Why stop at Intel Macs? heck if they could put M1 on a PCI expansion card, Let all Windows PC's run Mac/iOS software and the world is your oyster!
No... Just no. Apple is not going to indefinitely support both Intel and Apple Silicon based Macs. Running Windows on a Mac is very much an edge case and not something Apple should be concerned about AT ALL. If you want to run Mac OS, buy a Mac. If you want to run Windows, buy a PC. It's that simple.
There’s nothing special about the memory being on-package.
I wonder if that’s totally true....
anandtech’s review seems to show no advantage in terms of latency or bandwidth relative to other PCs (unless I’ve misinterpreted). So that seems to support your statement.
But might there be an advantage in terms of watts?
Right now, the on-package RAM is "normal" LPDDR4X. This is slightly exotic in that it only saves a relatively small amount of power compared to normal DDR4 or LPDDR4, and most vendors only care about that level of power draw reduction in phones. The current 13" MacBook Pro with 4TB3 uses it with off-package chips, though. On-package LPDDR4X has maybe a few microwatts less power draw due to the shorter traces compared to off-package, but that should be it.
On-package RAM can use stricter timing, but LPDDR4X isn't enough to need that. HBM and HBM2 do need it, and I'm very slightly disappointed Apple didn't go with that for their on-package RAM.
I think the choice to go with on-package was related more to the physical package technology they're using. Specifically, the M1 is physically a lot like the A12X and A12Z. I bet the M1 has almost exactly the same number of pins as the A12X. I think they wanted to go low-effort with the M1 and just use the same packaging technology they already used. That lets them prove out the core designs (not just CPU and GPU, but Thunderbolt, and so on) while TSMC works on the larger packaging needed for off-package RAM.
There’s nothing special about the memory being on-package.
I wonder if that’s totally true....
anandtech’s review seems to show no advantage in terms of latency or bandwidth relative to other PCs (unless I’ve misinterpreted). So that seems to support your statement.
But might there be an advantage in terms of watts?
Right now, the on-package RAM is "normal" LPDDR4X. This is slightly exotic in that it only saves a relatively small amount of power compared to normal DDR4 or LPDDR4, and most vendors only care about that level of power draw reduction in phones. The current 13" MacBook Pro with 4TB3 uses it with off-package chips, though. On-package LPDDR4X has maybe a few microwatts less power draw due to the shorter traces compared to off-package, but that should be it.
On-package RAM can use stricter timing, but LPDDR4X isn't enough to need that. HBM and HBM2 do need it, and I'm very slightly disappointed Apple didn't go with that for their on-package RAM.
I think the choice to go with on-package was related more to the physical package technology they're using. Specifically, the M1 is physically a lot like the A12X and A12Z. I bet the M1 has almost exactly the same number of pins as the A12X. I think they wanted to go low-effort with the M1 and just use the same packaging technology they already used. That lets them prove out the core designs (not just CPU and GPU, but Thunderbolt, and so on) while TSMC works on the larger packaging needed for off-package RAM.
Thanks -- that's informative and makes sense.
I was also disappointed they didn't use HBM. But maybe they will in the future? I've read that HBM3 (or whatever it's called) can be combined with less expensive (relative to using a silicon interposer) packaging to achieve high bandwidth, low latency, at a reasonable(-ish) cost.
Maybe they could put 8 or 16 GB of HBM3 in the SOC package, and then traditional user-upgradeable DIMMs on the motherboard (at least in desktop systems, anyway)
I sure hope the Mac Pros will have PCIe 4 slots. We use 32gig fiber cards in our editing systems.
Since the M1 already uses PCIe 4, it seems like any Mac with card slots will necessarily support PCIe4.
And were did you see that? The M1 chip has a custom SSD controller, That does not make it PCIe4. Unless you think that USB4 is the same, and it is not.
I sure hope the Mac Pros will have PCIe 4 slots. We use 32gig fiber cards in our editing systems.
Since the M1 already uses PCIe 4, it seems like any Mac with card slots will necessarily support PCIe4.
And were did you see that? The M1 chip has a custom SSD controller, That does not make it PCIe4. Unless you think that USB4 is the same, and it is not.
I sure hope the Mac Pros will have PCIe 4 slots. We use 32gig fiber cards in our editing systems.
Since the M1 already uses PCIe 4, it seems like any Mac with card slots will necessarily support PCIe4.
And were did you see that? The M1 chip has a custom SSD controller, That does not make it PCIe4. Unless you think that USB4 is the same, and it is not.
Apple.
That's for the MVMe storage Not for x4, x8 and x16 lane slots. That would call for a chip above the M1.
I sure hope the Mac Pros will have PCIe 4 slots. We use 32gig fiber cards in our editing systems.
Since the M1 already uses PCIe 4, it seems like any Mac with card slots will necessarily support PCIe4.
And were did you see that? The M1 chip has a custom SSD controller, That does not make it PCIe4. Unless you think that USB4 is the same, and it is not.
Apple.
That's for the MVMe storage Not for x4, x8 and x16 lane slots. That would call for a chip above the M1.
Do they need more lanes? Of course. The M1 is a low-end SoC. But you asked where did I find PCIe4. There it is.
I don't see even 32 happening for anything but the Mac Pro. I saw forum comments recently that games don't and can't use more than six cores. How much RAM (and heat) would you need to feed 32 cores?
Plus, the games offerings on Mac is pretty terrible compared to Windows (has been since the beginning) so people interested in desktop gaming don’t gravitate towards Mac anyway, creative professionals do. I agree 32 cores is for a small group mostly, e.g post production or heavy 3D rendering, but even there the GPU is the more important component (until you render to a film clip or something).
Comments
On-package RAM can use stricter timing, but LPDDR4X isn't enough to need that. HBM and HBM2 do need it, and I'm very slightly disappointed Apple didn't go with that for their on-package RAM.
I think the choice to go with on-package was related more to the physical package technology they're using. Specifically, the M1 is physically a lot like the A12X and A12Z. I bet the M1 has almost exactly the same number of pins as the A12X. I think they wanted to go low-effort with the M1 and just use the same packaging technology they already used. That lets them prove out the core designs (not just CPU and GPU, but Thunderbolt, and so on) while TSMC works on the larger packaging needed for off-package RAM.
I was also disappointed they didn't use HBM. But maybe they will in the future? I've read that HBM3 (or whatever it's called) can be combined with less expensive (relative to using a silicon interposer) packaging to achieve high bandwidth, low latency, at a reasonable(-ish) cost.
Maybe they could put 8 or 16 GB of HBM3 in the SOC package, and then traditional user-upgradeable DIMMs on the motherboard (at least in desktop systems, anyway)
And were did you see that? The M1 chip has a custom SSD controller, That does not make it PCIe4. Unless you think that USB4 is the same, and it is not.
That's for the MVMe storage Not for x4, x8 and x16 lane slots. That would call for a chip above the M1.
I agree 32 cores is for a small group mostly, e.g post production or heavy 3D rendering, but even there the GPU is the more important component (until you render to a film clip or something).