I'm running CPDN on a 3.60Ghz dual Xeon w/HT. I noticed significant differences between benchmarks with HT 'On' and HT 'Off'. With HT off, the Integer speed is doubled.

Should I keep HT off, or turn it back on?
/nja

---

The benchmarks mean very little in CP. Typically you should get about 15% more throughput with HT on. This means that each model takes about 70% longer but you are doing 2 models at the same time. If you are an enthusiatic CP cruncher, the higher throughput of using HT is good. If you are only going to crunch for a short period - enough to do a model without HT then perhaps off is best.

---

Visit BOINC WIKI for help

And join BOINC Synergy for all the news in one place.

The benchmarks are not used in CPDN, and an unreliable guide to how quickly work will be done. In practice, with HT turned on you will complete two WUs, run simultaneously, about 10-15% faster than two run sequentially with it turned off. Of course, you need general preferences to be set to use two processors (the default).

EDIT - snap!

---

> I'm running CPDN on a 3.60Ghz dual Xeon w/HT. I noticed significant
> differences between benchmarks with HT 'On' and HT 'Off'. With HT off, the
> Integer speed is doubled.
>
> Should I keep HT off, or turn it back on?
> /nja
>

Run HT with CPDN if you are running Windows XP. Turn off HT if you are running Windows 2000.

---

Thanks for all the replies.

1) Suspend Boinc
2) With Task Manager, set affintiy of boinc_gui.exe to one CPU.
3) Run Benchmarks...see that the results are the same with HT off. :-)
4) With Task Manager, set affintiy of boinc_gui.exe back to ALL CPUs.
5) Un-suspend Boinc.
6) With Task Manager, set affintiy of each boinc process to one CPU each.
7) Repeat step 6, as new boinc processes are created,...whenever. Just trying illiminate as much context switching as possible.

I found that if you leave the boinc_gui locked on one CPU when you resume processing. All the subsequent processes that are created are on that one CPU.

Thanks again for your comments,
/nja

---