DPAD - Multi core rigs & getting the best output

[Assimilator1]

I don't know how many of you folks watch the benchmarking thread over on the DPAD forums, but it's come to light that the DPAD client is very inefficient at multi threading.

Tests by [DPC] Eclipse~NaWA (thanks for the info)
His rig is a dual socket AMD with dual core Opterons (lucky git)

Check out his graphs & tables here ,thiers also more info in the thread I linked at the top.
Going from 1 client auto threading to 4 clients single threaded resulted in a 35% boost in output!:Q

Oh & I'm no longer un-employed, got a new job sorted now so I can order up my Core2 duo bits

 

[JonB]

I'll check that out. As a recent test I set the dual clients on my dual core Opty to run 4 threads (instead of auto). I was looking to improve output quality, as demonstrated by the Viewresults program from DPAD. I've noticed that my AMD based clients do fewer units but higher Mpts per unit. I haven't figured a reason.

congrats on the job.

 

[Assimilator1]

Interesting ,what do you mean by 'improve output quality' ?

Also I've noticed that my AMD based clients do fewer units but higher Mpts per unit.
Are you saying the AMD units are slower but get a better score for their work?

 

[JonB]

No, I'm not saying they are slower, but they do do fewer work units. They not only seem to be faster (more Mpts) at DPAD (at least the 64 bit processors) but my undocumented viewing of Results.txt files using Viewresults shows that my AMD clients consistently have a higher cluster of results to the right side of the graph. That is the "good" side of the graph. My Dell Intel Core Duo boxes are heavily shifted to the left side of the graph (lots of negative results and low Mpts per unit).

 

[Amaroque]

My X2's are faster than my C2D clock for clock by ~20%

 

[Assimilator1]

Damn! that's wierd ,& I've just bought a C2D! ,I hope Stephen sorts DPAD out so that C2Ds do better.Have you mentioned that to him?

 

[Assimilator1]

Well I've got my C2D upgrade up & running now

I went to run 2 clients of DPAD earlier ,it was then I realised I didn't know how to set CPU affinity for each client! ,lol ,how do you do that?

 

[Amaroque]

C2D is faster in 80% of everything else though... 1 GHz OCs are pretty easy (most of the time) too!

 

[Amaroque]

Set threads to auto in the config.txt. DPAD will use any, and all cores available then.

 

[Assimilator1]

Err no,did you not read my 1st post? heh,setting 1 client to auto is the slowest way to run DPAD ,that's why I need to know how to set CPU affinity

And yea I'm hoping for at least 3GHz

 

[Amaroque]

The only way I know how to change affinity with regards to DPAD, would be via task manager.

 

[Assimilator1]

Their's a way to do it via the commandline ,I just can't remember exactly how
Thanks anyway

 

[Darkone]

Assbatterer1,

Make 2 folders, each with the contents of the dpad zip in..

In the first directory, open the config file, set threads to 1

in the second directory, open the config file and set threads to auto

will run 2 instances on eon each core then.

------

 

[Assimilator1]

Dankone
Ok thx bud

Actually that's more or less what I did in the end,except on the 2nd client I set the threads to 1 too.
That shouldn't matter should it?

[edit]Stephen brooks has answered me in the DPAD forums & that is in fact what he suggested to do

 

[Amaroque]

:laugh: rofl, you guys are funny

 

[Assimilator1]

Can't think of good one for your name

Btw just cranked up my C2D to 2.5GHz ,on a single core basis it seems my XPM @2.5GHz was a little faster at DPAD!:Q
Early results show I'm getting about 224 Kpts/s vs 234Kpts/s for the XPM!

 

[Assimilator1]

A little bump for those that missed it

 

[VirtualLarry]

Originally posted by: JonB
No, I'm not saying they are slower, but they do do fewer work units. They not only seem to be faster (more Mpts) at DPAD (at least the 64 bit processors) but my undocumented viewing of Results.txt files using Viewresults shows that my AMD clients consistently have a higher cluster of results to the right side of the graph. That is the "good" side of the graph. My Dell Intel Core Duo boxes are heavily shifted to the left side of the graph (lots of negative results and low Mpts per unit).

That's interesting. It suggests a CPU-dependent bias in the results, which should not be so. Are these machines overclocked? Are they stable? I'm just wondering if there is a qualitative difference in the output that is dependent on the CPU. I wonder if there is some as-yet-undiscovered CPU bug that these results might indicate.

 

[VirtualLarry]

Originally posted by: Assimilator1


Can't think of good one for your name

Btw just cranked up my C2D to 2.5GHz ,on a single core basis it seems my XPM @2.5GHz was a little faster at DPAD!:Q
Early results show I'm getting about 224 Kpts/s vs 234Kpts/s for the XPM!

That's interesting too. I would not have guessed that clock-for-clock an older XP-M CPU would compete with a C2D.

 

[Assimilator1]

Yea their's something odd going on with DPAD & C2Ds ,I'm pretty sure I asked Stephen Brooks about it ,but I've got no replie yet....... I don't think anyway ,I've just remembered that DPADs forum doesn't have email notification:roll:

 

[Orange Kid]

I don't know if the DPAD client will accept switches, but the ones I used to use for seti was -cpu 0 and -cpu 1.

 

[Assimilator1]

Yea that's the sort of thing I was thinking of originally ,but I got around it anyhow
Though I'm not 100% sure that each client is dedicated to 1 core!
I did find a program specifically made to set CPU affinity, but I ended up not using it.The websites gone now too ,but if anyone wants it I d/led it.