new Folding@Home 4.0 final

[csimon]

You can get the new Fah4 client here (http://folding.stanford.edu/download.html)!

This is the final release non-beta or prerelease.

Thanks to Guha!!! :fold:

# Windows Console Edition ########################
##########################################

Folding@home Client Version 4.00

http://folding.stanford.edu

##########################################
##########################################

[GHoosdum]

Is it actually faster as you guys were discussing earlier?

[csimon]

Not sure if it faster ...but it certainly seems faster. I can't tell because I haven't run the 3.25 beta since the 4.0 betas came out but maybe you will test and see?
The one thing that this final client has over the older prerelease is that the flag -forceSSE is no longer case sensitive!
so your amd will now accept -forcesse.

[Thrax]

What will it list in the log when SSE is successfully forced on?

[GHoosdum]

What switches need to be run on this one? No -forceasm any more? -forcesse -advmethods (?)

And does it use the SSE2 extensions or not?

[GHoosdum]

OK - after testing 4.0 final vs. 3.25 beta - they are EXACTLY the same speed for completing frames on a 51.4 point gromac unit.

(Same WU, I just shut down and installed the 4.0 instead of 3.25 then ran it through a few frames before checking frame time again)

//edit: Thrax, it still says "Extra SSE boost OK." when SSE is turned on on mine.

[Straight_Man]

Well, what I got with 3 was that end frames took longer than first, on two boxes and two O\Ss that are radically different. With 4.0, the first percent took much longer than after it had been running 15% through a WU.

Second, I ended up using -forcesse -forceasm -advmethods on both the Barton and the Intel P4 boxes for best overall performance with 4.0.

GHoosdum, your Barton box is a better candidate for 4.0 than the XP2100+. Yes, 4.0 can use some of the SSE2 extensions, though I doubt it uses all by default as if it DID, the forceSSE would hav more bang on the P4 speed than it did. BUT, it checks also to see if it needs to send more often than the version 3 client did, and it does indeed truely try to send if there are completed WUs in work queue-- Vijay said it was set to basicly do a send attempt about every 1\3 WU if there were completed WUs in queue. When my router gets here, will try again and see if can send better, and just let the client handle resends if it logs them for me in final. I guess Monday at latest will be cycling WUs again, possibly late Monday.

Note that forcesse and forceasm are off by default, also, in the graphical client, and this is another clue to me that some SSE2 is being used as the processors that were having issues that I know of do not have a full SSE2 set builtin.

Overall, I was doing WUs of 1000 series in 3\4 the time of older Gromacs, and getting more of the newer ones than the old ones. Expect the new client\core to show percentage on new WU projects, not frames. They are 2.5 Million step workunits, so they do 25,000 steps per percentage point (1/100th of a WU instead of 1/400th for a 400 frame WU where reports are in frames) and they also do more per percentage than what was spoken of as frames in same time frame. I was turning in new Gromacs projects FASTER also, and the new Core_78 subversion picked up the pace some on these WUs when used with the new Pres for 4..0 even, maybe 5-10% improvement, think the CORE is using more SSE2 and SSE if available from client overall config and benchmarking and SSE validation.

So, to get a real speed test, do maybe 15-20% of a WU, take overall time versus similar project on older client from past logs. BUT, would test on Barton first, and expect less improvement, though some, on the XP2100+ than on the Barton. Figure on newer WUs that first percent will probably take LONGER(maybe 2 min longer or 18% longer, looks like the thing is self-testing some as it calcs, I had 12-13 min for first percentage, more like under 10 about 1/4 way into new WU series of 1000 numbering, and last percentage was very close to the one 1/4 of the way in) than one 25% of the way in, and that percent 100 should be about same time to 15 seconds variance from the one at 1/4 done WU.

Does that help explain some???

John.

[Black Hawk]

Ok wtf?! I was using v3.24 with -advmethods -forceasm -local -service and I got the SSE thing but now with v4 I get 3DNow. Do I need a diffrent switch or do I have to take out one?

[GHoosdum]

-forcesse

[Black Hawk]

so I switch forceasm for forcesse?

yeah...

[GHoosdum]

John... yes, that info does help. Sorry I didn't see your post at first, since "go to first unread" went straight to Black Hawk's post under it.

The box I was checking it on at first is the P4 box at work (hence the SSE2 question) - I haven't installed the 4.0 client on either my Barton or my TBred boxes at home yet, but I will.

On a sidenote, I'll get the new 92MM 60 CFM fan for my Primary rig tomorrow, so I can OC it again! Woohoo!

[mmonnin]

Uhhh blachawk, you got sse the first time and never after. You cant get SSE all the time on AMDs with 3.24. At least after a core update SSE was disabled.

[Thrax]

Nevermind. I see where it says SSE boost enabled in the log.

[csimon]

on my amd rig I use -advmethods -forcesse
on my p4 & xeons I use -advmethods

that way on all I get the extra SSE boost ok@

[Black Hawk]

mmonnin had this to say
Uhhh blachawk, you got sse the first time and never after. You cant get SSE all the time on AMDs with 3.24. At least after a core update SSE was disabled. Checking my logs, the last time I started up v3.24 I got SSE.

[GHoosdum]

Yes - every time it starts with 3.24 it runs with SSE. Then, each subsequent WU it processes until your next startup, runs with 3dNow.

[Black Hawk]

GHoosdum had this to say
Yes - every time it starts with 3.24 it runs with SSE. Then, each subsequent WU it processes until your next startup, runs with 3dNow. Nope. All have gone with SSE.

[csimon]

that only happens on amd procs.

[mmonnin]

Maybe its something with a new core because with 3.24 -forceasm would not get you SSE. Its not supposed to happen. I used it and it only got 3dnow! so I had to upgrade.