3 ms·

What is the biggest hurdle for the M2 Mac Pro? Is it connectivity that isn't suppported by the chip? E.g. when they want external DRAM and GPU, the M2 SoCs simp

What is the biggest hurdle for the M2 Mac Pro? Is it connectivity that isn't suppported by the chip? E.g. when they want external DRAM and GPU, the M2 SoCs simply don't have that capability so a whole new SoC would be required, with features that would only be used in the Pro, making it a questionable investment? Or is there a more technical hurdle, e.g. that supporting external Dram and third party GPU would require more fundamental changes than merely building a new SoC?

zozbot234 4y ago
Couldn't you connect external DRAM to a Mac Studio via Thunderbolt if you really wanted to? Or is that still missing the same memory coherence semantics as actual DRAM sticks?
- formerly_proven 4y ago
  Thunderbolt has around two orders of magnitude less bandwidth and two orders of magnitude higher latency than the DRAM in that machine. Viewed through the narrow <40 Gbit/s TB3/4 funnel, you'll have trouble telling DRAM and a PCIe 4 SSD apart.
ngcc_hk 4y ago
Could it be many still need CUDA and hence unless they can somehow fit in Nvidia they … But they do not want to. Hence not sure what is the point other than they can win the war against CUDA.
- alkonaut 4y ago
  CUDA or not, I don't think many will buy a mac pro unless it delivers what the last one did in terms of Graphics. And that probably requires at least using separate graphics card in the several-hundred-watt class. They obviously aren't going to shove a 400W graphics card into the SoC and they aren't going to shrink the power requirements by an order of magnitude for GPUs. An they also aren't going to enter the business of making discrete pro graphics cards. nvidia and AMD can afford to do that research work because it trickles down to consumer/gamer cards. But Apple surely can't motivate that research expense. So the only thing that makes sense is to either make no Mac pro tower workstation with discrete GPUs at all, or doing what they did last time, using third party cards.

zozbot234 4y ago

Couldn't you connect external DRAM to a Mac Studio via Thunderbolt if you really wanted to? Or is that still missing the same memory coherence semantics as actual DRAM sticks?

formerly_proven 4y ago
Thunderbolt has around two orders of magnitude less bandwidth and two orders of magnitude higher latency than the DRAM in that machine. Viewed through the narrow <40 Gbit/s TB3/4 funnel, you'll have trouble telling DRAM and a PCIe 4 SSD apart.

ngcc_hk 4y ago

Could it be many still need CUDA and hence unless they can somehow fit in Nvidia they … But they do not want to. Hence not sure what is the point other than they can win the war against CUDA.

alkonaut 4y ago
CUDA or not, I don't think many will buy a mac pro unless it delivers what the last one did in terms of Graphics. And that probably requires at least using separate graphics card in the several-hundred-watt class. They obviously aren't going to shove a 400W graphics card into the SoC and they aren't going to shrink the power requirements by an order of magnitude for GPUs. An they also aren't going to enter the business of making discrete pro graphics cards. nvidia and AMD can afford to do that research work because it trickles down to consumer/gamer cards. But Apple surely can't motivate that research expense. So the only thing that makes sense is to either make no Mac pro tower workstation with discrete GPUs at all, or doing what they did last time, using third party cards.