Hi. I have a duo-core machine and it crunches 2 WUs at a time just fine.

What I'd like to do is have both cores process only 1 WU at a time...process the same WU in fact. Why would I want to do this?...so the graphics will run/dance around faster. I would only do it for a few days while show it to some friends (I typically run the Advanced side of BOINC). This request is really just for a coolness factor.

Theoretically there is no difference in 2 cores processing 1 or 2 WUs at a time...simply the speed at which *each* WU will be fully processed.

Is there any way to do this? I have 3gigs of RAM, a stripped down XP Pro box, and an Intel duo-core chip and BOINC 6.2.19.

Thanks!

-Eric

---

I think the quick answer is no. I asked the same question some time ago and got this response:

No.

The case where that might occur would be with a hyperthreaded vs. not hyperthreaded CPU, because both threads must use the same math coprocessor.

---

i also think its the way RAH/BOINC is made. one task per core.

I think the quick answer is no. I asked the same question some time ago and got this response:

No.

The case where that might occur would be with a hyperthreaded vs. not hyperthreaded CPU, because both threads must use the same math coprocessor.

i think this is the good ol' difference between serial and parallel processing.

someone who knows better, correct me where i'm wrong, but...

i think that a Rosie wu must be processed serially on a core; but more than one wu can be (serially) processed in parallel on a multi-core cpu.

Nope, this would require that the Rosetta be designed for multithreading, and BOINC is not designed for scheduling multithreaded apps either. I believe BOINC support is coming in a future release of 6.x but I doubt a version Rosetta will be released to take advantage of it.

In order for multithreading to work, you need to (a) have an algorithm that is easily parallelizable and (b) a reason to do this since multithreading often has an inherent performance impact due to interprocess communication overhead (locking, memory, etc). Perhaps a project that needs results very fast or has very long computation times would see a reason for it.

The case where that might occur would be with a hyperthreaded vs. not hyperthreaded CPU, because both threads must use the same math coprocessor.

It doesn't quite work that way either, hyperthreading is the same as dual-core as far as software is concerned, except that there is not as much performance gain from going non-HT to HT as compared to single-core to dual-core. HT does not let you run a single thread faster. There was an old rumor flying around of a "reverse hyperthreading" in an upcoming AMD CPU which would do something akin to dynamically allocating resources between hardware threads but it hasn't materialized in Intel/AMD's CPU roadmaps. However, there is a feature in Sun's Rock processor called Hardware scout in which the second thread is used to speed up a single application. However, your not likely to see such a CPU designed for backend server applications in a typical desktop ;)

---