That roadmap implies no m3 Mac mini. I wonder if there will also be no m3 Mac Studio and if maybe the Mac Studio will get an m4 faster than we’d otherwise expect. And if all that turns out to be true, I wonder if the explanation might be that the first 3nm process from TSMC had yield issues, so maybe Apple wants to move to N3E ASAP
I wonder if we might see an intro of an M4 Max and Ultra for the Mac Studio and Mac Pro at WWDC. Introduce them on June 10th, available on June 21 or 28 (which is a bit longer than the normal announcement to availability span for Apple)... that's 10 or 11 weeks away. Considering the relative small numbers of units needed compared to MacBook Pros, this could be a possibility.
Also, putting an M3 Max and Ultra chip into a Studio in June, really makes no sense as the processors (at least for the Max) as it would already be eight months old.
LOL, this is never going to happen. The M4 is not that far along and the basic vanilla M4 would be the first one out of the fab due to less complexity compared to the Pro/Max/Ultra. Fabs rarely if ever start with the more advanced chips and then work backwards. Moreover, if the M4’s are ready now one has to question how many “AI” enhancements it actually has.
I would imagine Apple will have the vanilla M4 pretty much ready to go to production but it will not have much in terms of significant AI enhanced features that will be added to the Pro/Max/Ultra models in which I can see each iteration being more AI capable. This also allows Apple to continue working on further AI enhancement capabilities that will be part of the Ultra or Max.
I agree it is unlikely to happen in June, but never say never. The max and ultra are very low volume relative to other chips in apples lineup so it’s not crazy to think they could come earlier in the cycle than was the case for m1 and m2.
Heck, m3 max was out at the same time as m3, so there is that precedent.
If for some reason the M4 max was released so soon I doubt it would have compelling AI capabilities since it seems Apple has been a little late to the AI party.
Apple first included an NPU in the A11, back in 2017. They have increased the number of cores from 2 to 16 during that time. They’ve primarily used it for image related ML tasks.
Apple is only late to the GPT NLP game. But no later than they were to the mp3 player or smartphone games. It might be too early to count them out.
That roadmap implies no m3 Mac mini. I wonder if there will also be no m3 Mac Studio and if maybe the Mac Studio will get an m4 faster than we’d otherwise expect. And if all that turns out to be true, I wonder if the explanation might be that the first 3nm process from TSMC had yield issues, so maybe Apple wants to move to N3E ASAP
I wonder if we might see an intro of an M4 Max and Ultra for the Mac Studio and Mac Pro at WWDC. Introduce them on June 10th, available on June 21 or 28 (which is a bit longer than the normal announcement to availability span for Apple)... that's 10 or 11 weeks away. Considering the relative small numbers of units needed compared to MacBook Pros, this could be a possibility.
Also, putting an M3 Max and Ultra chip into a Studio in June, really makes no sense as the processors (at least for the Max) as it would already be eight months old.
LOL, this is never going to happen. The M4 is not that far along and the basic vanilla M4 would be the first one out of the fab due to less complexity compared to the Pro/Max/Ultra. Fabs rarely if ever start with the more advanced chips and then work backwards. Moreover, if the M4’s are ready now one has to question how many “AI” enhancements it actually has.
I would imagine Apple will have the vanilla M4 pretty much ready to go to production but it will not have much in terms of significant AI enhanced features that will be added to the Pro/Max/Ultra models in which I can see each iteration being more AI capable. This also allows Apple to continue working on further AI enhancement capabilities that will be part of the Ultra or Max.
I agree it is unlikely to happen in June, but never say never. The max and ultra are very low volume relative to other chips in apples lineup so it’s not crazy to think they could come earlier in the cycle than was the case for m1 and m2.
Heck, m3 max was out at the same time as m3, so there is that precedent.
If for some reason the M4 max was released so soon I doubt it would have compelling AI capabilities since it seems Apple has been a little late to the AI party.
Apple first included an NPU in the A11, back in 2017. They have increased the number of cores from 2 to 16 during that time. They’ve primarily used it for image related ML tasks.
Apple is only late to the GPT NLP game. But no later than they were to the mp3 player or smartphone games. It might be too early to count them out.
It seems no matter how many times it get spelled out, some people will still read that and think “but NPU and AI don’t have any of the same letters.”
That roadmap implies no m3 Mac mini. I wonder if there will also be no m3 Mac Studio and if maybe the Mac Studio will get an m4 faster than we’d otherwise expect. And if all that turns out to be true, I wonder if the explanation might be that the first 3nm process from TSMC had yield issues, so maybe Apple wants to move to N3E ASAP
I wonder if we might see an intro of an M4 Max and Ultra for the Mac Studio and Mac Pro at WWDC. Introduce them on June 10th, available on June 21 or 28 (which is a bit longer than the normal announcement to availability span for Apple)... that's 10 or 11 weeks away. Considering the relative small numbers of units needed compared to MacBook Pros, this could be a possibility.
Also, putting an M3 Max and Ultra chip into a Studio in June, really makes no sense as the processors (at least for the Max) as it would already be eight months old.
LOL, this is never going to happen. The M4 is not that far along and the basic vanilla M4 would be the first one out of the fab due to less complexity compared to the Pro/Max/Ultra. Fabs rarely if ever start with the more advanced chips and then work backwards. Moreover, if the M4’s are ready now one has to question how many “AI” enhancements it actually has.
I would imagine Apple will have the vanilla M4 pretty much ready to go to production but it will not have much in terms of significant AI enhanced features that will be added to the Pro/Max/Ultra models in which I can see each iteration being more AI capable. This also allows Apple to continue working on further AI enhancement capabilities that will be part of the Ultra or Max.
I agree it is unlikely to happen in June, but never say never. The max and ultra are very low volume relative to other chips in apples lineup so it’s not crazy to think they could come earlier in the cycle than was the case for m1 and m2.
Heck, m3 max was out at the same time as m3, so there is that precedent.
If for some reason the M4 max was released so soon I doubt it would have compelling AI capabilities since it seems Apple has been a little late to the AI party.
Apple first included an NPU in the A11, back in 2017. They have increased the number of cores from 2 to 16 during that time. They’ve primarily used it for image related ML tasks.
Apple is only late to the GPT NLP game. But no later than they were to the mp3 player or smartphone games. It might be too early to count them out.
Don’t think they need to count on being in it. They will have chatbot and LLM oriented features that will be very useful to some niche of people, but I’m having a hard time imagining it as a feature that will sell computers or phones.
LLMs seem adjacent to automation and scripting support. That has been a thing, a desire among many users, for at least 30 years. Automation and scripting is a niche feature all this time, it speaks to what the mass market does with computers, and how many things are not scriptable or automated for those workflows.
So wait and see. This is for Apple’s bread and butter consumer computing products. Server markets? LLMs look like a big product and a new instrument of torture. The non-deterministic outputs based on vague inputs will be interesting to see.
Meanwhile, steady as she goes with Apple. Add LLM features that make a difference to users.
That roadmap implies no m3 Mac mini. I wonder if there will also be no m3 Mac Studio and if maybe the Mac Studio will get an m4 faster than we’d otherwise expect. And if all that turns out to be true, I wonder if the explanation might be that the first 3nm process from TSMC had yield issues, so maybe Apple wants to move to N3E ASAP
I wonder if we might see an intro of an M4 Max and Ultra for the Mac Studio and Mac Pro at WWDC. Introduce them on June 10th, available on June 21 or 28 (which is a bit longer than the normal announcement to availability span for Apple)... that's 10 or 11 weeks away. Considering the relative small numbers of units needed compared to MacBook Pros, this could be a possibility.
Also, putting an M3 Max and Ultra chip into a Studio in June, really makes no sense as the processors (at least for the Max) as it would already be eight months old.
LOL, this is never going to happen. The M4 is not that far along and the basic vanilla M4 would be the first one out of the fab due to less complexity compared to the Pro/Max/Ultra. Fabs rarely if ever start with the more advanced chips and then work backwards. Moreover, if the M4’s are ready now one has to question how many “AI” enhancements it actually has.
I would imagine Apple will have the vanilla M4 pretty much ready to go to production but it will not have much in terms of significant AI enhanced features that will be added to the Pro/Max/Ultra models in which I can see each iteration being more AI capable. This also allows Apple to continue working on further AI enhancement capabilities that will be part of the Ultra or Max.
I agree it is unlikely to happen in June, but never say never. The max and ultra are very low volume relative to other chips in apples lineup so it’s not crazy to think they could come earlier in the cycle than was the case for m1 and m2.
Heck, m3 max was out at the same time as m3, so there is that precedent.
If for some reason the M4 max was released so soon I doubt it would have compelling AI capabilities since it seems Apple has been a little late to the AI party.
Apple first included an NPU in the A11, back in 2017. They have increased the number of cores from 2 to 16 during that time. They’ve primarily used it for image related ML tasks.
Apple is only late to the GPT NLP game. But no later than they were to the mp3 player or smartphone games. It might be too early to count them out.
It seems no matter how many times it get spelled out, some people will still read that and think “but NPU and AI don’t have any of the same letters.”
While true, the overall notion of Apple getting caught off guard here remains correct.
This has little to do with NPUs in 2017, which as Blastdoor correctly pointed out were primarily image implementations and not industry leading either.
Just why ML wasn't all over the phone from the get go is perhaps indicative of not having enough resources available or just wanting to drip feed everything over OS roll outs over the years.
Apple wasn't first with the NPUs and, seven years on, finds itself behind on current AI technologies.
Yes, it has done a lot with ML since 2017, but then again so has everybody else, which made the recent insistance on ML and deliberately avoiding the term AI stand out even more as a topic of commentary. The real point here is that so many of the front runners have shipped so much more over the last couple of years.
There is no doubt that Apple will put AI front and foremost for WWDC because it is under pressure to do so. That is industry and investment pressure and it is undeniable.
Now the only real question is what they have in the pipe and how it performs. And of course when users will get to use it.
Generative options seem to be a given. Probably multi-modal options which is where the current buzz is now. Integration with current Apple apps and system services is also a given. And of course more ML smarts.
Rumours suggest Siri will benefit greatly but it is incredible how little has been done with Siri over the years.
Pro app tie-ins? Will they try to monetise those options with a tiered system or make it free? Or free for a time (like with the satellite options).
There are still a lot of unknowns and perhaps the most pressing of those (as I just mentioned) is when exactly they will appear for users.
All of this M4 speculation with the associated AI hype is going to kill sales of current Mac models. Tying AI into this chip along with all of the high expectations regarding how AI is going to change the world is going to give people who don’t absolutely need a new Mac now motivation to wait. This is an example of a time when it would be beneficial for Apple to talk a little bit about future products and tell people that currently shipping M3 models will be able to take advantage of future AI software from Apple. I could see Apple building significantly more neural processors into the M4. And perhaps enhancing memory and on chip communications. But the core of Mac OS will be fine with current M3 systems. I haven’t seen any recent rumors about some sort of iOS/MacOS integration. That might be something an M4 would be needed for.
I'm I the only person who doesn't want AI on his devices? So for the interaction I have had with AI has been Artificial Stupid not intelligence. If Apple forces AI with the next operating systems I will not install it or purchase any device with it pre-installed. I already have all the intelligence I need. I use my brain.
You probably have some form of “AI” already in your computer. The problem with the term AI is that it ends up meaning whatever the marketing team wants it to mean, independent of the tech team’s definition of AI.
AI can be neural networks, knowledge bases, deep statistical analysis and even a rebranding of a complex algorithm (or maybe not complex at all in certain cases).
The other problem with the current approach of AI is that it ends up being technology oriented development, rather than need oriented development.
As for using your brain, I agree using what you have, but sometimes delegating a task can free you up to do other things. Of course there are times when an AI can try to be too smart, but then again that can be a simple algorithm - too many times in MS Office that I’ve been corrected in ways I didn’t want.
Humph. Well, I’m still hanging on to the 27” mid 2010 iMac by my fingernails, using OCLP to run Ventura. That is coming to an end with Apple’s typical throttling of the previous OS with successive ‘security updates’ - and Sonoma isn’t going to run at all on the machine by the looks. So, I’ll be keeping a beady eye on WWDC to see what might be in the pipeline.
This M4 Gurman rumour could well knock my decision to get the M2 mini on the head (I’ve long ago given up on a direct 27” iMac replacement) - it has certainly made me put it on hold until mid-June. And I really have to accept that Apple’s marketing strategy has changed, and that I’m going to have to live with it - or not. Unfortunately for me, ‘or not’ has been occurring to me frequently since March this year .
As to AI, a lot of the motivation driving it appears to be the desire by managements to, to put it bluntly, sack people.
Half a terabyte is a lot of unified memory. It might be worth checking this report against memory roadmaps from SK Hynix, Samsung, and Micron.
Not compared to the memory you can connect to an Intel or AMD chip, it isn't. You already knew that, of course, because the REAL Mac Pro could use 1.5TB RAM, plus whatever video RAM was on your GPU boards.
"Unified" memory just means it's sharing RAM between the CPU and GPU. You know, the thing that used to be called "integrated graphics" that we used to acknowledge wasn't as good as a real GPU with its own RAM, but now that Apple does it, we pretend to fool ourselves?
Half a terabyte is a lot of unified memory. It might be worth checking this report against memory roadmaps from SK Hynix, Samsung, and Micron.
Not compared to the memory you can connect to an Intel or AMD chip, it isn't. You already knew that, of course, because the REAL Mac Pro could use 1.5TB RAM, plus whatever video RAM was on your GPU boards.
"Unified" memory just means it's sharing RAM between the CPU and GPU. You know, the thing that used to be called "integrated graphics" that we used to acknowledge wasn't as good as a real GPU with its own RAM, but now that Apple does it, we pretend to fool ourselves?
No, this is incorrect.
On-chip GPUs of the past partitioned off a section of system memory to serve as memory for the GPU. If a system has 8 GB of RAM, the on-chip GPU would partition off, say 1 GB, to serve as GPU memory, resulting in a decrease of system memory. To transfer GPU assets, programs would have to copy from system memory to the GPU memory partition. Some on-chip GPUs may have been limited on much
With Apple’s unified memory, there isn’t a partition and therefore there isn’t a transfer of assets to do. The GPU just accesses system memory wherever it needs to. Any processing unit in the SoC has uniform memory access to system memory.
Modern x86 SoCs may have addressed this and is using unified memory architectures, I think? I don’t really know and need to go find out.
For discrete GPUs on a PCIe card, where transfers have to be made, there is limitation of how many MB could be transferred per copy/transfer command and introduces inefficiencies. Various companies are trying to institute protocols where a dGPU can access system memory directly and remove transfer size limits, like resizable bars or system addressable memory protocols, which enable a dGPU to address system right across the PCIe bus.
Regarding 192 GB to a possible 512 GB memory on an Apple SoC, the GPU will be able to access something like 80% of it. So an Apple GPU could have 150 to 400 GB of memory it can directly address.
Most consumer dGPUs will be limited to something up to 32 GB. So it’s a natural advantage for certain workflows. Server GPUs have low hundreds of GB of memory, but not many people will have that on their desks.
There will be not M4 introductions until the A18 and A18 Pro chips are introduced. The iPhone’s capabilities are always going to take center stage. As Willie Sutton reportedly said, “That’s where the money is.”
September 2024, A18 and A18 Pro (iPhone 16, iPhone 16 Pro) October/November 2024, M4, M4 Pro, M4 Max (MacBook Pro line, MacMini, MacStudio) March 2025, M4 MacBook Airs, 24 inch iMac
That roadmap implies no m3 Mac mini. I wonder if there will also be no m3 Mac Studio and if maybe the Mac Studio will get an m4 faster than we’d otherwise expect. And if all that turns out to be true, I wonder if the explanation might be that the first 3nm process from TSMC had yield issues, so maybe Apple wants to move to N3E ASAP
I wonder if we might see an intro of an M4 Max and Ultra for the Mac Studio and Mac Pro at WWDC. Introduce them on June 10th, available on June 21 or 28 (which is a bit longer than the normal announcement to availability span for Apple)... that's 10 or 11 weeks away. Considering the relative small numbers of units needed compared to MacBook Pros, this could be a possibility.
Also, putting an M3 Max and Ultra chip into a Studio in June, really makes no sense as the processors (at least for the Max) as it would already be eight months old.
I have a M2 Studio Ultra bought it Nov. 2023, Apple please release the M3 Studio Ultra with 256 gigs of UMA memory if you don't release to the public you can't make a sell, Apple at this time is in a position to leverage the M series computers laptop/desktop unique qualities and have a presence in this area of computing a presence they don't have in for example AAA gaming. Apple can't afford to let this pass it is more critical than playing AAA games and it will also require Apple write/port/create support software.
Apple needs to offer more powerful Apple Silicon computers the time is now not later, and as usual Apple has been acquiring small companies to support existing in house projects within Apple.
Comments
This has little to do with NPUs in 2017, which as Blastdoor correctly pointed out were primarily image implementations and not industry leading either.
Just why ML wasn't all over the phone from the get go is perhaps indicative of not having enough resources available or just wanting to drip feed everything over OS roll outs over the years.
Apple wasn't first with the NPUs and, seven years on, finds itself behind on current AI technologies.
Yes, it has done a lot with ML since 2017, but then again so has everybody else, which made the recent insistance on ML and deliberately avoiding the term AI stand out even more as a topic of commentary. The real point here is that so many of the front runners have shipped so much more over the last couple of years.
There is no doubt that Apple will put AI front and foremost for WWDC because it is under pressure to do so. That is industry and investment pressure and it is undeniable.
Now the only real question is what they have in the pipe and how it performs. And of course when users will get to use it.
Generative options seem to be a given. Probably multi-modal options which is where the current buzz is now. Integration with current Apple apps and system services is also a given. And of course more ML smarts.
Rumours suggest Siri will benefit greatly but it is incredible how little has been done with Siri over the years.
Pro app tie-ins? Will they try to monetise those options with a tiered system or make it free? Or free for a time (like with the satellite options).
There are still a lot of unknowns and perhaps the most pressing of those (as I just mentioned) is when exactly they will appear for users.
This M4 Gurman rumour could well knock my decision to get the M2 mini on the head (I’ve long ago given up on a direct 27” iMac replacement) - it has certainly made me put it on hold until mid-June. And I really have to accept that Apple’s marketing strategy has changed, and that I’m going to have to live with it - or not. Unfortunately for me, ‘or not’ has been occurring to me frequently since March this year
As to AI, a lot of the motivation driving it appears to be the desire by managements to, to put it bluntly, sack people.
Most consumer dGPUs will be limited to something up to 32 GB. So it’s a natural advantage for certain workflows. Server GPUs have low hundreds of GB of memory, but not many people will have that on their desks.
September 2024, A18 and A18 Pro (iPhone 16, iPhone 16 Pro)
October/November 2024, M4, M4 Pro, M4 Max (MacBook Pro line, MacMini, MacStudio)
March 2025, M4 MacBook Airs, 24 inch iMac
I have a M2 Studio Ultra bought it Nov. 2023, Apple please release the M3 Studio Ultra with 256 gigs of UMA memory if you don't release to the public you can't make a sell, Apple at this time is in a position to leverage the M series computers laptop/desktop unique qualities and have a presence in this area of computing a presence they don't have in for example AAA gaming. Apple can't afford to let this pass it is more critical than playing AAA games and it will also require Apple write/port/create support software.
Apple needs to offer more powerful Apple Silicon computers the time is now not later, and as usual Apple has been acquiring small companies to support existing in house projects within Apple.
https://9to5mac.com/2024/04/22/apple-startup-acquire-ai-compression-and-computer-vision/ Datakalab a small French company (11 people will be joining Apple).
https://www.hardware-corner.net/guides/mac-for-large-language-models/
https://www.unite.ai/from-siri-to-realm-apples-journey-to-smarter-voice-assistants/ On the Edge.