I'm a bit suspicious that 2.0_B may have been a step backward on my Gallatin hypertheaded system--but have not made truly careful observations, nor testing.
The system in question is a Gallatin (Northwood-descended, but with the 2 Megabyte L3 onboard cache--sold as the first Extreme Edition P4), Windows XP Pro, but with 533 MHz memory (slow for a Gallatin). It runs Einstein (current release version) as the predominant BOINC application, with SETI currently given 10%, so nearly always when SETI runs it is paired with an Einstein on the other virtual processor. The specific KWSN 2.0_B ap I ran for a couple of weeks is:
SaH_5.15_KWSN_SSE2-Intel_Ben-Joe_2.0_B.exe
The previous KWSN ap to which I have now reverted and against which these informal comparisons apply is:
setiathome-5.15-kwsn-sse2-p4.exe
(which I'll call R-1.3, as it reports itself so)
I have suspected for some time that 2.0_B when running considerably slows down the paired Einstein WU, as compared to the throughput when the thread is running either an Einstein or the previous KWSN SETI.
This is disturbing, as historically (and very widely reported) SETI and Einstein were synergistic HT pairs, getting greatest total system throughput with one SETI and One Einstein runing.
It is not completely implausible that an otherwise desirable programming change can greatly harm HT behavior. Over on Einstein, there was one particular revision in Akos's wonderful series of improvement which abruptly lowered what had been up to then a mild HT gain to a very considerable HT loss (in both cases running a pair of Einstein instances). So there is precedent, not that the precedent proves anything about the current case.
One other symptom, which may be evidence:
With R-1.3 my CPU temperature is unaltered when the system switches from running a pair of Einstein's to one Einstein, one SETI. But with 2.0_B the CPU temperature drops about 2.5 degrees Fahrenheit, consistently. I can actually look at the temperature graph in my ASUSProbe log and tell when the system started and stopped running a SETI time slice by looking at the temperature graphs.
I grant that I've not provided remotely convincing evidence--I started the topic in case it might attract comment from others, possibly others who have had a bit of suspicion.
Even if I am right, it may not apply equally to all hyperthreaded models, which vary considerably.
