Clustering Macs for Compute Power

Any particular tasks in mind here that need those computes?

There’s no path for client consolidation nor for distributed computing integrated with current macOS, with the existing apps.

I’m well aware of clustering, and macOS just doesn’t have any offerings here.

Xsan and SMB shares only get you so far.

Not short of adding something akin to Apache Spark or Apache Hadoop. Or SETI@home.

None of which will be particularly integrated with any of the typical macOS client apps.

Xgrid worked well for some stuff and could make for faster builds with Xcode, but so too did SSD upgrades. But few other apps used it.

For some of what’s involved, have a look at whatever of the the Xgrid and distcc stuff that’s still around, and at the Apache stuff mentioned earlier.

Compute-intensive work loads such as image rendering will have means for distributing load and that may well include Apache Hadoop, but these efforts don’t usually involve macOS.

You would probably consider lower-end Mac or iOS or iPadOS clients and a more powerful server and running one or more guests in a virtual machine, if you were building this configuration anew.

Graphics buses and the processor-to-RAM buses are far faster than your network.

Primary cache is co-resident on the processor, and last-level cache is right next to the processor at its furthest distance, and the latency increases as the signal distance from the processor increases.

If you’re interested in caches and bandwidth and latency and coherence, here’s some reading:

https://software.rajivprab.com/2018/04/29/myths-programmers-believe-about-cpu-caches/

https://www.realworldtech.com/coherency/ (multiprocessors and from 2006, but still pretty good)

https://people.eecs.berkeley.edu/~rcs/research/interactive_latency.html (very visual)

The Mac hosting providers use Mac system hardware, either dedicated systems or with whatever macOS guests are permissible by license when used on Apple hardware.

Chunking stuff up within an app, that app—Blender rendering, as somebody else here has mentioned—dictates what distributed computing or clustering implementation gets used.

Forrest wrote:

As you point out, using virtual machines and such are less expensive and more efficient.

FWIW, macOS guests do not support accelerated graphics.

For blender I use crowd render and seems to work ok , I tie in one other iMac and a windows machine .

Welcome.