Hardware Canucks

Hardware Canucks (http://www.hardwarecanucks.com/forum/)
-   HardwareCanucks F@H Team (http://www.hardwarecanucks.com/forum/hardwarecanucks-f-h-team/)
-   -   SMP variants... (http://www.hardwarecanucks.com/forum/hardwarecanucks-f-h-team/47095-smp-variants.html)

BrainEater October 7, 2011 01:09 PM

SMP variants...
 
Ok , I have some questions regarding SMP. I'm going to do some testing , but I thought I'd ask here as well if anyone has tried this.

I run an Intel q9450 at stock speed.

Running mpich -advmethods in windows XP I was getting ~5000 ppd...Now , running the latest SMP console client in windows 7 (same flags) , I'm getting about 3k , a drop of 1500-2000 ppd.

I'm curious what my ppd will goto if I drop the -advmethods flag and do no frills SMP.Can you do -big without -advmethods ? Can I actually run -bigadv on a 4 core/thread to any advantage ?

Once this current unit finishes , imma drop the -advmethods flag and see what happens.

:thumb:

SugarJ October 7, 2011 01:15 PM

-bigadv is supposed to require a minimum of 6 cores visible to the OS. You can add the flag, but you'll still only get SMP units with your quad-core. The only reason the i7's can run them is because they have HT.

BrainEater October 7, 2011 01:17 PM

Yep thats what I thought....

Is it possible to 'save' a work unit ? I.E. have the exact same unit for benching ?

3.0charlie October 7, 2011 02:21 PM

There is a way, but I never managed to do it.

frontier204 October 7, 2011 03:33 PM

Quote:

Originally Posted by BrainEater (Post 555579)
Yep thats what I thought....

Is it possible to 'save' a work unit ? I.E. have the exact same unit for benching ?

Yep it's possible, although it's easier to do it for console clients:
Configure the client to ignore deadlines, i.e. answer YES to the prompt in the config:
Quote:

Ignore any deadline information (mainly useful if system clock frequently has errors) (no/yes)
That will prevent the client from timing out if you try to run the work unit after the deadline.

You'll also want to make sure the client doesn't connect to the Internet, so you don't upload multiple copies of the WU.

After reconfiguring, make sure the F@H client and core are closed, then make a backup copy of the entire F@H folder containing the work unit you want to save. For the sake of the project, you'll want to finish the work unit from the "original" copy of your F@H folder.

You can then restore copies of that backup and run them, and it will be the same WU, and it will start from the same % complete as when you made the backup. Note that you'll want to benchmark based on time per frame and not PPD, because your points count will be messed up because it will be calculated off the deadline of the work unit.

MARSTG October 7, 2011 06:43 PM

Quote:

Originally Posted by SugarJ (Post 555578)
-bigadv is supposed to require a minimum of 6 cores visible to the OS. You can add the flag, but you'll still only get SMP units with your quad-core. The only reason the i7's can run them is because they have HT.

Actually there is quite a debate there, and of course people at Stanford don't want to get the necessary support, but it has been proved that Phenom II X6 clocked sufficiently high, over 3.8 GHz, can actually finish -bigadv units in time, before the preferred deadline. Done by saving the units on core i7 systems and manually transferring them on machines with PII X6. Even with HT, the 2 virtual cores share the same FPU, so for core i7 we have 4 FPUs while PII X6 has 6 FPUs. Officially -bigadv requires at least 8 threads to be seen, not cores, i7 has only 4 real cores. That is why i was considering getting an Ubuntu installation over an X6 to make it a -bigadv producer. but given the fact Bulldozer comes with 8 real cores and 8 FPUs I shall wait patiently.
there is a bootable CD for testing, but I don't think is with -bigadv, called notfred, I red something it has like 4 types of units, one of them being AMBER.

MARSTG October 7, 2011 07:01 PM

Quote:

Originally Posted by BrainEater (Post 555574)
, running the latest SMP console client in windows 7 (same flags) , I'm getting about 3k , a drop of 1500-2000 ppd.

On that config is kinda low. Maybe you would have better success using FAH GPU tracker which is not using Version 7 that would give you more PPD. the CPU is folding alone in the system or there are other GPU clients running?

Dead Things October 7, 2011 07:59 PM

Quote:

Originally Posted by MARSTG (Post 555654)
it has been proved that Phenom II X6 clocked sufficiently high, over 3.8 GHz, can actually finish -bigadv units in time, before the preferred deadline

Gaming the system is discouraged.

MARSTG October 8, 2011 08:16 AM

is there any advantage, PPD wise, if you fold in Linux/VMware compared to Windows for the same client? let's say SMP?

krazyups October 8, 2011 10:37 AM

Absolutely ... Native Linux > VMware > Windows in terms of PPD.


All times are GMT -7. The time now is 02:46 PM.