Hi Bjarne
My statement related specifically to DG's comment about latency when working with live players, not when recording MIDI. If you're working with live musicians and have your host running on a separate system, then as long as the musicians are monitoring themselves 'direct' or via a low latency host buffer setting, MIR latency will not be an issue. In fact DG intends to run everything on one system, in which case it clearly IS an issue, because MIR and the host would share the same buffer settings.
I'll have to check into the bypass function, I wasn't aware of it, but have not been using MIR for the last few days. When I ran some stress tests I found 512 samples (sorry, I keep quoting buffers in ms, brain not working) not to be a problem, and that involved MOL, and it also invovled audio being routed back to Logic on a separate machine and then from Logic onto another separate machine running ProTools HD via lightpipe. 1024 could become probelmatic, no doubt.
I would have thought that on a well specified i7 machine you would get very good performance at 256 samples, indeed it may deliver close to the Xeon spec I use. The only reason I ended up with a Xeon machine is because of the price of 4GB Ram modules for the i7 config, because I wanted the system to have 24GB. When I compared specs, I could get the dual Xeon system for the same price as a 24GB i7 rig because of the savings on Ram cost.
Cheers
Jules