Finally found a way to lock the affinity of each unit automatically :)

I found this great affinity/priority program Bill2's Process Manager. Over the last 27hours on my machine it has used 39 seconds of CPU time, it does however use some Ram, currently 120Mb and 100MB virtual on mine (see sig). The great feature of it is you can set a rule for any process (32bit or 64bit, XP or Vista). You can set the rule to be a certain core or a different core for each of the same process . So I set the rule for AK_v8_win_x64_SSE3.exe to "Distribute each instance on a different CPU". Rough guess at this early stage is 5% gain . It does check each rule every 10min by default to make sure it's abiding, however this can be changed to check every 60min, you can also change default priority of any process. Give it a shoot and tell me what you think. Also make sure to choose English language when installing (default french).

I've been out of it for a few days - why do you want to set the affinity?
I can understand changing the priority of a WU but the affinity?

Anyway, since I recently had to reinstall (HDD failure) I'll give this a try instead of the one I was using before ...

I'm know expert here but it has to do with the process having to switch back and forth between cores which results in a slow down as the information is passed from for instance core 0's cpu cache to core 1's cpu cache. Suffice to say without setting the affinity the process(es) switch back and forth between the 2 cores to share the load.

Never thought about it that way ...

Hmm, interesting. I'll try it on my back-up rig and see. Worse than can happen is it'll crash my system and I'll have to reinstall, lol.

Btw, good to see ya QI!

I can see where affinity would be an advantage with a Phenom system - different core speeds if you're OC'ing it right. And I have considered running an F@H GPU app while running SETIx4. Might be good to have the SETI apps fixed and let the GPU app run on any core whenever it needs CPU access.


[hijack]

Finally limping along with the new HDD and a re-install. I still think I'm going to use one of my newer rigs as a new main rig but I've been procrastinating so it'll take a week of testing first to get the right combo. I've got three boards (790FX/SB600 and 2x 790GX/SB750 - one ASUS, one Biostar) and three chips (9850 and 2x 9950). One of the 790GX boards will become an HTPC, but it's still a lot of testing ...

[/hijack]

Nebulous said:

Hmm, interesting. I'll try it on my back-up rig and see. Worse than can happen is it'll crash my system and I'll have to reinstall, lol.

Btw, good to see ya QI!

Click to expand...

I'll leave the potential reinstall to others discretion. Just thought 5% of our RAC would be a nice little boost (350,263 + 5% = 367,776 )

is this another program to go OVER the AK optimizer program?

Is the proggy in German?

silent bob said:

is this another program to go OVER the AK optimizer program?

Click to expand...

Not sure what your refering to...you mean the drop in SSE optimized SETI client made by AK?

Is the proggy in German?

Click to expand...

The default language is french. When I installed it I choose only to install English.

Also be sure under options to turn the Automatic priority setting off (untick the checkbox).

I have noticed in Vista it does not start up with admin privileges, if the process manager starts with almost all processes showing in red then I have to restart it with admin privileges to allow it to change the affinities.

Psykoikonov said:

Not sure what your refering to...you mean the drop in SSE optimized SETI client made by AK?

Click to expand...

yes , on my quad , in Task show each core being maxed , 25% each

silent bob said:

yes , on my quad , in Task show each core being maxed , 25% each

Click to expand...

I have not tested on a quad, should work the same. Basically without setting affinity the 4 instances of the client don't run on a single cpu core, each instance runs 25% on each core. Which means each client is actually moving (very quickly) from core to core as the work unit progresses, which causes the client to move from one cache to another cache of each core. This effect can be negated by locking the affinity of each client to run only on one cpu core, which should improve times for completion.

Currently I am using this app to lock one instance of SETI and one instance of FAH GPU2 to Core 1 and one instance of FAH CPU to Core 0, this way FAH CPU client doesn't "over power" SETI resulting in no work being performed, however the FAH GPU client gets as much processor time as it wants and SETI uses the rest.

edit: As soon as the "Fall Brawl" is over I will go back to crunching 2 cores of SETI and keep the FAH GPU client, at the time I'll do more testing to see how much faster it is to lock each instance. First impressions before installing the FAH CPU client was around 5%.

ok so I got almost everything going , 50% of a quad
100% of a quad
FX 60 split with a GPU client
plus loaded Dimes on a quad and FX60