I expect the Mini to get the base M2 with an option for up to 32GB RAM. That’s it. No mythical headless Mac. It will remain a consumer device.
Agree, hope you’re right about the memory, generally I think a Mini Pro is unlikely because most people who can afford that silicon will just get a MacBook Pro or an iMac Pro.
There’s a place for the Mini between the Air and the iMac. Not so much when it comes to the higher end.
an 8+2 chip isn't that powerful, barely a pro chip by any means. M2 will likely see similar bottlenecks, I don't know there's any LPDDR5 that can do 16GiB on a single die.
Seems a bit premature to make assumptions about M2. The A15 is not necessarily indicative of what to expect. The rumors that say the M2 will be a "marginal" advance beyond M1 are extrapolating from A14 versus A15. In other words, bullshit. In reality the M-series is a new branch of the tree and it could introduce elements that later find their way into the A-series, instead of vice-versa. TSMC and Apple are working together on this. Think different!
I expect the Mini to get the base M2 with an option for up to 32GB RAM. That’s it. No mythical headless Mac. It will remain a consumer device.
Agree, hope you’re right about the memory, generally I think a Mini Pro is unlikely because most people who can afford that silicon will just get a MacBook Pro or an iMac Pro.
There’s a place for the Mini between the Air and the iMac. Not so much when it comes to the higher end.
an 8+2 chip isn't that powerful, barely a pro chip by any means. M2 will likely see similar bottlenecks, I don't know there's any LPDDR5 that can do 16GiB on a single die.
Seems a bit premature to make assumptions about M2. The A15 is not necessarily indicative of what to expect. The rumors that say the M2 will be a "marginal" advance beyond M1 are extrapolating from A14 versus A15. In other words, bullshit. In reality the M-series is a new branch of the tree and it could introduce elements that later find their way into the A-series, instead of vice-versa. TSMC and Apple are working together on this. Think different!
Like, what, the M2 isn’t going to based on Blizzard and Avalanche? On the other hand, all major releases are set to 2023, both Intel and AMD. Actually, there’s little to no rumors about the M2 Pro/Max, but we do heard about the M3 early on.
I think we’ll see an M1 Mac Pro before the M2 comes out, or maybe an announcement at the same time. There will be four release/refresh stages for macOS devices (I posted this in the Air thread as well):
[1] Each generation of Macintosh Silicon will appear first in the MacBook Air and the Mac Mini. Both will be silent masterpieces of technology with minuscule failure rates, with no fan, utterly reliable.
[2] Next come the iMac and the MacBook, with the colors and the same silicon as the Air and Mini. These are consumer Macs, with lower prices and higher failure rates.
[3] Then the MacBook Pro gets its refresh with the new Pro and Max configurations.
[4] Finally, the iMac Pro and the Mac Pro complete the cycle, with multiple dies and GPU advances. Depending on what they do with the Mac Pro, this stage could be split into two phases.
All of this takes place over a cycle of about 18 months, with some flexibility built into it. Apple Silicon will not make promises it can't keep. It won't be like clockwork, and it won't be an annual cycle. macOS, however, will stay on an annual cycle, because it has to keep up with more than just changes in the M series, but that doesn't mean the hardware will.
The 16" MacBook Pro will get two dies, the cooling is sufficient. I've been figuring out the thermals and I'm confident it will.
The power would be high for this. The wall power of the highest 16" has been measured at around 100W, with an extra M1 Max, it would be 170W. The power supply is only 140W. The 16" MBP will be able to get close to a Duo chip on 3nm in a couple of years, there won't be waiting times like on Intel. I reckon the options will be:
mini, 24" iMac, Air, 13" Pro = M1 14"/16" MBP = M1 Pro, M1 Max 27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB) Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
I think we’ll see an M1 Mac Pro before the M2 comes out, or maybe an announcement at the same time. There will be four release/refresh stages for macOS devices (I posted this in the Air thread as well):
[1] Each generation of Macintosh Silicon will appear first in the MacBook Air and the Mac Mini. Both will be silent masterpieces of technology with minuscule failure rates, with no fan, utterly reliable.
[2] Next come the iMac and the MacBook, with the colors and the same silicon as the Air and Mini. These are consumer Macs, with lower prices and higher failure rates.
[3] Then the MacBook Pro gets its refresh with the new Pro and Max configurations.
[4] Finally, the iMac Pro and the Mac Pro complete the cycle, with multiple dies and GPU advances. Depending on what they do with the Mac Pro, this stage could be split into two phases.
All of this takes place over a cycle of about 18 months, with some flexibility built into it. Apple Silicon will not make promises it can't keep. It won't be like clockwork, and it won't be an annual cycle. macOS, however, will stay on an annual cycle, because it has to keep up with more than just changes in the M series, but that doesn't mean the hardware will.
The 16" MacBook Pro will get two dies, the cooling is sufficient. I've been figuring out the thermals and I'm confident it will.
The power would be high for this. The wall power of the highest 16" has been measured at around 100W, with an extra M1 Max, it would be 170W. The power supply is only 140W. The 16" MBP will be able to get close to a Duo chip on 3nm in a couple of years, there won't be waiting times like on Intel. I reckon the options will be:
mini, 24" iMac, Air, 13" Pro = M1 14"/16" MBP = M1 Pro, M1 Max 27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB) Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
Yeah, hard to believe, but it's doable. No way we can put two GPUs in a laptop, and throttling to an extent is acceptable. At least the 16+4 CPU will be implemented and won't throttle by itself. I'd expect somewhere 60+60 when running both, then about 90 watts for GPU alone.
Apple also has a proud tradition of shipping a smaller charger, so relying on batteries isn't out of the question 😂
The 14" has a much smaller cooler with less power, but was able to match the 16" in performance. All of that indicates the 16" still hasn't reached its full potential.
On the other hand, 8+2 is NOT equivalent to Intel's 12th-gen i9s, they're more or less just high-power ultrabooks, not enough for proper mobile workstations.
I think we’ll see an M1 Mac Pro before the M2 comes out, or maybe an announcement at the same time. There will be four release/refresh stages for macOS devices (I posted this in the Air thread as well):
[1] Each generation of Macintosh Silicon will appear first in the MacBook Air and the Mac Mini. Both will be silent masterpieces of technology with minuscule failure rates, with no fan, utterly reliable.
[2] Next come the iMac and the MacBook, with the colors and the same silicon as the Air and Mini. These are consumer Macs, with lower prices and higher failure rates.
[3] Then the MacBook Pro gets its refresh with the new Pro and Max configurations.
[4] Finally, the iMac Pro and the Mac Pro complete the cycle, with multiple dies and GPU advances. Depending on what they do with the Mac Pro, this stage could be split into two phases.
All of this takes place over a cycle of about 18 months, with some flexibility built into it. Apple Silicon will not make promises it can't keep. It won't be like clockwork, and it won't be an annual cycle. macOS, however, will stay on an annual cycle, because it has to keep up with more than just changes in the M series, but that doesn't mean the hardware will.
The 16" MacBook Pro will get two dies, the cooling is sufficient. I've been figuring out the thermals and I'm confident it will.
The power would be high for this. The wall power of the highest 16" has been measured at around 100W, with an extra M1 Max, it would be 170W. The power supply is only 140W. The 16" MBP will be able to get close to a Duo chip on 3nm in a couple of years, there won't be waiting times like on Intel. I reckon the options will be:
mini, 24" iMac, Air, 13" Pro = M1 14"/16" MBP = M1 Pro, M1 Max 27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB) Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
You also mentioned HBM. I just talked to my friend, then we realized the biggest bottleneck for any Apple Silicon is RAM capacity. M2 will likely be stuck on 16GiB, M1 Max is 64GiB, 16+4 @ 128. 4-die workstations only allow 256GiB and M3 @ 32/128/256/512. Apple will be abandoning 1.5TiB of RAM for sure. Hopefully, HBM will ease the pain but might be too expensive.
I think we’ll see an M1 Mac Pro before the M2 comes out, or maybe an announcement at the same time. There will be four release/refresh stages for macOS devices (I posted this in the Air thread as well):
[1] Each generation of Macintosh Silicon will appear first in the MacBook Air and the Mac Mini. Both will be silent masterpieces of technology with minuscule failure rates, with no fan, utterly reliable.
[2] Next come the iMac and the MacBook, with the colors and the same silicon as the Air and Mini. These are consumer Macs, with lower prices and higher failure rates.
[3] Then the MacBook Pro gets its refresh with the new Pro and Max configurations.
[4] Finally, the iMac Pro and the Mac Pro complete the cycle, with multiple dies and GPU advances. Depending on what they do with the Mac Pro, this stage could be split into two phases.
All of this takes place over a cycle of about 18 months, with some flexibility built into it. Apple Silicon will not make promises it can't keep. It won't be like clockwork, and it won't be an annual cycle. macOS, however, will stay on an annual cycle, because it has to keep up with more than just changes in the M series, but that doesn't mean the hardware will.
The 16" MacBook Pro will get two dies, the cooling is sufficient. I've been figuring out the thermals and I'm confident it will.
The power would be high for this. The wall power of the highest 16" has been measured at around 100W, with an extra M1 Max, it would be 170W. The power supply is only 140W. The 16" MBP will be able to get close to a Duo chip on 3nm in a couple of years, there won't be waiting times like on Intel. I reckon the options will be:
mini, 24" iMac, Air, 13" Pro = M1 14"/16" MBP = M1 Pro, M1 Max 27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB) Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
You also mentioned HBM. I just talked to my friend, then we realized the biggest bottleneck for any Apple Silicon is RAM capacity. M2 will likely be stuck on 16GiB, M1 Max is 64GiB, 16+4 @ 128. 4-die workstations only allow 256GiB and M3 @ 32/128/256/512. Apple will be abandoning 1.5TiB of RAM for sure. Hopefully, HBM will ease the pain but might be too expensive.
16 GByte LPPDR5 DRAM has been sampling for awhile. Apple maybe using them in the M1 Pro/Max. Either that or they are putting 2 8-GByte LPDDR5 packages per 128-bit channel in the M1 Pro/Max. Once the 16 GB LPDDR5 DRAM packages become available, which should be 2022, Apple can have up to 32 GB for the M2 if they use the same RAM packaging design (2 packages adjacent to the die). If they are using 8 GB LPDDR5 packages in the M1 Pro/Max, it will also mean their RAM capacities can double too, so 128 GB for "M2 Max" machines, and they can move up to 512 GB in a 4-die MCM with 16 channels.
HBM2e will be RAM capacity limited. It's either SK Hynix or Micron, maybe Samsung, forgot which, that are making 12-layer stacks of HBM2e, with each layer at 2 GByte. So 12 layers makes 24 GB per stack. 8 HBM2e stacks makes 192 GB. 8 stacks is about as much as all the exotic MCM CPUs and GPUs will go in 2022. If Apple uses HBM for the Duo and Quad MCMs, 256 GB is about as much as they will go. That's like 16 stacks of 8-hi HBM. 8 TB/s of bandwidth, if they can design a memory controller that interface with that much memory bandwidth.
It would be very interesting if they use HBM, but it sounds like a costly solution for them. I'm thinking if they could stack 16 GB LPDDR5 in 4 layers, they can have 512 GB with 8 channels of LPDDR5, or 1 TB if they used all 4 channels of every die. They will be able to use normal PCB substrates, no need for silicon interposers or bridges, and it will have over 1 TB/s bandwidth, assuming all the memory controller stuff between 4 dies is worked out. Just all around cheaper, and sounds something like they would be more apt to do.
I think we’ll see an M1 Mac Pro before the M2 comes out, or maybe an announcement at the same time. There will be four release/refresh stages for macOS devices (I posted this in the Air thread as well):
[1] Each generation of Macintosh Silicon will appear first in the MacBook Air and the Mac Mini. Both will be silent masterpieces of technology with minuscule failure rates, with no fan, utterly reliable.
[2] Next come the iMac and the MacBook, with the colors and the same silicon as the Air and Mini. These are consumer Macs, with lower prices and higher failure rates.
[3] Then the MacBook Pro gets its refresh with the new Pro and Max configurations.
[4] Finally, the iMac Pro and the Mac Pro complete the cycle, with multiple dies and GPU advances. Depending on what they do with the Mac Pro, this stage could be split into two phases.
All of this takes place over a cycle of about 18 months, with some flexibility built into it. Apple Silicon will not make promises it can't keep. It won't be like clockwork, and it won't be an annual cycle. macOS, however, will stay on an annual cycle, because it has to keep up with more than just changes in the M series, but that doesn't mean the hardware will.
The 16" MacBook Pro will get two dies, the cooling is sufficient. I've been figuring out the thermals and I'm confident it will.
The power would be high for this. The wall power of the highest 16" has been measured at around 100W, with an extra M1 Max, it would be 170W. The power supply is only 140W. The 16" MBP will be able to get close to a Duo chip on 3nm in a couple of years, there won't be waiting times like on Intel. I reckon the options will be:
mini, 24" iMac, Air, 13" Pro = M1 14"/16" MBP = M1 Pro, M1 Max 27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB) Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
You also mentioned HBM. I just talked to my friend, then we realized the biggest bottleneck for any Apple Silicon is RAM capacity. M2 will likely be stuck on 16GiB, M1 Max is 64GiB, 16+4 @ 128. 4-die workstations only allow 256GiB and M3 @ 32/128/256/512. Apple will be abandoning 1.5TiB of RAM for sure. Hopefully, HBM will ease the pain but might be too expensive.
16 GByte LPPDR5 DRAM has been sampling for awhile. Apple maybe using them in the M1 Pro/Max. Either that or they are putting 2 8-GByte LPDDR5 packages per 128-bit channel in the M1 Pro/Max. Once the 16 GB LPDDR5 DRAM packages become available, which should be 2022, Apple can have up to 32 GB for the M2 if they use the same RAM packaging design (2 packages adjacent to the die). If they are using 8 GB LPDDR5 packages in the M1 Pro/Max, it will also mean their RAM capacities can double too, so 128 GB for "M2 Max" machines, and they can move up to 512 GB in a 4-die MCM with 16 channels.
HBM2e will be RAM capacity limited. It's either SK Hynix or Micron, maybe Samsung, forgot which, that are making 12-layer stacks of HBM2e, with each layer at 2 GByte. So 12 layers makes 24 GB per stack. 8 HBM2e stacks makes 192 GB. 8 stacks is about as much as all the exotic MCM CPUs and GPUs will go in 2022. If Apple uses HBM for the Duo and Quad MCMs, 256 GB is about as much as they will go. That's like 16 stacks of 8-hi HBM. 8 TB/s of bandwidth, if they can design a memory controller that interface with that much memory bandwidth.
It would be very interesting if they use HBM, but it sounds like a costly solution for them. I'm thinking if they could stack 16 GB LPDDR5 in 4 layers, they can have 512 GB with 8 channels of LPDDR5, or 1 TB if they used all 4 channels of every die. They will be able to use normal PCB substrates, no need for silicon interposers or bridges, and it will have over 1 TB/s bandwidth, assuming all the memory controller stuff between 4 dies is worked out. Just all around cheaper, and sounds something like they would be more apt to do.
I mean the single RAM chip, that means those skinny ones on the M1. I doubt we can see that in 2022, sometimes the industry moves pretty slow, but we’ll see. I also don’t believe the existence of M2 Pro chips. I’m in odds but Apple doesn’t need to follow on a yearly basis, especially when all it does is 10% increase depending on the thermal. M3s, with 3nm and ARMv9 are a much bigger deal.
I think we’ll see an M1 Mac Pro before the M2 comes out, or maybe an announcement at the same time. There will be four release/refresh stages for macOS devices (I posted this in the Air thread as well):
[1] Each generation of Macintosh Silicon will appear first in the MacBook Air and the Mac Mini. Both will be silent masterpieces of technology with minuscule failure rates, with no fan, utterly reliable.
[2] Next come the iMac and the MacBook, with the colors and the same silicon as the Air and Mini. These are consumer Macs, with lower prices and higher failure rates.
[3] Then the MacBook Pro gets its refresh with the new Pro and Max configurations.
[4] Finally, the iMac Pro and the Mac Pro complete the cycle, with multiple dies and GPU advances. Depending on what they do with the Mac Pro, this stage could be split into two phases.
All of this takes place over a cycle of about 18 months, with some flexibility built into it. Apple Silicon will not make promises it can't keep. It won't be like clockwork, and it won't be an annual cycle. macOS, however, will stay on an annual cycle, because it has to keep up with more than just changes in the M series, but that doesn't mean the hardware will.
The 16" MacBook Pro will get two dies, the cooling is sufficient. I've been figuring out the thermals and I'm confident it will.
The power would be high for this. The wall power of the highest 16" has been measured at around 100W, with an extra M1 Max, it would be 170W. The power supply is only 140W. The 16" MBP will be able to get close to a Duo chip on 3nm in a couple of years, there won't be waiting times like on Intel. I reckon the options will be:
mini, 24" iMac, Air, 13" Pro = M1 14"/16" MBP = M1 Pro, M1 Max 27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB) Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
How did the 32c GPU perform again? I believe it's pretty close to a mobile 3080 isn't it? May just bump to 40c/48c after all.
I think we’ll see an M1 Mac Pro before the M2 comes out, or maybe an announcement at the same time. There will be four release/refresh stages for macOS devices (I posted this in the Air thread as well):
[1] Each generation of Macintosh Silicon will appear first in the MacBook Air and the Mac Mini. Both will be silent masterpieces of technology with minuscule failure rates, with no fan, utterly reliable.
[2] Next come the iMac and the MacBook, with the colors and the same silicon as the Air and Mini. These are consumer Macs, with lower prices and higher failure rates.
[3] Then the MacBook Pro gets its refresh with the new Pro and Max configurations.
[4] Finally, the iMac Pro and the Mac Pro complete the cycle, with multiple dies and GPU advances. Depending on what they do with the Mac Pro, this stage could be split into two phases.
All of this takes place over a cycle of about 18 months, with some flexibility built into it. Apple Silicon will not make promises it can't keep. It won't be like clockwork, and it won't be an annual cycle. macOS, however, will stay on an annual cycle, because it has to keep up with more than just changes in the M series, but that doesn't mean the hardware will.
The 16" MacBook Pro will get two dies, the cooling is sufficient. I've been figuring out the thermals and I'm confident it will.
The power would be high for this. The wall power of the highest 16" has been measured at around 100W, with an extra M1 Max, it would be 170W. The power supply is only 140W. The 16" MBP will be able to get close to a Duo chip on 3nm in a couple of years, there won't be waiting times like on Intel. I reckon the options will be:
mini, 24" iMac, Air, 13" Pro = M1 14"/16" MBP = M1 Pro, M1 Max 27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB) Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
You also mentioned HBM. I just talked to my friend, then we realized the biggest bottleneck for any Apple Silicon is RAM capacity. M2 will likely be stuck on 16GiB, M1 Max is 64GiB, 16+4 @ 128. 4-die workstations only allow 256GiB and M3 @ 32/128/256/512. Apple will be abandoning 1.5TiB of RAM for sure. Hopefully, HBM will ease the pain but might be too expensive.
16 GByte LPPDR5 DRAM has been sampling for awhile. Apple maybe using them in the M1 Pro/Max. Either that or they are putting 2 8-GByte LPDDR5 packages per 128-bit channel in the M1 Pro/Max. Once the 16 GB LPDDR5 DRAM packages become available, which should be 2022, Apple can have up to 32 GB for the M2 if they use the same RAM packaging design (2 packages adjacent to the die). If they are using 8 GB LPDDR5 packages in the M1 Pro/Max, it will also mean their RAM capacities can double too, so 128 GB for "M2 Max" machines, and they can move up to 512 GB in a 4-die MCM with 16 channels.
HBM2e will be RAM capacity limited. It's either SK Hynix or Micron, maybe Samsung, forgot which, that are making 12-layer stacks of HBM2e, with each layer at 2 GByte. So 12 layers makes 24 GB per stack. 8 HBM2e stacks makes 192 GB. 8 stacks is about as much as all the exotic MCM CPUs and GPUs will go in 2022. If Apple uses HBM for the Duo and Quad MCMs, 256 GB is about as much as they will go. That's like 16 stacks of 8-hi HBM. 8 TB/s of bandwidth, if they can design a memory controller that interface with that much memory bandwidth.
It would be very interesting if they use HBM, but it sounds like a costly solution for them. I'm thinking if they could stack 16 GB LPDDR5 in 4 layers, they can have 512 GB with 8 channels of LPDDR5, or 1 TB if they used all 4 channels of every die. They will be able to use normal PCB substrates, no need for silicon interposers or bridges, and it will have over 1 TB/s bandwidth, assuming all the memory controller stuff between 4 dies is worked out. Just all around cheaper, and sounds something like they would be more apt to do.
I mean the single RAM chip, that means those skinny ones on the M1. I doubt we can see that in 2022, sometimes the industry moves pretty slow, but we’ll see. I also don’t believe the existence of M2 Pro chips. I’m in odds but Apple doesn’t need to follow on a yearly basis, especially when all it does is 10% increase depending on the thermal. M3s, with 3nm and ARMv9 are a much bigger deal.
That's what I mean too. The M1 is using 64 Gbit density LPDDR4X "chips". Ie, 8 GByte RAM packages. What you see is a package. Inside of it could be one RAM die, 2 dies, 4 dies, etc, depending on RAM density. 8 GByte RAM packages is a practical upper limit for mass market LPDDR4X RAM. Two of them in the M1 makes 16 GByte total.
128 Gbit LPDDR5 densities have been sampling (test production) for awhile now. This will enable those "skinny RAM chips" adjacent to the M1 SoC to be 16 GByte capacity. So, Apple's successor to the M1 can have up to 32 GB RAM capacity. The M1 Pro/Max are using LPDDR5 memory, so it is already supported. That can come as soon as the next round of updates for the MBA, iMac, Mac mini.
I'm thinking Apple may actually be using 16 GByte LPDDR5 RAM in the M1 Pro/Max, but there has been no verification of it. It could just be 2 8 GByte LPDDR5 chips inside the packaging adjacent to the SoC. All the RAM vendors have 128 Gbit density as sampling or a scarce high end SKU, so it's really uncertain if Apple's using 128 Gbit LPDDR5 densities.
I think we’ll see an M1 Mac Pro before the M2 comes out, or maybe an announcement at the same time. There will be four release/refresh stages for macOS devices (I posted this in the Air thread as well):
[1] Each generation of Macintosh Silicon will appear first in the MacBook Air and the Mac Mini. Both will be silent masterpieces of technology with minuscule failure rates, with no fan, utterly reliable.
[2] Next come the iMac and the MacBook, with the colors and the same silicon as the Air and Mini. These are consumer Macs, with lower prices and higher failure rates.
[3] Then the MacBook Pro gets its refresh with the new Pro and Max configurations.
[4] Finally, the iMac Pro and the Mac Pro complete the cycle, with multiple dies and GPU advances. Depending on what they do with the Mac Pro, this stage could be split into two phases.
All of this takes place over a cycle of about 18 months, with some flexibility built into it. Apple Silicon will not make promises it can't keep. It won't be like clockwork, and it won't be an annual cycle. macOS, however, will stay on an annual cycle, because it has to keep up with more than just changes in the M series, but that doesn't mean the hardware will.
The 16" MacBook Pro will get two dies, the cooling is sufficient. I've been figuring out the thermals and I'm confident it will.
The power would be high for this. The wall power of the highest 16" has been measured at around 100W, with an extra M1 Max, it would be 170W. The power supply is only 140W. The 16" MBP will be able to get close to a Duo chip on 3nm in a couple of years, there won't be waiting times like on Intel. I reckon the options will be:
mini, 24" iMac, Air, 13" Pro = M1 14"/16" MBP = M1 Pro, M1 Max 27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB) Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
You also mentioned HBM. I just talked to my friend, then we realized the biggest bottleneck for any Apple Silicon is RAM capacity. M2 will likely be stuck on 16GiB, M1 Max is 64GiB, 16+4 @ 128. 4-die workstations only allow 256GiB and M3 @ 32/128/256/512. Apple will be abandoning 1.5TiB of RAM for sure. Hopefully, HBM will ease the pain but might be too expensive.
16 GByte LPPDR5 DRAM has been sampling for awhile. Apple maybe using them in the M1 Pro/Max. Either that or they are putting 2 8-GByte LPDDR5 packages per 128-bit channel in the M1 Pro/Max. Once the 16 GB LPDDR5 DRAM packages become available, which should be 2022, Apple can have up to 32 GB for the M2 if they use the same RAM packaging design (2 packages adjacent to the die). If they are using 8 GB LPDDR5 packages in the M1 Pro/Max, it will also mean their RAM capacities can double too, so 128 GB for "M2 Max" machines, and they can move up to 512 GB in a 4-die MCM with 16 channels.
HBM2e will be RAM capacity limited. It's either SK Hynix or Micron, maybe Samsung, forgot which, that are making 12-layer stacks of HBM2e, with each layer at 2 GByte. So 12 layers makes 24 GB per stack. 8 HBM2e stacks makes 192 GB. 8 stacks is about as much as all the exotic MCM CPUs and GPUs will go in 2022. If Apple uses HBM for the Duo and Quad MCMs, 256 GB is about as much as they will go. That's like 16 stacks of 8-hi HBM. 8 TB/s of bandwidth, if they can design a memory controller that interface with that much memory bandwidth.
It would be very interesting if they use HBM, but it sounds like a costly solution for them. I'm thinking if they could stack 16 GB LPDDR5 in 4 layers, they can have 512 GB with 8 channels of LPDDR5, or 1 TB if they used all 4 channels of every die. They will be able to use normal PCB substrates, no need for silicon interposers or bridges, and it will have over 1 TB/s bandwidth, assuming all the memory controller stuff between 4 dies is worked out. Just all around cheaper, and sounds something like they would be more apt to do.
I mean the single RAM chip, that means those skinny ones on the M1. I doubt we can see that in 2022, sometimes the industry moves pretty slow, but we’ll see. I also don’t believe the existence of M2 Pro chips. I’m in odds but Apple doesn’t need to follow on a yearly basis, especially when all it does is 10% increase depending on the thermal. M3s, with 3nm and ARMv9 are a much bigger deal.
That's what I mean too. The M1 is using 64 Gbit density LPDDR4X "chips". Ie, 8 GByte RAM packages. What you see is a package. Inside of it could be one RAM die, 2 dies, 4 dies, etc, depending on RAM density. 8 GByte RAM packages is a practical upper limit for mass market LPDDR4X RAM. Two of them in the M1 makes 16 GByte total.
128 Gbit LPDDR5 densities have been sampling (test production) for awhile now. This will enable those "skinny RAM chips" adjacent to the M1 SoC to be 16 GByte capacity. So, Apple's successor to the M1 can have up to 32 GB RAM capacity. The M1 Pro/Max are using LPDDR5 memory, so it is already supported. That can come as soon as the next round of updates for the MBA, iMac, Mac mini.
I'm thinking Apple may actually be using 16 GByte LPDDR5 RAM in the M1 Pro/Max, but there has been no verification of it. It could just be 2 8 GByte LPDDR5 chips inside the packaging adjacent to the SoC. All the RAM vendors have 128 Gbit density as sampling or a scarce high end SKU, so it's really uncertain if Apple's using 128 Gbit LPDDR5 densities.
I expect the Mini to get the base M2 with an option for up to 32GB RAM. That’s it. No mythical headless Mac. It will remain a consumer device.
Agree, hope you’re right about the memory, generally I think a Mini Pro is unlikely because most people who can afford that silicon will just get a MacBook Pro or an iMac Pro.
There’s a place for the Mini between the Air and the iMac. Not so much when it comes to the higher end.
an 8+2 chip isn't that powerful, barely a pro chip by any means. M2 will likely see similar bottlenecks, I don't know there's any LPDDR5 that can do 16GiB on a single die.
Seems a bit premature to make assumptions about M2. The A15 is not necessarily indicative of what to expect. The rumors that say the M2 will be a "marginal" advance beyond M1 are extrapolating from A14 versus A15. In other words, bullshit. In reality the M-series is a new branch of the tree and it could introduce elements that later find their way into the A-series, instead of vice-versa. TSMC and Apple are working together on this. Think different!
Like, what, the M2 isn’t going to based on Blizzard and Avalanche? On the other hand, all major releases are set to 2023, both Intel and AMD. Actually, there’s little to no rumors about the M2 Pro/Max, but we do heard about the M3 early on.
Sorry, this is just wrong. I missed this response earlier. Correct — the M2 is not going to be based on the A15, it will use a 4nm process like the A16, the codenames for the M2 are Staten and Rhodes. Staten is due in mid 2022, Rhodes (Pro/Max) is expected in early 2023.
There has been very little on the M3, except the likely true leaked information about it using TSMC’s still-in-development 3nm process, and it being likely to be the first generation to support quad-die configurations.
I expect the Mini to get the base M2 with an option for up to 32GB RAM. That’s it. No mythical headless Mac. It will remain a consumer device.
Agree, hope you’re right about the memory, generally I think a Mini Pro is unlikely because most people who can afford that silicon will just get a MacBook Pro or an iMac Pro.
There’s a place for the Mini between the Air and the iMac. Not so much when it comes to the higher end.
an 8+2 chip isn't that powerful, barely a pro chip by any means. M2 will likely see similar bottlenecks, I don't know there's any LPDDR5 that can do 16GiB on a single die.
Seems a bit premature to make assumptions about M2. The A15 is not necessarily indicative of what to expect. The rumors that say the M2 will be a "marginal" advance beyond M1 are extrapolating from A14 versus A15. In other words, bullshit. In reality the M-series is a new branch of the tree and it could introduce elements that later find their way into the A-series, instead of vice-versa. TSMC and Apple are working together on this. Think different!
Like, what, the M2 isn’t going to based on Blizzard and Avalanche? On the other hand, all major releases are set to 2023, both Intel and AMD. Actually, there’s little to no rumors about the M2 Pro/Max, but we do heard about the M3 early on.
Sorry, this is just wrong. I missed this response earlier. Correct — the M2 is not going to be based on the A15, it will use a 4nm process like the A16, the codenames for the M2 are Staten and Rhodes. Staten is due in mid 2022, Rhodes (Pro/Max) is expected in early 2023.
There has been very little on the M3, except the likely true leaked information about it using TSMC’s still-in-development 3nm process, and it being likely to be the first generation to support quad-die configurations.
And my skepticism: 1). Just a reminder that N3 will be Q1 2023, 4nm M2 in 2023 sounds really slow. All the big opponents will show up in 2023, do we really expect ASi will fight them with an inferior process? I don't trust that "18-month" rumor because of this. 2). “Avalanche” and "Blizzard" are codenames for the CPU architecture, "Staten" and "Rhodes" sound like referring to the entire SoC. Some sites say it has the same CPU as the M1. 3). Let's not forget about the GPU, which still lags behind in terms of ray tracing. No words on M2 for that. So I doubt the M2 will be based on A16. However, these new chips could be called M2, maybe.
I expect the Mini to get the base M2 with an option for up to 32GB RAM. That’s it. No mythical headless Mac. It will remain a consumer device.
Agree, hope you’re right about the memory, generally I think a Mini Pro is unlikely because most people who can afford that silicon will just get a MacBook Pro or an iMac Pro.
There’s a place for the Mini between the Air and the iMac. Not so much when it comes to the higher end.
an 8+2 chip isn't that powerful, barely a pro chip by any means. M2 will likely see similar bottlenecks, I don't know there's any LPDDR5 that can do 16GiB on a single die.
Seems a bit premature to make assumptions about M2. The A15 is not necessarily indicative of what to expect. The rumors that say the M2 will be a "marginal" advance beyond M1 are extrapolating from A14 versus A15. In other words, bullshit. In reality the M-series is a new branch of the tree and it could introduce elements that later find their way into the A-series, instead of vice-versa. TSMC and Apple are working together on this. Think different!
Like, what, the M2 isn’t going to based on Blizzard and Avalanche? On the other hand, all major releases are set to 2023, both Intel and AMD. Actually, there’s little to no rumors about the M2 Pro/Max, but we do heard about the M3 early on.
Sorry, this is just wrong. I missed this response earlier. Correct — the M2 is not going to be based on the A15, it will use a 4nm process like the A16, the codenames for the M2 are Staten and Rhodes. Staten is due in mid 2022, Rhodes (Pro/Max) is expected in early 2023.
There has been very little on the M3, except the likely true leaked information about it using TSMC’s still-in-development 3nm process, and it being likely to be the first generation to support quad-die configurations.
And my skepticism: 1). Just a reminder that N3 will be Q1 2023, 4nm M2 in 2023 sounds really slow. All the big opponents will show up in 2023, do we really expect ASi will fight them with an inferior process? I don't trust that "18-month" rumor because of this. 2). “Avalanche” and "Blizzard" are codenames for the CPU architecture, "Staten" and "Rhodes" sound like referring to the entire SoC. Some sites say it has the same CPU as the M1. 3). Let's not forget about the GPU, which still lags behind in terms of ray tracing. No words on M2 for that. So I doubt the M2 will be based on A16. However, these new chips could be called M2, maybe.
Sorry, I didn’t mean to imply Staten and Rhodes were parallel to Avalanche and Blizzard. We don’t know the codenames for the A16 (4nm, N4P) right? Avalanche and Blizzard are the old codenames for the A15 (5nm, N5P) right?
I don’t know what you’re referring to about the N4P — my understanding is it’s an improvement over N5P, but not hugely so. What Apple does with that (i.e., A15 versus A16, and M1 versus M2) is a different question.
Maybe Apple is holding out for N3P for the M3? So their timetable will lag behind that of the N3. Really I doubt Apple is very worried about whatever Intel or AMD is doing at this point. They are focused on their own thing.
I expect the Mini to get the base M2 with an option for up to 32GB RAM. That’s it. No mythical headless Mac. It will remain a consumer device.
Agree, hope you’re right about the memory, generally I think a Mini Pro is unlikely because most people who can afford that silicon will just get a MacBook Pro or an iMac Pro.
There’s a place for the Mini between the Air and the iMac. Not so much when it comes to the higher end.
an 8+2 chip isn't that powerful, barely a pro chip by any means. M2 will likely see similar bottlenecks, I don't know there's any LPDDR5 that can do 16GiB on a single die.
Seems a bit premature to make assumptions about M2. The A15 is not necessarily indicative of what to expect. The rumors that say the M2 will be a "marginal" advance beyond M1 are extrapolating from A14 versus A15. In other words, bullshit. In reality the M-series is a new branch of the tree and it could introduce elements that later find their way into the A-series, instead of vice-versa. TSMC and Apple are working together on this. Think different!
Like, what, the M2 isn’t going to based on Blizzard and Avalanche? On the other hand, all major releases are set to 2023, both Intel and AMD. Actually, there’s little to no rumors about the M2 Pro/Max, but we do heard about the M3 early on.
Sorry, this is just wrong. I missed this response earlier. Correct — the M2 is not going to be based on the A15, it will use a 4nm process like the A16, the codenames for the M2 are Staten and Rhodes. Staten is due in mid 2022, Rhodes (Pro/Max) is expected in early 2023.
There has been very little on the M3, except the likely true leaked information about it using TSMC’s still-in-development 3nm process, and it being likely to be the first generation to support quad-die configurations.
And my skepticism: 1). Just a reminder that N3 will be Q1 2023, 4nm M2 in 2023 sounds really slow. All the big opponents will show up in 2023, do we really expect ASi will fight them with an inferior process? I don't trust that "18-month" rumor because of this. 2). “Avalanche” and "Blizzard" are codenames for the CPU architecture, "Staten" and "Rhodes" sound like referring to the entire SoC. Some sites say it has the same CPU as the M1. 3). Let's not forget about the GPU, which still lags behind in terms of ray tracing. No words on M2 for that. So I doubt the M2 will be based on A16. However, these new chips could be called M2, maybe.
Sorry, I didn’t mean to imply Staten and Rhodes were parallel to Avalanche and Blizzard. We don’t know the codenames for the A16 (4nm, N4P) right? Avalanche and Blizzard are the old codenames for the A15 (5nm, N5P) right?
I don’t know what you’re referring to about the N4P — my understanding is it’s an improvement over N5P, but not hugely so. What Apple does with that (i.e., A15 versus A16, and M1 versus M2) is a different question.
Maybe Apple is holding out for N3P for the M3? So their timetable will lag behind that of the N3. Really I doubt Apple is very worried about whatever Intel or AMD is doing at this point. They are focused on their own thing.
Well, 13900K will have 24-core (8P+16e). Their E-core is equivalent to Skylake @ 2.4GHz. 14900K next year will top at 40-core plus N3, 2-die ASi won't match that. Granted the 139K/149K is nowhere mainstream, but you can see the scale of the competition.
Surely nobody wants to think "switching is a grave mistake" isn't it?
Comments
mini, 24" iMac, Air, 13" Pro = M1
14"/16" MBP = M1 Pro, M1 Max
27" iMac = M1 Pro ($1999 - 16GB/512GB), M1 Max ($2,799 - 32GB/512GB), M1 Max Duo ($4,999 - 64GB/512GB)
Mac Pro = M1 Max Duo ($4,999 - 64GB/512GB), M1 Max Quad ($7,999 - 128GB/512GB)
I assume the Duo/Quad chips will use HBM memory. If not, they could be cheaper.
The iMac form factor could easily handle an M1 Max Quad but maybe they want the Mac Pro option to have accessible SSDs and to use the 32" display.
Apple also has a proud tradition of shipping a smaller charger, so relying on batteries isn't out of the question 😂
If we checked what's inside, we noticed that fans in the 16" are significantly larger.
M1 Max 16":
Touch bar 16": https://www.notebookcheck.com/fileadmin/_processed_/d/d/csm_DSC08864_3b15979d82.jpg
The 14" has a much smaller cooler with less power, but was able to match the 16" in performance. All of that indicates the 16" still hasn't reached its full potential.
On the other hand, 8+2 is NOT equivalent to Intel's 12th-gen i9s, they're more or less just high-power ultrabooks, not enough for proper mobile workstations.
HBM2e will be RAM capacity limited. It's either SK Hynix or Micron, maybe Samsung, forgot which, that are making 12-layer stacks of HBM2e, with each layer at 2 GByte. So 12 layers makes 24 GB per stack. 8 HBM2e stacks makes 192 GB. 8 stacks is about as much as all the exotic MCM CPUs and GPUs will go in 2022. If Apple uses HBM for the Duo and Quad MCMs, 256 GB is about as much as they will go. That's like 16 stacks of 8-hi HBM. 8 TB/s of bandwidth, if they can design a memory controller that interface with that much memory bandwidth.
It would be very interesting if they use HBM, but it sounds like a costly solution for them. I'm thinking if they could stack 16 GB LPDDR5 in 4 layers, they can have 512 GB with 8 channels of LPDDR5, or 1 TB if they used all 4 channels of every die. They will be able to use normal PCB substrates, no need for silicon interposers or bridges, and it will have over 1 TB/s bandwidth, assuming all the memory controller stuff between 4 dies is worked out. Just all around cheaper, and sounds something like they would be more apt to do.
I also don’t believe the existence of M2 Pro chips. I’m in odds but Apple doesn’t need to follow on a yearly basis, especially when all it does is 10% increase depending on the thermal. M3s, with 3nm and ARMv9 are a much bigger deal.
128 Gbit LPDDR5 densities have been sampling (test production) for awhile now. This will enable those "skinny RAM chips" adjacent to the M1 SoC to be 16 GByte capacity. So, Apple's successor to the M1 can have up to 32 GB RAM capacity. The M1 Pro/Max are using LPDDR5 memory, so it is already supported. That can come as soon as the next round of updates for the MBA, iMac, Mac mini.
I'm thinking Apple may actually be using 16 GByte LPDDR5 RAM in the M1 Pro/Max, but there has been no verification of it. It could just be 2 8 GByte LPDDR5 chips inside the packaging adjacent to the SoC. All the RAM vendors have 128 Gbit density as sampling or a scarce high end SKU, so it's really uncertain if Apple's using 128 Gbit LPDDR5 densities.
https://www.micron.com/products/dram/lpdram/part-catalog/mt62f2g64dafl-031-wt
I don't know whether that "128Gb" means one of those 2-die package or just one silicon die. Manufacturers always refer to the entire package.
1). Just a reminder that N3 will be Q1 2023, 4nm M2 in 2023 sounds really slow. All the big opponents will show up in 2023, do we really expect ASi will fight them with an inferior process? I don't trust that "18-month" rumor because of this.
2). “Avalanche” and "Blizzard" are codenames for the CPU architecture, "Staten" and "Rhodes" sound like referring to the entire SoC. Some sites say it has the same CPU as the M1.
3). Let's not forget about the GPU, which still lags behind in terms of ray tracing. No words on M2 for that.
So I doubt the M2 will be based on A16. However, these new chips could be called M2, maybe.
I don’t know what you’re referring to about the N4P — my understanding is it’s an improvement over N5P, but not hugely so. What Apple does with that (i.e., A15 versus A16, and M1 versus M2) is a different question.
Maybe Apple is holding out for N3P for the M3? So their timetable will lag behind that of the N3. Really I doubt Apple is very worried about whatever Intel or AMD is doing at this point. They are focused on their own thing.
Surely nobody wants to think "switching is a grave mistake" isn't it?