New Gromacs core for Mac

[bousozoku]

There is a new folding@home Gromacs core that will be generally available next week.

It is enhanced for all machines and I'm sure you'll be impressed with the results, especially on G3s. Currently, it's only taking 37 percent of the time to process a work unit on my dual G4/800. 83 minutes per percent became 33. :)

Anyway, you can download it, earlier than would happen automatically, by removing FahCore_78.exe from the usual directory.

[Vlade]

Thats AWESOME, this should REALLY boost our team when everyone gets the new core. I'm going to upgrade right now, it seams like you just quit folding, delete the FAH_CORE78, and start up again.

EDIT : It has more than doubled my speed!!!

[chibianh]

so what would happen if you don't delete the 78 core now? Would it still download the new one next week?

[Rower_CPU]

How about G5 compatibility?

[MrMacMan]

Originally posted by chibianh
so what would happen if you don't delete the 78 core now? Would it still download the new one next week?
Yes, if not when you get a new unit, but most likely next week.

This has improved my production by a lot.

Anyone with a mac, get this, it kicks.

[Vlade]

I think this could easily add 4000+ points per day to our team, What do you guys think?

[bousozoku]

There is G5 compatibility, as well as much improved performance for G4s and G3s.

[csubear]

I didn't know they used gromacs for your guys folding software. I am more familar with the app from molecular dynamics(MD) . We use gromacs for MD simulations at where I work(all mac by the way). What is this new update you guys are talking about. I know that there is a mac beta that we are using, and that these is a new beta tuned for the G5. Maybe this update you guys are talking about is based of the beta? Could you guys explain this folding program?

[stoid]

I have the 65 core, will it upgrade my core to 78??

[Vlade]

Originally posted by stoid
I have the 65 core, will it upgrade my core to 78??

Yes and No. There are 2 types of units, Tinker and Gromacs. The 65 core is for tinkers (what your using now), when you finish the work unit your on, you will most likely get a new Gromacs core (78). You have no choice but to finish your tinker unit, and your next unit should be about 3-4 times as fast if its gromacs.

[bousozoku]

Originally posted by csubear
I didn't know they used gromacs for your guys folding software. I am more familar with the app from molecular dynamics(MD) . We use gromacs for MD simulations at where I work(all mac by the way). What is this new update you guys are talking about. I know that there is a mac beta that we are using, and that these is a new beta tuned for the G5. Maybe this update you guys are talking about is based of the beta? Could you guys explain this folding program?

Stanford University has a distributed application which attempts to simulate protein folding. They have been working with Gromacs for nearly two years and have been using it to speed the simulations. I doubt that they've used the beta of which you speak, but of course, it's possible. It's more likely given the timeframe that they've simply taken the new wrapper code which utilises Gromacs and re-compiled the whole thing with gcc 3.3 for G5 compatibility.

[csubear]

you are are most likly right about the beta . I am sure that they have just recomplied it with gcc 3.3 . It is a shame though. We are using the gromacs 3.1.5 beta for a while now and it is very nice. All the inner loops are compiled with altivec code, and that makes gromacs scream on our G4's. I relly want to see what it will do with a G5.

[Powerbook G5]

I deleted 65 but instead of getting 78, it just re-downloaded 65 again.

[bousozoku]

Originally posted by Powerbook G5
I deleted 65 but instead of getting 78, it just re-downloaded 65 again.

Yes, of course. Tinker work units use FahCore_65.exe, not FahCore_78.exe, which is reserved for Gromacs work units.

When you receive your next Gromacs work unit, it will most likely download a new version of FahCore_78.exe for your folding pleasure.

[Tiauguinho]

Ill test this new Protein on my Dual G5 and will make a report of the performance of the beast compared to my Dual G4 1GHz.

[nw_mike]

According to discussions over at http://forum.folding-community.org/ They recommended not using Gromacs/(Altivec) enhanced core. So can we now?

Also, what determines what cores get downloaded? Right now my G5 is running 2 x FahCore_65.exe?

TIA

G5

[mc68k]

Originally posted by nw_mike
Also, what determines what cores get downloaded? Right now my G5 is running 2 x FahCore_65.exe?a protein is downloaded that best matches your processor type and speed.

according to the protein (http://folding.stanford.edu/psummary.html) that is chosen, you get a certain core.

TINKER = core65
GROMACS = core78

[bousozoku]

Originally posted by nw_mike
According to discussions over at http://forum.folding-community.org/ They recommended not using Gromacs/(Altivec) enhanced core. So can we now?
...
TIA

G5

It wasn't a recommendation on the G5, it was an impossibility to complete a Gromacs work unit using a G5. Now, it's possible.

[nw_mike]

Thanks to bousozoku & mc68k

[bousozoku]

Now that I've completed 3 Gromacs WUs with the new core being at least partially responsible for quick turnaround, they've given me Tinker WUs all round.

Is this a Wintel conspiracy? :D

[MrMacMan]

Indeed it is!

Why did I also get a Tinker unit?

And WHY IS IS PROCESSING SO SLLLOOOWWW..

I aprox 80% slower then normal!

:eek:

[Powerbook G5]

I always get stuck with the tinker WUs, it's really annoying. It just sucks that Folding is sooooo slow on Macs, I wish they had a better client for OS X.

[bousozoku]

Originally posted by Powerbook G5
I always get stuck with the tinker WUs, it's really annoying. It just sucks that Folding is sooooo slow on Macs, I wish they had a better client for OS X.

Well, the Tinker core could have been written better, but it's mostly because the G3 and G4 stink at floating point numbers, that Tinker WUs go so slowly. :(

IBM does not go for that and clock for clock, the latest G3 processors have better floating point capabilities than the G4.

When you get a Gromacs WU, you'll see how much better that core is. :)

[MrMacMan]

Indeed either kill off the tinker units all together or update the tinker core.

It really sucks... :o

[bousozoku]

I have asked them repeatedly to re-compile the Tinker core with newer tools. I didn't ask them to change it in any other way.

I'm not sure whether they used a FORTRAN compiler, of which there are few on Mac OS X, or converted the code to C. If it's the latter method, they could re-compile it with the latest C compiler and find better efficiency. It should be worth a couple of hours work.

[Makosuke]

I'm rather new to the whole F@H thing, but I've been trying to read up on it and I think I'm getting the general idea now. Since it would seem that the new Gromacs core is available at this point, I went ahead and used mc68k's spiffy script to install F@H on my dual G5, and I've got it grinding away with what appears to be the new gromacs core (at least I didn't get a tinker unit).

At this point, each processor seems to be taking about 9 minutes or so per frame (that is what they're called, isn't it?) of a 100 frame, 51 point protein (p352_nat_305_99)... which made me wonder how much better than the old core that would be (and how it compares to Jaguar, which I'm not running).

Does it work to calculate minutes/point for comparative purposes, or are the points not consistent enough for that to be meaningful? If so, what are some before and after numbers of people who've been running for a while? Guess that'd make mine around 18 minutes/point (per processor), if that's at all meaningful.

Sorry in advance if I'm spewing meaningless drivel--just ignore the newbie.

[Plutronics]

You Make Perfect sense to me !!
The old Gromacs core wouldn't even run on the G5, but the new core on the G4s are running 230 % to 270 % faster !! Even My G3s are running 230 % better !
Panther seems to speed a lot of things up on the G5s but Folding only a Tad.
Welcome to the Team !!
Wooo Hooo !!!! :D :D

[Plutronics]

If you use the -advmethods flag , I think you'll get more Gromacs WU. Or maybe I'm Lucky ? I've got NINE gromacs and not a single (S)tinker. :D :D
I Use :
./OSX-3.25 -local -pause -advmethods -verbosity 9

[bousozoku]

I just noticed something after my G3 returned results. It got a Gromacs work unit, but did not automatically download the new core. I removed the old FahCore_78.exe and it, of course, downloaded the new version then.

You may want to verify that you're running the new one with any Gromacs WU.

[stoid]

I haven't had a Gromacs core ever, in all my WUs, and I must say I'm disappointed. I just ran some tests and found out the AltiVec enhanced my laptop will scream at nearly 5.3 GigaFlops however, while non-AltiVec, the poor guy is demoted to a lowly .63 GigaFlops. Needless to say, I would love to get one that uses a Gromacs core to get the nearly 10x boost in speed.

BTW if my calculations are correct, the BigMac's current flop rate would put it at around 30 seconds to complete a 70 point WU and it could return about 214,000 points a day! Even then though it would take almost 150 days for it to fold us into first place...:D

[Powerbook G5]

I want a 1100 node G5 cluster...my PowerBook seems to turn out a percentage point every 25-30 minutes, which ends up taking forever to complete a WU.

[stoid]

Is that one percent, or one frame? When I realized that my 70 point WU was doing one frame every 20 minutes instead of one percent, I got even more let down.

edit: Uhhhh, didn't mean to burst anyone's bubble there. Sorry if I did.

[Powerbook G5]

As in one percent of the whole WU. I'm not sure how to check how many frames it does. The graphical client just shows percentage until the WU is complete and generally it takes about 25 minutes or so to go up a percent.

[stoid]

It is the console unit that shows frames.

i.e.

[17:38:23] Finished a frame (126)
[18:36:05] Finished a frame (127)


I have QT Pro doing a large video file conversion at the same time, but usually it takes only 20 minutes per frame.

[Bob Knob]

[00:14:03] Completed 1675000 out of 2500000 steps (67)
[00:20:55] Completed 1700000 out of 2500000 steps (68)

I had a few other apps going during this, but I seem to average 01% every six to seven minutes on each processor.
BTW - Dual 2GHz G5 only 1GB of ram till the end of the month.

[MacBandit]

So how do I know I have the new core. I'm guessing it's still caries the name FahCore_78.exe

[bousozoku]

Originally posted by MacBandit
So how do I know I have the new core. I'm guessing it's still caries the name FahCore_78.exe

In the console client, after the line "Folding@home Gromacs Core", there's another line which says "Version 1.53" or "Version 1.48" or whatever.

[MacBandit]

Originally posted by bousozoku
In the console client, after the line "Folding@home Gromacs Core", there's another line which says "Version 1.53" or "Version 1.48" or whatever.

How do I get those lines?

The only commands I know for the console client are.

start
stop
pause
resume
work

Never mind I just found that I have the new client. I just needed to look at the log on in the F@H folders

[MrMacMan]

Yeah to restate.

If you have the Console (or mc68k/bous script) you need to trash the FahCore_78.exe while your current unit has stopped or paused.
After doing this Folding will download a new FahCore_78.exe this is better optomized and you will fold faster.


Hope that clears that up.

BTW, the folder is located in your hope directory, aka:
~/F@H (or)
~/F@H1

[Plutronics]

The old '78 Core is 3 Megabytes
The new Core is 2.3 MB with a 10/31/03 Modified Date
:D :D

[MacBandit]

Originally posted by Plutronics
The old '78 Core is 3 Megabytes
The new Core is 2.3 MB with a 10/31/03 Modified Date
:D :D

I noticed that but the mod and creation date for mine is today the day I downloaded it.

I didn't know what to think of the size difference. I kind of wondered if it increased over time due to archiving history or something.

[Vlade]

Originally posted by MacBandit
I didn't know what to think of the size difference. I kind of wondered if it increased over time due to archiving history or something.

Small code is fast code, so thats probably the reason that it shrunk so much.

[MacBandit]

Originally posted by Vlade
Small code is fast code, so thats probably the reason that it shrunk so much.

Agreed. Forget my original comment it was just stupid. I was thinking that maybe the program stored some sort of data within itself and grew in size over time but then the sizes of everyones F@H.exe file would be different which isn't the case.

[MrMacMan]

Originally posted by Vlade
Small code is fast code, so thats probably the reason that it shrunk so much.

Indeed, even if any code wasn't rewritten if it gets smaller, it gets faster.

:)

[MacBandit]

Well it is definitely faster. I had two units working one on each processor. I had at least 3 24 hour days of work into each of them to get to 66%. After installing the new client they both finised in less than 8 hours.

One of my latest projects is a 51 pointer and in an hour and fifteen minutes I have completed 6%. That's a big increase over what I was getting before. That means I'll complete a 51 point unit in under 21 hours. That's about 1/3 the time it use to take.

Now I just have to go update my commandeered iMac and get 2 or 3 G5s and maybe a Powerbook or two working for me.

[Golem]

Finally I get gromacs and see what my g5 can really fold like. I had 9 or 10 70pt tinkers in a row at approx 2 a week processed. Never a gromacs to be seen. Sudddenly I am getting 50 pt gromacs and I have processed 5 in the last 3 days. It has put my production through the roof from 300-350 average a week to 700 since I last checked it 4 days ago.

[MacBandit]

Originally posted by Golem
Finally I get gromacs and see what my g5 can really fold like. I had 9 or 10 70pt tinkers in a row at approx 2 a week processed. Never a gromacs to be seen. Sudddenly I am getting 50 pt gromacs and I have processed 5 in the last 3 days. It has put my production through the roof from 300-350 average a week to 700 since I last checked it 4 days ago.

What's your machine?

My Dual/1.42 turns out a 60point gromacs at about 8-9 hours/cpu.

I can get about 1100 points a week just off my Dual/1.42.

[Golem]

Originally posted by MacBandit
What's your machine?


1.6 g5

[MacBandit]

Originally posted by Golem
1.6 g5

That's pretty good then. That's a 27% increase in output over just one of my 1.42GHz processors with only a 13% increase in processor MHz.

[MrMacMan]

It certainly has improved a heck of a lot.

My units were taking 4-5 days... now 2-3.

Really good.

[bousozoku]

Improvement is an understatement. With two machines and three processors prior to Gromacs, I was seeing a combined 16 something points per day. A few days back, I saw a high of 86 something points per day and I seem to be averaging about 55.