The 25% I'm seeing is in line with what I observed when I first got a Pentium 4 with hyperthreading. With just a single core it's easier to measure. I ran a number-crunching program and measured it's time to execute, then I ran two copies of the program. The two simultaneously ran slower, of course, but overall I got about 125% the computational rate.
But as I previously mentioned, the dual Xeons with HT was a different story because using two processors it would assign two executing threads to the same processor, losing the benefit of dual processors. We always ran with HT disabled on these systems.
But as I previously mentioned, the dual Xeons with HT was a different story because using two processors it would assign two executing threads to the same processor, losing the benefit of dual processors. We always ran with HT disabled on these systems.
