Peak Peek - Apple officially announces new SE, new iPad Air, M1 Ultra, Mac Studio and Studio Display

[Imbadatnames]

14 minutes ago, leadeater said:

Not always that simple because current product stack a 12700 vs 12700k is not just locked vs unlocked as they have different clocks and TDP (PL1/PL2), they are entirely different products. So the additional problem comes when trying to compare a future 13700 vs 12700K, which is faster?

Was my point. You can’t complain at apple for having a fairly straightforward step up in performance naming scheme and then praise Intel/AMD for having numbers that do the same if not a worse job due to the inflated product stack. Haven’t even touched Y and U chips yet 

[hishnash]

11 minutes ago, leadeater said:

I would posit that just means the defect rate is so low in the CPU portion of the die it's a non issue, they do bin GPU cores and that is most likely because there actually is a significant enough defect rate there to warrant it.

 

Or when there is a defect with the cpu or cache if they binned it they would need to discard 1/2 of the cpu cores.  Or a defect in the TB controller, would they then sell a Mac with 1 less TB port? 

They know they cheapest option will sell in the highest volume as well so if only 2% of dies have a defect they can't sell that as the entry level config (unless they want to start damaging perfectly good dies).   

 

14 minutes ago, leadeater said:

Having some defective dies in this specific way isn't quite the same thing as having enough supply of them to underpin an entire product offering and not potentially end up having to supply with with otherwise useful and more valuable dies.

I think it depends on the volume they expect to sell, i expect the numbers of Dedicated Gpus they sell for the macpro will be very low, given many macpro users will be happy with the on package gpu and will be buying the macPro for the PCIe slots to put other cards into it.  Lets say its 2% of M1 max dies, even if 50% of macPro users opt to get a dedicated GPU add in card compared to all the M1 Max dies being mad for 16" MBPs that will still be less than 1% of the total M1 Max production.  

there is a reason apple used an off the shelf (much more costly) FPGA for the afterburner card rather than have dedicated silicon made, they know the volumes they would sell it do not justify the many many millions $ needed in tooling (masks etc).  The total volume of macPros that will be sold will be very low that is why I can't see them making a dedicated GPU die as there is no way on earth they will make up the tooling costs given the need to re-fresh that every few years. 

 

[leadeater]

1 hour ago, hishnash said:

there is a reason apple used an off the shelf (much more costly) FPGA for the afterburner card rather than have dedicated silicon made, they know the volumes they would sell it do not justify the many many millions $ needed in tooling (masks etc).

I saw that more as the end justifies the means, they needed something sooner than later and also needed to gauge market demand along with only needing a specific task accelerated. This is what FPGAs are actually for. A more complex and larger GPU is a bit different and to top it off what the MPX card did the SoCs do themselves.

 

So what exactly would actually utilize a dedicated GPU? If you need more encode/decode streams than a M1 Ultra can support. Compute GPU acceleration? Well an add-in card pretty much entirely negates all the benefits of the SoC and why the GPU is so good, even CUDA unified memory is useless to a large extent without NVLink directly between the GPUs and in some cases (PowerPC) between the CPU(s).

 

If/When dedicated GPUs come Apple will have much better plans around it than simply repurposing. I have a strong sense other interconnect technology and protocols are going to be part of it to get true usefulness out of them and not introduce interoperability and utilization problems between the SoC GPU and any add-in GPUs.

[igormp]

4 hours ago, sazrocks said:

I'm mainly concerned with what apparently is a NUMA memory architecture with M1 ultra. Each chiplet seems to keep its own memory controller meaning there is going to be a latency difference between accessing memory attached to the same chiplet vs the other chiplet. This could lead to really inconsistent performance scaling across workloads, as it did for Threadriper 1st gen and Epyc 1st gen.

They're likely just going to just interleave the memory at the hardware level, so the OS won't even be aware of that, the latency should be small enough to do so.

 

4 hours ago, LAwLz said:

Maybe we'll see a 3 chiplet design? 

Unlikely, the M1 only has a single D2D connection. The only possibility that I can think of is using 2x M1 Ultras connected through some slower, external bus, and me thinks that's not going to happen.

 

3 hours ago, Arika S said:

Will the M2 be faster than ALL M1 variants? or will it slot in under the Ultra and above the Pro? Or will it be the same as the Ultra just with extra feature? 

No, the M2 will likely be faster than the M1 but slower than all of the other variants.

 

2 hours ago, Arika S said:

What i am pissed at Intel for is their letter identifiers, because on desktop you have

• Base
• K
• F
• KF

Mobile is even worse.

 

If i can look at 2 products from the same company on a store shelf next to each other and not immediately be able to tell which one is the "better" one, then they fucked up.

As others said, in that case those aren't meant to identify tiers, but features. The thing that one would need to know which is the better models is in the numbers (i3/i5/i7/i9, 12600, 12700, 12900, etc).

[hishnash]

1 hour ago, leadeater said:

Well an add-in card pretty much entirely negates all the benefits of the SoC and why the GPU is so good

Depends on the task, if your application supports multi GPU then you can make use of many add in cards while still making use of the SoC GPU. Apps targeting the macPro already need to do this since all of the high end GPU options on the macPro are multi gpu options (the duo cards apple sell expose themlsves as multi gpu). 

 

[hishnash]

1 hour ago, igormp said:

They're likely just going to just interleave the memory at the hardware level, so the OS won't even be aware of that, the latency should be small enough to do so.

 

So the OS schedular is aware of this, (at least the linux on M1 team are aware of this) there are 2 cpu clusters each with its own L2 cache so the schedular already handles that correctly.  I expect the os schedular will be aware of what die it is running on since it will know what core complex it is running the task on. MacOS has supported much more complex NUMA targets (multi die Xeons) so this should not be an issue at all, even on older intel systems the schedular cared about what core to put a task basked on the tasks priority and the boost etc state of the core.  
 

[Kisai]

8 hours ago, harryk said:

Steve Jobs famously had an obsession with making impossibly powerful miniature computers. Famously they all are regarded as failures. Let's see how the latest iteration fares. Personally I have high hopes for the Mac Studio and will likely purchase one in the coming future.

 

 

None of those were failures. Just the G4 cube and the Mac Pro cylinder aren't what the market asked for. The M1 Ultra is about 90% of what people asked for. 

 

1. It's a small box that has front ports

2. It has 4 TB ports in the back and 2 USB-A ports. So it technically has only 2 less ports than the Trashcan Mac Pro. This doesn't get us away from donglehell, but it's close enough. I would have rather a 3.5mm headphone jack be on the front support TRRS headset (speaker+mic), with the connector on the rear be an optical S/PDIF (TOSLINK). It's close enough, but falls a little short for professional use out of the box. And the reason for wanting this is because you want a way to test the surround sound mix on actual high quality speakers.  However I do realize that SPDIF falls short of being able to do Dolby Atmos on top of Dolby Digital Plus (8-channel 18Mbits/sec), and DTS-MA/DTS:X , but I'm not seeing the Mac Studio being used as a theatrical player, just editing/mastering. So perhaps this might mean connecting the HDMI instead to a surround processing box, adding latency. At least an optical out permits something rather than nothing at all.

3. It lacks user-upgradable internal storage. Seeing as this would likely be the first failure point, this pretty much requires you to buy more storage than you need, since you'll never be able to upgrade it.

 

The market wants two kinds of mac's for professional use:

A. Something that is not an iMac, but otherwise is far more powerful than a MacMini (which was never able to meet a windows desktop or iMac at parity with the intel chips)

B. Something that is a "big boy tower" for workstation use.

[hishnash]

4 minutes ago, Kisai said:

 

3. It lacks user-upgradable internal storage. Seeing as this would likely be the first failure point, this pretty much requires you to buy more storage than you need, since you'll never be able to upgrade it.

 

I expect almost everyone who will use this will have all real data stored either on the network or a TB attached drive. There are now TB drives that can aggregate multiple TB links to provide higher performance!

The internal drive will be used just for applications etc, 

The fact that apple are still selling the macPro (and said they will be updating this at a later point) means the issues of no having use-upgrabilty are offset by having that other product. 

[WolframaticAlpha]

4 hours ago, Arika S said:

performance in in ascending order. and something that the layperson would easily be able to determine.

Lol, I was talking to someone and they thought that M1 Max is better than M1 ultra because Max is maximum lol.

[WhitetailAni]

7 hours ago, HenrySalayne said:

The Mac Studio is just as ugly as the Surface Laptop Studio.

Come on, Apple. This is not industry leading design, this is just a hideous brick.

They just took the Mac mini and stretched it.

Kinda lazy, but I can see why. What would you have them do? Extend out? Extend back?

[saltycaramel]

“We’re adding ONE LAST CHIP to the M1 family, and it’s gonna blow your mind.”

 

That would imply the M1 Ultra is the last chip of the A14/M1 generation.

 

I think the Mac Pro will use M2 Max 10-core, M2 Ultra 20-core and M2 Ultra 40-core. (I don’t think it’s possible for the Mac Pro to have CPU-parity with the humble small form factor Mac Studio, it’s gotta have a 40-core option)

 

And the M2 Max ought to have the UltraFusion chiplet interconnect on TWO sides of the die (as opposed to the M1 Max having it only on one side), otherwise you wouldn’t be able to pair 4 of them together.

 

So to recap we yet have to see one last M1 based Mac: the new M1 Pro/Max Mac Mini with plastic white top and external power brick.

 

After that, it’s all M2/A15 generation

- M2 Macbook Air new design

- M2 Macbook Pro 13”

- M2 Mac Mini low-end

- M2 Pro/Max Macbook Pro 14”/16”

- M2 Max/Ultra_20c iMac Pro 7K miniLED

- M2 Max/Ultra_20c/Ultra_40c Mac Pro

 

 

[rikitikitavi]

1 hour ago, WolframaticAlpha said:

Lol, I was talking to someone and they thought that M1 Max is better than M1 ultra because Max is maximum lol.

It does sound logical. Ultra is a new word in Apple Marketing Dictionary and ppl don't know yet that Maximum is less than Ultra lol

 

I guess someone in the department thought that M1 8-core, 10-core, 20-core sounds non-Apple and too confusing for the audience that actually cares about such... 

[rikitikitavi]

11 minutes ago, FakeKGB said:

Kinda lazy, but I can see why. What would you have them do? Extend out? Extend back?

Add colour.

[WhitetailAni]

4 minutes ago, rikitikitavi said:

Add colour.

But then it won't match the Studio Display.

Also, these are professionals. They don't need people to focus on their computers, they need people to focus on what they make.

[saltycaramel]

6 hours ago, Arika S said:

Will the M2 be faster than ALL M1 variants? or will it slot in under the Ultra and above the Pro? Or will it be the same as the Ultra just with extra feature? 

The M2 will have a faster single-core performance than all M1 variants.

But it doesn’t even matter much. People will shop for Macs and buy what’s currently available in the particular Mac form factor they’re interested in, not shop for CPUs. 

[HenrySalayne]

15 minutes ago, FakeKGB said:

They just took the Mac mini and stretched it.

Kinda lazy, but I can see why. What would you have them do? Extend out? Extend back?

The Mac Studio has the wrong proportions, changing these does not retain the original design.

The Mac Mini works because it's a disc. The Mac Studio is a brick. This is a complete break from their design language in the past decade (and even longer).

[BlueChinchillaEatingDorito]

44 minutes ago, FakeKGB said:

They just took the Mac mini and stretched it.

Kinda lazy, but I can see why. What would you have them do? Extend out? Extend back?

Though at the same time, it's hard to give them flack when on the PC side... OEMs have been using the same chassis for well over a decade. All they have really done to "innovate" for the past decade is change the plastic faceplates every so often to have the illusion they're doing something. Same case, same cooling solution. I'm looking at you Dell. 

[leadeater]

2 hours ago, hishnash said:

Depends on the task, if your application supports multi GPU then you can make use of many add in cards while still making use of the SoC GPU. Apps targeting the macPro already need to do this since all of the high end GPU options on the macPro are multi gpu options (the duo cards apple sell expose themlsves as multi gpu). 

Multi GPU is one thing, data sync and timing is another if required. Not everything can just be split and mass assigned, not that things can't. Just a lot of what if and depends which is why I think Apple will seek best to make that not a thing

[saltycaramel]

13 minutes ago, HenrySalayne said:

The Mac Studio has the wrong proportions, changing these does not retain the original design.

The Mac Mini works because it's a disc. The Mac Studio is a brick. This is a complete break from their design language in the past decade (and even longer).

The MacStudio is built around the fact that it needs to sit under the display and that it has front facing usb-c/thunderbolt ports and SDXC reader. Given that, that’s the only possible shape. 

[RedRound2]

8 hours ago, LAwLz said:

Edit: should have checked the rest of the announcement. They confirmed that they will have another chip for the Mac Pro. So we will get higher end stuff. 

They specifically said that M1 Ultra is the last M1 chip. The Mac Pro might have M2 and that would actually mean that they would have the entire M2 lineup ready from the get-go by WWDC (I dont think anyone is quite sure yet how the upgrade cycle of the Apple Silicon chips is gonna play out)

[WhitetailAni]

4 minutes ago, RedRound2 said:

They specifically said that M1 Ultra is the last M1 chip. The Mac Pro might have M2 and that would actually mean that they would have the entire M2 lineup ready from the get-go by WWDC (I dont think anyone is quite sure yet how the upgrade cycle of the Apple Silicon chips is gonna play out)

M1 Ultra likely didn’t take long. If the M1 Max has interprocessor built in no matter what, then it shouldn’t have taken long to Gorilla Glue two together, do some software work, and slap an Ultra on it.

[LAwLz]

6 hours ago, leadeater said:

I still really doubt defective dies will ever be used as GPUs. Not only would that be needlessly expensive having dies isolated to just CPU core defects and otherwise being useful is very low likelihood and those are already going to the 8 core models. Just seems extremely unlikely to me compared to actual GPU dies which could very well just be based on the M1 with the CPU cores and anything else not needed removed.

Not sure if you seen the announcement yet, but they announced that they will update the iPad Air with the full-fat version of the M1. 

Since they can afford to put the non-cutdown version of the M1 in their midrange products, my guess is that their yields are really good. 

[saltycaramel]

15 minutes ago, RedRound2 said:

They specifically said that M1 Ultra is the last M1 chip. The Mac Pro might have M2 and that would actually mean that they would have the entire M2 lineup ready from the get-go by WWDC (I dont think anyone is quite sure yet how the upgrade cycle of the Apple Silicon chips is gonna play out)

 

The MacPro isn’t coming anytime soon imho. 

Even if the M2s start to show up this Spring (and we have reports from devs that are already starting to see them in the logs, because they’re being tested on Apple campus, and mass production of M2 laptops started in January), the bigger M2 Pro, M2 Max and M2 Ultra will be 9-12 months away at least. 
Just like Xeon architectures catch up later…you begin from the smaller chips and gradually move to the bigger ones..

 

So I believe we’ll see the MacPro at WWDC 2023 at the earliest…and we may even get another Intel MacPro in the meantime. (btw yesterday the MacPro got a new GPU cto option, the AMD W6600X)

[hishnash]

1 hour ago, leadeater said:

Multi GPU is one thing, data sync and timing is another if required. Not everything can just be split and mass assigned, not that things can't. Just a lot of what if and depends which is why I think Apple will seek best to make that not a thing

For sure, but worth noting the GPU in the M1 Ultra is already faster than any single GPU they sell for the macPro (its not faster than the duo but those are multi gpu).  So as a drop in for WWDC this year they could use the M1 Ultra with M1 ultra based add in GPUs (for those that need more than the gpu on the SoC) and then next year when they have a M2 Ultra (or exteram) release that. Rather than waiting another year and missing there deadline. 

[Kisai]

13 minutes ago, saltycaramel said:

 

The MacPro isn’t coming anytime soon imho. 

Even if the M2s start to show up this Spring (and we have reports from devs that are already starting to see them in the logs, because they’re being tested on Apple campus, and mass production of M2 laptops started in January), the bigger M2 Pro, M2 Max and M2 Ultra will be 9-12 months away at least. 
Just like Xeon architectures catch up later…you begin from the smaller chips and gradually move to the bigger ones..

 

So I believe we’ll see the MacPro at WWDC 2023 at the earliest…and we may even get another Intel MacPro in the meantime. (btw yesterday the MacPro got a new GPU cto option, the AMD W6600X)

I'm not expecting a "M2" brand honestly. I actually expect them to still call it a M1 unless there is actual functionality added (eg native AV1 encoding and decoding, which the M1 lacks both, and is about as fast as an Intel CPU without hardware decoding in decoding AV1.) 

 

Honestly, the lack of AV1 encoding is the reason why I'm not buying new hardware right now. This mac is cool and all, but for my true use case, I need to be able to stream and record AV1, and for that, I'd be willing to wait until something does. Present AV1 software encoders are simply too poor to use for streaming.