So, the question is, assuming RAM isn\'t a problem, does crunching go quicker (in the total throughput sense) with two models per CPU or with one, when running on a NetBurst architecture chip?

Running two models with hyperthreading increases throughput by about 15% on a single processor P4. On dual Xeons with hyperthreading, there doesn\'t appear to be any advantage due to front side bus/memory access/contention problems.

Thanks for the info. Maybe once my 4 models are out of the way I\'ll get two more models, not four, especially as I\'m thinking of trying the attribution project and four models would be somewhat heavy!

Running two models with hyperthreading increases throughput by about 15% on a single processor P4. On dual Xeons with hyperthreading, there doesn\'t appear to be any advantage due to front side bus/memory access/contention problems.

Does this mean that I can run 8 models on my Q6600?
I have 4 GB 800MHz ram

The Core processors don\'t have hyperthreads (it\'s only the old Pentium 4 \'netburst\' range which did).

---

I'm a volunteer and my views are my own.
News and Announcements and FAQ