Windows SMP Beta goes Open

Darryn · 18 Mar 2007 at 00:49

Here it is guys...

All you multi core Windows users need this.

SiriusB · 18 Mar 2007 at 00:56

Does this mean the Linux SMP Guide isn't needed any more?

Darryn · 18 Mar 2007 at 01:06

I dont know if the Linux SMP client is any more or less efficient than the Windows SMP client.

From my experience the Linux OS itself is more efficient than Windows, so Id imagine the Linux client would get slightly more PPD due to less system tasks being active in the background.

Mattus · 18 Mar 2007 at 01:18

Excellent, been waiting for this!

SiriusB said:
Does this mean the Linux SMP Guide isn't needed any more?

Nope, it just means we need a Windows one as well

Joe42 · 18 Mar 2007 at 02:37

I thought this was miles off...

Does it still have the really tight deadlines so machines have to be on 24/7 to run it?

Btw if anyone wants a page on my folding website for windows/linux smp guides they're welcome...

Steevo38 · 18 Mar 2007 at 07:41

C2D machine switched to WinSMP

Looks like only one client running is the way to go as seeing 100% cpu usage.

Anyone got any idea how many points a 2651 is worth? At 15 mins per frame they had better be worth some serious pointage to be better value than the Linux SMP wu's.......... :eek:

Fold on team

Steevo

babyface uk · 18 Mar 2007 at 08:12

super, lets see what the ol quad core can do

Babyface UK

Snapshot · 18 Mar 2007 at 08:24

I've just loaded it on my Pentium D 820-based PC. It's running absolutely flat out, using ~256MB RAM and taking 30 min per frame. I need a better HSF and exhaust fan in that case! I'll try it on the C2Ds when they finish their current Linux WUs.

oceaness · 18 Mar 2007 at 08:24

How are the deadlines?

I have an AMD X2 3800 @ 2.5 GHz and an E6300 @ stock (waiting for better case and CPU cooler - I can't be bothered to go through overclocking it twice) which will be upgrading to SMP at approximately 4:30pm

I also have a Northwood, I know HT isn't true dual core but will it work/be fast enough.

On with the Fold!

jaric · 18 Mar 2007 at 09:11

According to http://forum.folding-community.org/ftopic18534.html
kasson says
Quote:
Currently we have dedicated win SMP work units. Three projects are being assigned at this time:
Project 2610: 1523 points, deadline 4 days
Project 2651: 1760 points, deadline 4 days
Project 2652: 1148 points, deadline 3 days

Particularly as this is a beta client, we reserve the possibility to adjust point values and/or deadlines in the future.

Steevo38 · 18 Mar 2007 at 09:16

Answers my question. Cheers.......

FreeStream · 18 Mar 2007 at 10:35

Any ideas on PPD?

Im doing 1.9k with a 3.456Ghz Conroe in linux

Steevo38 · 18 Mar 2007 at 10:44

Now getting 13.5 mins per frame on my E6300 @3.15
So roughly 1870ppd here.

diogenese · 18 Mar 2007 at 10:46

How do you close the thing down without it having a paddy when you restart it and get "improper shutdown" messages?

joeyjojo · 18 Mar 2007 at 10:47

Steevo38 said:
Now getting 13.5 mins per frame on my E6300 @3.15
So roughly 1870ppd here.

Might as well install this. Just started another 1760 pointer though, will have to wait another 32 hours for it to finish.

2bullish · 18 Mar 2007 at 11:05

I take it this client runs at 100% cpu usage, therefore, running this and the gpu client is redundant?

Steve_O · 18 Mar 2007 at 11:08

diogenese said:
How do you close the thing down without it having a paddy when you restart it and get "improper shutdown" messages?

Fantastic to see that's still an issue :rolleyes:

The cores can take a while to stop - best to check Task Manager and see what's going on there before doing anything else

edit:

19) Unless you plan to reboot, when you stop the client with CTRL-C be sure that the processes FahCore_a1.exe and mpiexec and/or smpd have stopped. Many folks are finding that they need to kill mpiexec manually and or stop/start the smpd service. If you restart the client without doing this, the WU may be deleted. A common message is MISSING_WORK_FILES.

From the known bugs list - linky

2bullish said:
I take it this client runs at 100% cpu usage, therefore, running this and the gpu client is redundant?

on a quad core it's alright - on a dual core it's not a good idea at all

(ooops at home for the weekend and didn't realised my bro's machine was logged in as him

- rich99)

diogenese · 18 Mar 2007 at 11:13

I've just found on the Stanford site that they recommend the use of the -forceasm flag.

The SMP core can get confused about disabling SSE, so we suggest running with the -forceasm flag if you notice that the SSE was disabled unnecessarily.

edit: and just noticed this too :rolleyes:

6) The SMP core can get confused about disabling SSE, so we suggest running with the -forceasm flag if you notice that the SSE was disabled unnecessarily. In fact, SSE will re-enable itself even though the messages say that isn't happening, so -forceasm is no longer required.

Its all running nice on 5-20% of my e6600 now, just got to wait a few hours for the linux one to finish on -oneunit to see it's full potential

rich99million · 18 Mar 2007 at 11:21

diogenese said:
I've just found on the Stanford site that they recommend the use of the -forceasm flag.

In the updated Known Bugs list they now say it's not necessary to use -forceasm, it can't hurt though

http://forum.folding-community.org/ftopic18210.html

6) The SMP core can get confused about disabling SSE, so we suggest running with the -forceasm flag if you notice that the SSE was disabled unnecessarily. In fact, SSE will re-enable itself even though the messages say that isn't happening, so -forceasm is no longer required.

In terms of successfully shutting down the client without losing work it certainly can be tricky - given the relatively poor performance of the SMP client on my X2 I'm sticking with my GPU client for now - not being able to reboot the machine for more than a day is not my idea of a stable client

FreeStream · 18 Mar 2007 at 11:54

Its running!!!

2.3k PPD at 10:59 min frame time on a 1760 pointer