Are there any DC projects that actually utilize dual-core?

[jasong]

There are plenty of projects out there that can clone themselves to use more than one cpu, but are there any projects out there where it's one workunit utilizing all the cpus simultaneously?

[the-mk]

for instance: distributed.net, BOINC (that are the last one I've used with my AMD 3800 X2)

[PCZ]

Jasong

Most projects only use one core.
To use all your cores you run multiple instances of the client.

Some clients detect how many cores you have and run several instances in parallel, for example Boinc and dnet are 2 clients that do this.
SOB can also spawn a fixed number of instances when run as a service.
This is manually set though.

DPAD is multi threaded and uses all your cores on a single task.
This isn't very efficient though and higher output can be achieved by limiting the client to one process and running multiple instances.


Basically the rule is one process/instance per core.
Splitting a single task accross multi cores doesn't work very well.
The single task gets a speed boost but greater overall production is achieved by running X {x = number of cores] tasks in parallel.

[cwhyl]

In CPDN Carl have made some tests on parallel computing on two cores. It didn't work out very well but seems to be on the "wish list".

[tnerual]

one more project detecting the number of cpu is Rieselsieve boinc ... but it's still running one instance per cpu / core

runing very well on my quad (number 18 in the top computer in rieselsieve)

and for info there is some multi-processor app like a LLR prime verification that are running one thread on all your processor, but to achieve 1.5 time the work of one processor you have to use 4 processor ...

it's usefull to verify very large prime on some "beast" with 16 or 32 processor ... but it's better to run one instance per core ...

[jasong]

one more project detecting the number of cpu is Rieselsieve boinc ... but it's still running one instance per cpu / core

runing very well on my quad (number 18 in the top computer in rieselsieve)

and for info there is some multi-processor app like a LLR prime verification that are running one thread on all your processor, but to achieve 1.5 time the work of one processor you have to use 4 processor ...

it's usefull to verify very large prime on some "beast" with 16 or 32 processor ... but it's better to run one instance per core ...
Since you all said basically the same thing, I'm just going to quote the most recent post.

To be perfectly honest, I had forgotten about the slowdown. Thanks for the info.