That really depends on what applications you run, and how many in parallel, or if it's one main app, how much it can make use of threads.
If you switch HT on, you need to switch to an SMP kernel, and with that you incur some additional overhead that may or may not eat up your gain. But it's worth a try.
The current wisdom seems to be that you get the most gain on dual-P4 machines (where you already run SMP, so you get the addt'l cycles for free). I run a cluster of dual P4's and get some 15-20% gain, as if I had 2.3 CPU's instead of 2. Again, that is specific to what runs on them.
I would try it and benchmark it. You can always go back.
mlp
|