I was thinking of adding the SMP client to a system and am curious as to how well it performs. Could you share with us the specs and output of the system(s) you are running the SMP on?
It's a Mac Pro with two dual-core 'Woodcrest' Xeon CPUs, running at 2.66 GHz. I have 2 GBytes of RAM. Although it's not relevant for Folding, my GPU is an Nvidia GeForce 7300GT with 256 MBytes.
Since I installed the SMP client it's run seven work units in three days for Projects 3024, 3025, 3026. Each unit has taken about 10 hours and they are worth over 600 points each. These work units are set up specifically for high availability 2 or 4 core systems, and their preferred completion is 1 day, with a deadline of 2 days. My system's productivity, measured using the points values, has gone up by a factor of about 20 compared with its previous level using the normal single CPU console client.
While it is running, if my system is just doing normal office-type work, the four cores are utilised about 50-60% by FaH. One runs a bit higher, presumably doing traffic cop overhead work.
The SMP software frightened the life out of me when I first started to run it, as I was monitoring system activity and noticed that the network traffic was off the scale. I thought I would be kicked off the Internet by my ISP. However, it turns out this is just loopback traffic between the cores, and never leaves my system.
I'm a bit concerned that the old system I want to put it on won't be up to the task of meeting the SMP work unit deadlines for finishing and reporting back to Stanford.
There's been some discussion in the FaH forum about whether lower availability, or lower spec, machines can deliver the tight deadlines that are set for the SMP work units. I get the impression that you really need to have a fairly current rig to do it, using Intel's Core-2 Duo (or the new quads that are emerging).
I apologise that I didn't realise the SMP client is not available for Windows yet. I'm so used to Mac stuff trailing Windows that it never occurred to me that I might actually be ahead of the curve