x86 options for M class Macs?

I have Windows XP going in UTM. It runs amazingly fast. But then, it's also a very simple OS compared to Windows 10 or 11. While I can load those as Intel versions in UTM, painfully slow doesn't begin to describe how slow those OS versions are. They're basically unusable. And that's on an M4 Pro mini.

I did try the VMware, and while the ARM versions of Win 10 and 11 run at essentially the same speed as macOS, it's impossible to install an Intel version of Windows. Not even if you choose other/legacy. There you can at least select an Intel ISO of Windows, but it still won't install. Same with VirtualBox or Parallels. They won't make any attempt to install Intel Windows releases on an Mx Mac.

All of that seems rather strange since the entire purpose of a VM is to emulate a machine. UTM gives the user at least some Intel CPU emulation. Why don't the "upper class" VMs, such as Parallels and VMware?

I think you are going to have two Macs on your desk, if the issue is Windows drivers with no ARM version. Unless the vendors produce drivers for the ARM architecture, I think you might be SOL in a way that programs that might run under x86 conversion might not be.

You can try using UTM to run the Intel version of Windows, but this will likely run very slow since UTM is just an easy to use GUI wrapper for using the free open source QEMU software which is a system emulator & virtualizer.

https://mac.getutm.app/

I expect they haven't found a way to make Apple ARM --> x64 microcode translation fast enough to make it worth their while. Their big selling feature is making other CPU-compatible OS run on a Mac that otherwise wouldn't.

They will probably put their support behind Microsoft developing a fully-ARM compatible Windows before trying to solve a hard problem like remapping RISC to Intel CISC microcode without unworkable overhead (or having the hypervisor pin the SoC CPUs by consuming the entire speculative pipeline).

In theory that would work, but it would be very architecture dependent. Modern x64 - especially Intel's - chips have so many optimized opcode extensions for SSE, MMX, etc that I'm not sure the opcode rewriting could be done effectively.

And then there is the hardware side of the chips - very different pipelining and speculative execution paths, Apple's unified memory model for the CPU, GPU and neural cores - I haven't looked that deeply but what I have seen appears almost more like a Havard model architecture.

Like I said, it is probably not impossible to do, but I can't imagine Parallels or Broadcom deciding it is economically profitable (unless some university PG does it first, like what happened with Clang)