On 4/27/11 10:53 AM, John Reiser wrote:
On 04/27/2011 07:34 AM, Adam Jackson wrote:
> But assuming all is right with the world,
> host-accelerated 3D would almost assuredly be faster than software guest 3D.
Does that include host software 3D with a guest that talks hardware 3D?
I have trouble deciphering what you could mean by that.
If you mean "guest passes through to host, host happens to be running
software 3D", that would probably be slower than "guest runs software 3D
directly". It'd be a tradeoff between which side has a more efficient
software implementation, how many CPUs you allocate to the guest, how
expensive your hypercalls are, etc. And it's certainly not a scenario I
was considering in the scope of "all is right with the world".
If you mean "host sets up device passthrough for a guest, but runs
software 3D for itself", that's a thing you could do I guess, but it's
not clear why you'd want to, or what you'd be comparing against to
determine relative speed.