It's been a while since Apple has released a multiprocessor computer (not since the PPC era, I think). Is it even possible to do so on Apple Silicon? If so, would there be a performance benefit worth the extra cost? What would be the likelihood that Apple would do so?
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Is a multiprocessor AS system possible?
- Thread starter Axemantitan
- Start date
- Sort by reaction score
There is no logical reason to have a multiprocessor system when Apple silicon is not a CPU but a SOC. The loss in efficiency versus simply having a larger wafer with more cores would be a lot.
Absolutely possible, though likely you would need an external crossbar chip. Not at all practical - the die is much smaller than the reticle, so smarter to just put more cores on the die, and avoid the performance problems caused by multiprocessors.
Apple has been granted advanced patents with regards to combining multiple processing chips (aka chiplets) into one unified compute cluster. They are much more likely to go this route instead of multiprocessors. The later come with major performance and efficiency disadvantages.
Back in the 1990's, where processors only tended to have a single core, it made sense to have multi-processor computers. These days CPUs have multiple cores. AMD even has 64-core x86 processors. I think that they days of multi-processor desktop & laptops are long gone.
Back many, many years ago, mainframe systems had a variety of approaches to dual and quad processors. Ranging from one box with two physical processors, or two boxes each with one processor, through to what amounted to two entirely separate computers which just communicated and could move workload between the two (or four) computers. (We knew them as nodes.)
Something towards a nodal system could make much sense if trying to scale up AS.
Something towards a nodal system could make much sense if trying to scale up AS.
Multi-processor systems still have a place in server (and, to a limited degree, HPC) market, where you need to handle a lot of independent parallel threads. But they offer no benefit whatsoever in the regular customer designs, as you rightfully point out.
It’s still a thing. Install one of many open-source batch servers on your array of macs and enjoy a makeshift cluster. This kind of scalability is very limited of course.
The nodal architecture I was thinking about was somewhat tighter than a makeshift cluster but, logically, much the same. Little things that mattered - like real time clocks, were tightly synchronised to a microsecond. (Each RTC clicked every four microseconds, and each one was set to start at 0, 1, 2 or 3. So the same time could never occur on two nodes. Guaranteed uniqueness.)
Yes, I was avoiding muddying the waters by mentioning servers, since Apple aren't in that market.
There were Mac Pros from the 1st generation that had dual Xeons in them
He was responding to the first sentence of my original post, where I thought the last multiprocessor systems were PPC.