4 ms·

What's in it for Apple? I'm not trying to be glib, here, but unless there were some Mac only server functionality, nobody would buy an Apple ARM powered datacen

by jfb 5y ago

What's in it for Apple? I'm not trying to be glib, here, but unless there were some Mac only server functionality, nobody would buy an Apple ARM powered datacentre machine.

mcintyre1994 5y ago
What if Rosetta 2 was that Mac only server functionality? I don't know that they'd do it, but from M1 Mac reviews it sounds like M1 + Rosetta 2 will run at least some x86 code faster and more power efficiently than any competitor. I don't know how feasible it is to scale that up to a datacenter though, and I expect MacOS licensing costs would wipe away any power efficiency gains. But I do wonder if they could hypothetically scale up and beat the best Intel/AMD have to offer just using Rosetta 2 to translate x86 code.
gpm 5y ago
Eh, the Asahi linux people already have linux running on this chip. What's in it for Apple is money and better economies of scale for chips. But I don't really think it fits Apple's MO so I doubt they'll do it.
- mumblemumble 5y ago
  MO is my thought, too. Getting back into server hardware would require supporting a vastly different kind of client, and possibly require them to start doing some things that they, as a company, might find distasteful. Supporting equipment they stopped making a long time ago, for example. Including maintaining stockpiles of replacement parts.
- AceJohnny2 5y ago
  > Eh, the Asahi linux people already have linux running on this chip More specifically, people are running Linux on the CPu cores. The M1 is a system-on-chip, and according to the floorplan [1], the CPUs are maybe 1/5th of the chip. There are many other features that aren't unlocked, such as GPU (which is a brand new architecture) or power management. The latter is key to exploiting the chip to its full performance envelope. I don't expect Asahi to get anywhere further than proof-of-concept before it becomes obsolete by the march of the silicon industry. [1] https://images.anandtech.com/doci/16226/M1.png https://images.anandtech.com/doci/16226/M1.png
  - gpm 5y ago
    I think it depends on how much changes between generations. So far it seems like most of this isn't really new, but almost exactly what's been on the iDevices for a long time. If they don't re-architect substantially between generations I can see the Asahi project keeping up. The GPU stuff is new, but it seems like good progress is being made: https://rosenzweig.io/blog/asahi-gpu-part-3.html https://rosenzweig.io/blog/asahi-gpu-part-3.html For data centers, it helps that GPU and so on is just less important. It's wasted Silicon, but the CPU is competitive even before considering the added features so that's not the end of the world. There's a good chance that Apple can use that to their advantage too, by using chips with broken GPUs/NNPUs in the DC... or designing a smaller chip for data centers... or one with more cores... or so on.
epistasis 5y ago
If Linux was supported, it would be an interesting competitor to AWS's graviton instances. As for what's in it for Apple, it would be the profit from selling a hopefully large number of chips, but adding official Linux support and also commuting to an entire new market for a minimum of three years is probably far higher a cost on focus than any potential profits.
usrusr 5y ago
With a sufficiently wide lead in energy efficiency, just selling hardware without any follow up lock-in harvest can be attractive even for a company as spoiled as Apple. They'd likely want to make the modules offered sufficiently big to make them unattractive for desktop use or else they'd risk cannibalizing their lock-in market.
easton 5y ago
They get a gigantic boost in internal datacenter performance if they can jam a bunch of Apple Silicon chips into a rack mounted server and boot Linux on it. If they can get a similar boost in performance at lower power efficiency in the chip that is going in the Mac Pro, taking that chip and putting it on a board with Ethernet and Power wouldn't be a ton of engineering cost and then they could massively reduce the power consumption and cooling costs of the datacenters. And then they could resell the box they designed as a server, either with Linux support out of the box (unlikely, but since in this mystical scenario they'd have to write kernel patches for Linux to get it to boot...) or with a build of macOS that could be set up headlessly, in order to recoup the development costs. Apple shops that want machines to run Compressor or Xcode's build server would eat that up.
_ph_ 5y ago
First of all, Apple could save a huge amount of money replacing Intel based servers with their own chips. Both on the CPU price, expecially the Xeons are really expensive as well as on electricity consumption, probably the largest running cost of data centers. Then the gains of scale, making a CPU just of the Mac Pro would mean too low production numbers, but with data center usage would drive those up - especially if Apple also sold it to other customers, e.g. bringing the Xserve back. For the OS they could run Linux virtualized or they give the native Linux on Mac developers a hand.
- jfb 5y ago
  The amount of money that Apple would save to put a bunch of 2U ARM servers into racks is dwarfed by the costs of building such systems. Seriously, nobody (to the first approximation) is going to buy a macOS server, and so there's no reason for Apple to do this.

by jfb 5y ago

What's in it for Apple? I'm not trying to be glib, here, but unless there were some Mac only server functionality, nobody would buy an Apple ARM powered datacentre machine.

mcintyre1994 5y ago

What if Rosetta 2 was that Mac only server functionality? I don't know that they'd do it, but from M1 Mac reviews it sounds like M1 + Rosetta 2 will run at least some x86 code faster and more power efficiently than any competitor. I don't know how feasible it is to scale that up to a datacenter though, and I expect MacOS licensing costs would wipe away any power efficiency gains. But I do wonder if they could hypothetically scale up and beat the best Intel/AMD have to offer just using Rosetta 2 to translate x86 code.

gpm 5y ago

Eh, the Asahi linux people already have linux running on this chip. What's in it for Apple is money and better economies of scale for chips. But I don't really think it fits Apple's MO so I doubt they'll do it.

mumblemumble 5y ago
MO is my thought, too. Getting back into server hardware would require supporting a vastly different kind of client, and possibly require them to start doing some things that they, as a company, might find distasteful. Supporting equipment they stopped making a long time ago, for example. Including maintaining stockpiles of replacement parts.
AceJohnny2 5y ago
> Eh, the Asahi linux people already have linux running on this chip More specifically, people are running Linux on the CPu cores. The M1 is a system-on-chip, and according to the floorplan [1], the CPUs are maybe 1/5th of the chip. There are many other features that aren't unlocked, such as GPU (which is a brand new architecture) or power management. The latter is key to exploiting the chip to its full performance envelope. I don't expect Asahi to get anywhere further than proof-of-concept before it becomes obsolete by the march of the silicon industry. [1] https://images.anandtech.com/doci/16226/M1.png https://images.anandtech.com/doci/16226/M1.png
- gpm 5y ago
  I think it depends on how much changes between generations. So far it seems like most of this isn't really new, but almost exactly what's been on the iDevices for a long time. If they don't re-architect substantially between generations I can see the Asahi project keeping up. The GPU stuff is new, but it seems like good progress is being made: https://rosenzweig.io/blog/asahi-gpu-part-3.html https://rosenzweig.io/blog/asahi-gpu-part-3.html For data centers, it helps that GPU and so on is just less important. It's wasted Silicon, but the CPU is competitive even before considering the added features so that's not the end of the world. There's a good chance that Apple can use that to their advantage too, by using chips with broken GPUs/NNPUs in the DC... or designing a smaller chip for data centers... or one with more cores... or so on.

epistasis 5y ago

If Linux was supported, it would be an interesting competitor to AWS's graviton instances. As for what's in it for Apple, it would be the profit from selling a hopefully large number of chips, but adding official Linux support and also commuting to an entire new market for a minimum of three years is probably far higher a cost on focus than any potential profits.

usrusr 5y ago

With a sufficiently wide lead in energy efficiency, just selling hardware without any follow up lock-in harvest can be attractive even for a company as spoiled as Apple. They'd likely want to make the modules offered sufficiently big to make them unattractive for desktop use or else they'd risk cannibalizing their lock-in market.

easton 5y ago

They get a gigantic boost in internal datacenter performance if they can jam a bunch of Apple Silicon chips into a rack mounted server and boot Linux on it. If they can get a similar boost in performance at lower power efficiency in the chip that is going in the Mac Pro, taking that chip and putting it on a board with Ethernet and Power wouldn't be a ton of engineering cost and then they could massively reduce the power consumption and cooling costs of the datacenters. And then they could resell the box they designed as a server, either with Linux support out of the box (unlikely, but since in this mystical scenario they'd have to write kernel patches for Linux to get it to boot...) or with a build of macOS that could be set up headlessly, in order to recoup the development costs. Apple shops that want machines to run Compressor or Xcode's build server would eat that up.

_ph_ 5y ago

First of all, Apple could save a huge amount of money replacing Intel based servers with their own chips. Both on the CPU price, expecially the Xeons are really expensive as well as on electricity consumption, probably the largest running cost of data centers. Then the gains of scale, making a CPU just of the Mac Pro would mean too low production numbers, but with data center usage would drive those up - especially if Apple also sold it to other customers, e.g. bringing the Xserve back. For the OS they could run Linux virtualized or they give the native Linux on Mac developers a hand.

jfb 5y ago
The amount of money that Apple would save to put a bunch of 2U ARM servers into racks is dwarfed by the costs of building such systems. Seriously, nobody (to the first approximation) is going to buy a macOS server, and so there's no reason for Apple to do this.