my $2000 if they give me original M1 having 8 GB on-chip RAM in MBP with user upgradable DDR5 RAM up to 128 GB and replaceable NVMe SSD.

lol

If anything the Mac Pro will be soldered too. Maybe expansion using CXL.

 

[JasonLD]

lol

If anything the Mac Pro will be soldered too. Maybe expansion using CXL.

Apple isn't part of CXL consortium. Mac Pro could still have a additional DDR5 slots on top of having embedded LDDR5 for further memory expansion. Otherwise, I can't see how Apple could provide memory capacity beyond 256GB for Mac Pro.

 

[jpiniero]

Apple isn't part of CXL consortium.

That doesn't mean much. I would say most likely there isn't any expansion beyond Thunderbolt.

 

[JasonLD]

That doesn't mean much. I would say most likely there isn't any expansion beyond Thunderbolt.

Apple has no need for CXL anyways, since Apple is already doing their own thing looking at M1 Ultra.
It could still provide either PCIe 4.0 or 5.0 based expansion slots, providing if they keep the same chassis. Otherwise, Mac Pro wouldn't differenciate much from the Mac Studio.

 

[Ajay]

That doesn't mean much. I would say most likely there isn't any expansion beyond Thunderbolt.

Uh, that won't fly for the Mac Pro. At a minimum there will be a need to support PCIe for AIBs including graphics. Also, RAM support for at least a 1TB. The market for Mac Pros is small and very dependent on large amount of cores/IO/memory and high performance graphics. If Apple doesn't provide that, then much of that segment will be ceded to the Windows PC market.

I supposed that many of the current PCIe based AIBs could be redesigned to use Thunderbolt, but that won't work for graphics AIBs (not enough bandwidth). Also, the costs of a redesign may well kill of some professional AIBs, as the new price point might be too high to achieve necessary sales.

Otherwise, the Mac Pro will just be a Mac Studio Max with quad M1 Max SoCs. (or whatever SoC comes out next).

 

[jpiniero]

Uh, that won't fly for the Mac Pro. At a minimum there will be a need to support PCIe for AIBs including graphics.

I'd be surprised if Apple ever allows a discrete GPU. So if you are talking PCIe slots it's most likely accelerator cards and perhaps CXL.

Otherwise, the Mac Pro will just be a Mac Studio Max with quad M1 Max SoCs.

It wouldn't be M1 but something similar does seem like the most likely outcome. With a different design that allows for more chiplets of course.

 

[Doug S]

They don’t ignore it - its just not as high a priority for them as it is for the Win/x86 market. It’s a profitable business line for Apple, but still niche compared to the Windows OS market. If the Apple Mac line commanded 30% of the market - the game developers themselves would be knocking on Apple's door for better game support, and they themselves would better optimize their games for the metal GFX API.

Apple is huge in the gaming market, just not the gaming market people here care about. The sort of hardcore gamers that visit these forums and breathlessly await every review of the next generation of Intel or AMD CPU look down on the casual / mobile gaming market, but there's more money there than there is in the hardcore gaming market.

 

[Ajay]

Apple is huge in the gaming market, just not the gaming market people here care about. The sort of hardcore gamers that visit these forums and breathlessly await every review of the next generation of Intel or AMD CPU look down on the casual / mobile gaming market, but there's more money there than there is in the hardcore gaming market.

*thinks about wife's gaming habits* Good point! There was some old statistic that 70% of the apps in the IOS app store were games, and that they accounted for 90% of sales! IDK what it's like today.

 

[Roland00Address]

Apple can have my $2000 if they give me original M1 having 8 GB on-chip RAM in MBP with user upgradable DDR5 RAM up to 128 GB and replaceable NVMe SSD.

Is anyone making so dimm ddr5 chips with more than 32gb per stick ? (thus 64 gbs if you have 2 slots max)

 

[Heartbreaker]

Uh, that won't fly for the Mac Pro. At a minimum there will be a need to support PCIe for AIBs including graphics. Also, RAM support for at least a 1TB. The market for Mac Pros is small and very dependent on large amount of cores/IO/memory and high performance graphics. If Apple doesn't provide that, then much of that segment will be ceded to the Windows PC market.

I really can't see Apple supporting third party GPUs going forward, and I don't know how many PCIe cards remain that are important in the Mac Pro market.

I also don't know how they handle RAM. Anything with Slots will be slower, plus they already have 16 channels. Do they go to 20 channels with slots to make up for the lower speed of slotted RAM? (Yikes).

Otherwise, the Mac Pro will just be a Mac Studio Max with quad M1 Max SoCs.

I really think it might be something closer to that. But perhaps Multiple Ultras?

Maybe a proprietary card cage with super high speed backplane, where you have One to Four M2-Ultras configured each on their own card so you can add another Complete Ultra later.

Each Ultra can have up to 256GB of soldered in LPDDR/HBM.

4 Ultra cards = up to 1 TB of RAM. 4X CPU and 4X GPU of Ultra.

Mind you anything to access any other card will have NUMA penalties so trade-offs rise as you move beyond one Ultra.

Really hard to guess what Apple has up it's sleeve for the new Mac Pro. Which makes it exciting.