SMP Affinity Changer

[rilian]

*hondo* A-Ch on Q6600 usually gives +10%PPD if you run two FAH SMP clients (which give 8 fahCore_a1.exe processes)

i will send your info to developer and post reply later.

could you check time per 1% on both clients for more than 2 hours with and without A-CH and post results from both FAH SMP clients?

[*hondo*]

Many thanks rilian, it's 23:14 PM here now, so if its OK with you I'll do that tomorrow.

By which time both current WUs will be finished, in fact they will complete around 01:20am

I love these SMPs, because they are so productive

Thanks again.

[rilian]

sure, don't hurry

it's almost 2 AM here

[*hondo*]

sure, don't hurry

it's almost 2 AM here

Thanks : Are you nocturnal ?

[bruce]

bruce please check your last post. A-Ch does not work on 8-core hardware so far. i guess you need to divide by 2 all core-related part

oops, I'm wrong, but not completely

Change it to read: Affinity changer seems to work best with a single copy of FAH-SMP (four copies of FahCore_a1) on 4-core hardware.

It also works best where nothing else is running. If you have more copies of FahCore_a1 running than you have physical cores, or you run other things that compete for the same cache, the benefit diminishes rapidly. For example, a quad that's heavily used as a file server or web server might actually have POORER performance in FAH with Affinity changer than without it, but if you remove the word "heavily" it's a good thing.

[*hondo*]

*hondo* A-Ch on Q6600 usually gives +10%PPD if you run two FAH SMP clients (which give 8 fahCore_a1.exe processes)

i will send your info to developer and post reply later.

could you check time per 1% on both clients for more than 2 hours with and without A-CH and post results from both FAH SMP clients?

Hi rilian I hope this gives the information you requested, can you please make sure any response is pretty simple, I only do F@H to help the cause, I don't study how FAH works and therefore lack the knowledge of you guy's on here.
By the way, I'm running FAH on seperate Hds one on C:\ the other on D:\

Thanks now *hondo*

Both Projects are from P2653
Core : SMP Gromacs
Frames : 100
Credit : 1760 X 2


Figures below in blue are without Aff / Ch installed
Both comparisons took place over a 4.75 hours, for each time period.

FAH on C Drive –
Bench marks for C:\ from Fahmon
Min. Time / Frame : 13mn 32s - 1872.71 ppd
Avg. Time / Frame : 13mn 34s - 1868.11 ppd
Cur. Time / Frame : 13mn 36s - 1863.53 ppd
R3F. Time / Frame : 13mn 36s - 1863.53 ppd
Eff.. Time / Frame : 13mn 40s - 1854.44 ppd


FAH On D Drive
Bench marks from Fahmon
Min. Time / Frame : 13mn 25s - 1888.99 ppd
Avg. Time / Frame : 13mn 28s - 1881.98 ppd
Cur. Time / Frame : 13mn 29s - 1879.65 ppd
R3F. Time / Frame : 13mn 29s - 1879.65 ppd
Eff.. Time / Frame : 13mn 40s - 1854.44 ppd

These figures below were generated while using Aff / chan
Bench marks for C:\ from Fahmon
Min. Time / Frame : 13mn 33s - 1870.41 ppd
Avg. Time / Frame : 13mn 41s - 1852.18 ppd
Cur. Time / Frame : 13mn 50s - 1832.10 ppd
R3F. Time / Frame : 13mn 50s - 1832.10 ppd
Eff . Time / Frame : 14mn 13s - 1782.70 ppd


These figures were generated when using Aff / chan
Bench marks for D:\ from Fahmon
Min. Time / Frame : 13mn 25s - 1888.99 ppd
Avg. Time / Frame : 13mn 35s - 1865.82 ppd
Cur. Time / Frame : 13mn 30s - 1877.33 ppd
R3F. Time / Frame : 13mn 30s - 1877.33 ppd
Eff. Time / Frame : 13mn 27s - 1884.31 ppd

[rilian]

talked with nik4eva today:

1) as for *hondo*'s problem, we did not find solution yet. *Hondo*, could you check if fah processes are assigned to different CPUs when A-Ch is on ?

2) as for dual quads / 8-core intel processors - we started working on it! it will be supported in next version

[Shakita]

Nice. I'm waiting for that next version eagerly so that I can try running 4 clients. Any idea when that will be ready?

[DriveEuro]

hondo...

One SMP instance on a Q6600 that runs 2 instances of SMP scores a lot lower than 3300PPD. That's what is confusing.

Personally, I'm running 3.4ghz and two P2653s on my Q6600 and I only manage 2170-2190ppd. That's why everyone is second guessing you. It would take a LOT of horsepower to score 3300ppd on each instance with a Q6600.

Understand?


(ugh, just realized I was responding to a post on the previous page... take it as you may. sorry)

[The_Barnacle]

Rilian,

First off, thanks for promoting this great tool and to the developer for creating it.

Is there any way to add a feature that allows the user to choose which SMP clients get assigned to which cores as opposed to simply being locked at 0,2 and 1,3? When running on a quad core it would be nice if the end user could experiment and find which cores are paired and put them on the same project.

The_Barnacle

[MoneyGuyBK]

The_Barnacle, Welcome to the Folding Forums

Have you tried to set Affinity manually through Task Manager?
Right click on the process and select cores.... I have been doing that with my VM machine running two SMPs on a Quad.


Peace

[rilian]

8-cpu version is going to be released very soon

[rilian]

Affinity changer now supports 2-quad systems

Check out 1.0.5 version here
Check out 64-bit version (previous 1.0.4) on the same link

http://distributed.org.ua/forum/index.p ... topic=1149

Russian/Ukrainian topics here

http://distributed.org.ua/forum/index.php?showforum=29

All your smp affinities are belongs to Nik4eva!

[compuguy1088]

Affinity changer now supports 2-quad systems

Check out 1.0.5 version here
Check out 64-bit version (previous 1.0.4) on the same link

(SNIP)

Russian/Ukrainian topics here

(SNIP)

All your smp affinities are belongs to Nik4eva!

Your site is down....getting 500 error here in the US

[Flathead74]

Works ok, here.