smp setup guide

[MikeB12]

I know this info is dated and most of us know how to set this up...
but from another thread, I think some of the new gpu2 folders just now joining may have been left out of the smp instructions last year.. and we all know it can be a pita to setup the first time.

so I'm just posting this guide I did a while back, to help the new guys out, if they are considering running a smp cpu client with the gpu 2 client.

this is only for the quads, btw.. once you get this setup, make sure you go into settings on your gpu2 client and check off the "slightly higher core priority"... otherwise this smp cpu client may choke your gpu2 ppd... the gpu2 needs some affinity priority so it can jump in the front of the line for cpu requests.
************************************************** *******

************************************************** *******

************************************************** *******

************************************************** *******

************************************************** *******

anyway here's the guide for setting up SMP...

dl the client and run setup.exe to install the client.
http://www.stanford.edu/%7Ekasson/folding/Folding@Home%20Windows%20SMP%20Client.EXE
You can use the default loc for XP,
but in VISTA or WINDOWS SERVER 2003 don't choose the default.
choose C:\fold or some other place besides C:\Program Files.


************************************************** ********************************

The software install is done at this point, now you need to make it work.

On xp make sure you have net framework 2 installed, if not, run windows update and grab it from optional updates.
In Vista, net framework 2 is native.

Create a windows user, make them administrator, create a password (no blanks). must be admin, and must have a password.
http://img.photobucket.com/albums/v492/B12MHudson/1-12.jpg

************************************************** *********************************

Open your install location (in my example xp, I installed in C:\Program Files\Folding
the 2 highlighted files are the ones you need to be concerned with.
REMEMBER: in VISTA\WINDOWS SERVER 2003 do not install in C:\Program Files..
http://img.photobucket.com/albums/v492/B12MHudson/2-5.jpg

************************************************** *********************************

(this next action is mpi, and is required every time you intend to start the client. you must get "mpi is working" tyo proceed to next step)
-Run install.bat
-type the username you created above and hit ENTER.
-type the password you created for the username and hit ENTER.
-retype the password to confrim and hit ENTER.
-you should get "mpi is working" as a message.
-Hit ENTER to proceed and the window will close.

(the first time you run install.bat it may ask to you to allow traffic, say yes if it does)
http://img.photobucket.com/albums/v492/B12MHudson/3-3.jpg

************************************************** *********************************

(this next action starts the client)
Click/run fah.exe.
The first time you run this it will ask you to config. input your username and team#.
The most important thing you can do here is answer NO to everything except "accept large work assignments=YES"..
http://img393.imageshack.us/img393/7748/26ye0.jpg

************************************************** *********************************

A successful launch will make it all the way to where it displays "SSE Boost OK" "Completed 0 percent". Any errors or ends and something is amiss, If you get an error don't get frustrated. (see note **). Here's what it looks like launching error free.
http://img296.imageshack.us/img296/6826/114ar2.jpg

************************************************** *********************************

Once this client is up and running you should see 4 threads of fahcore_a1...
http://img106.imageshack.us/img106/2458/51re3.jpg

**NOTE
additional faq resources:
stanford smp faq http://folding.stanford.edu/FAQ-SMP.html


wiki smp faq http://fahwiki.net/index.php/How_do_I_install_the_SMP_Console_client_in_Windows _Vista%3F

[Marvin_The_Martian]

You're forgetting something, enter adv methods and enable ' always opt for scientific cores ' aka the -advmethods switch since most smp wu's will be scientific cores. Also I like your pictures but why not do it right and put in our team number :shock:

Btw, isn't this indeed a very rusty old repost? I would have been impressed if you put in some AC stuff, some estimates about production per hw type ( x2's vs c2d's / phenom vs c2q ect ) and so on.

[MikeB12]

it's just a simple guide I posted for Synthetics thread... and yes it's quite old. a copy/paste from one I did last year...

I doubt the AC dual smp stuff is needed anymore with all the gpu2 points running. I'm still running 3 quads on dual smps, but that's just because I don't have any decent gpu2's to add. as soon as I add a gpu2, I'll drop back to one smp and one gpu2....

We don't need the adv method switch on smp, they still get the same wu's. mostly 2665 for windows now.. the last client to utilize adv methods was the V6.10beta single core client.

so I'm just keeping it simple for the new gpu2 guys on quads, that want to run a cpu client with gpu2, and don't know how to setup smp. this makes it easier than weeding through the fah faq.

I think we've run the gambit on AC and dual smps, it's becoming old news now.. gpu2 is replacing it.. let's just keep this a simple one for the new gpu2 guys that don't have any smp setup experience.

Like I said, I mainly copy/pasted here for synthetic'spost about running gpu2 and cpu clients together. Which cpu client and how on a quad..

[Marvin_The_Martian]

Well I remember when Vijah posted telling us to use the ' -always use advanced methods' option, AC can still help ppd for allot of people who don't have a gpu2 client to run but I'm starting to nitpick on your guide and I feel out of place. It's a nice idea and touch, but for it to be really usefull it needs work in my opinion. Luckily, my opinion is just my own and I think some people might have some profit from this.

http://www.xtremesystems.org/forums/showthread.php?t=178597

Btw did you read that, followed the links?

If you wanted a guide for smp plus gpu2, write such a guide. What you posted is not, it's just smp. Now, where is that question so I can try and help since I am running that already :)

:up:

I really do like the enthousiasm and I would love for you to take this as it is ment, constructive criticism, and nothing more. You're a fellow folder, and as such, I already liked you before we both were born :D

[[XC] Synthetickiller]

I know I posted the question that you are responding to, so thank you for the help. I do need a little help though.

After clicking install.bat and going through the name, password, etc.... I encounter this error.
http://i31.tinypic.com/2csci91.jpg

I am in Vista x64. Any ideas?

[Marvin_The_Martian]

UAC off? Did you run it as admin? Did your firewall pop up ( it should! ) and ask new permissions?

Sounds like a network failure to me.

[[XC] Synthetickiller]

Sorry guys, i just did the wiki, turned off the UAC and it works and I am now setting it up. If anyone reads this, in vista, you must shut off UAC.

[MikeB12]

that's cool, I only posted it for standard smp setup. the xs faq only has a service setup guide.

no big deal, if it's not needed or wanted, just let it slide away out of sight....

Kevin already did a gpu2 guide...

like I already said, it's old news for most of us. but for a few new gpu2 guys, they don't know which cpu client to run, or how to set it up. all they need is the smp setup and the core setting in gpu2.

if it's not used by anyone, then let it disappear. no harm, no foul..


edit: glad you got it figured out Synthetic...

[Marvin_The_Martian]

Lol don't worry, you do know ' RTFM' isn't only the most used answer, it's most often the right one :up:

The setup thread UnG put up holds all the info you need, though nothing yet if I'm right about the gpu2 client.

[[XC] Synthetickiller]

I got it working. Now I'm just waiting for FahMon to report progress and PPD so I know how long my quad will have to work. My cpu usage is at 100%, and I don't see much of a hit if any to the GPU client.

I would like to mention, although I am getting 100% CPU usage, the task manager reports very low usage (spikes to 8%) for fahcore_11.exe. My temps prove this as well. Very strange.

Shouldn't it take about a day to crank out a WU w/ a q6600 at 3.6ghz?

[MikeB12]

you'll finish in much less than a day on one smp client. I'm running 2 and it takes about a day on q6600@3.4ghz... just wait til fahmon gets a couple frames in and you'll have your answer ppd and eta... I assume the fahcore_11.exe spikes you're seeing are the gpu2 cpu requests.. which is good. 25% would all of one core... and you're only at 8%, so gpu2 is only needing about a 1/3 of that core to use. nice... the rest is being used for smp...

post back and let us know what your machine ppd tally is... (gpu2 and smp)

[Marvin_The_Martian]

I got it working. Now I'm just waiting for FahMon to report progress and PPD so I know how long my quad will have to work. My cpu usage is at 100%, and I don't see much of a hit if any to the GPU client.

I would like to mention, although I am getting 100% CPU usage, the task manager reports very low usage (spikes to 8%) for fahcore_11.exe. My temps prove this as well. Very strange.

Shouldn't it take about a day to crank out a WU w/ a q6600 at 3.6ghz?

Like Mike said looks like you're fine. The cpu usage is what I expected with your card, maybe even lower and that's only good. If you put fahmon on 'last frame only' it will give you rough ppd within the first smp frame ( I think less then 10m at those clocks, I can't break 3.3g steady and run 3g 24/7 because I don't like pushing more and more volts through my nb to get the fsb stable. I wouldn't mind a simple vtt+ mod for p5w dh, since it's that voltage and vFsb which is in need of tweaking. ).

You should be breaking 9K total with that setup easy from what I seen.

[[XC] Synthetickiller]

MikeB12,

I'm getting 7694.75ppd. The GPU client is fast as we all know. The SMP client takes 23 hrs to finish a WU, about 2077ppd. Not too bad, but not what I was hoping for. I have another client on a single core 3800+ doing only 88ppd.

I don't have another quad or spare intel board, but if I dropped 2.7ghz X2 w/ an 8500gt 512, would that be a decent machine? A 5200+ brisbane is 77 bucks, 8500gt 512mb gddr3 is 63 bucks. If the GPU client final version is not CPU dependent on xp, i'd rather spend 130 on a GPU like the zotac 8800gts for 120 after rebate. That would put out more WUs and PPD in general.

[Marvin_The_Martian]

You're only asking MikeB12? Ok I'll shut up then :)

*whistles and walks away*

[Kingcarcas]

It's a PITA, even more so if you're using Vista. My X-Fi control panel was off, ATI display driver would always reload, no vsync if a game doesn't have the option, lower PPD, i'm back with XP SP3 :shrug:

[Marvin_The_Martian]

It's a PITA, even more so if you're using Vista. My X-Fi control panel was off, ATI display driver would always reload, no vsync if a game doesn't have the option, lower PPD, i'm back with XP SP3 :shrug:

That's not consistent with my experience I got nothing but praise for vista especially about it's handling of the gpu client. Though I know the issues when you use the card for 3d aren't all solved for a dedicated folder I would advice everyone to go for Vista X64 if their going to run a gpu2 client.

Edit: offcourse you're talking about the other client, ATI and CUDA doesn't mix (yet?)... sorry for the confusion, I wouldn't know which client/os to pick for an Ati card I don't own one :(

[MikeB12]

Synthetic,
concerning your amd x2, those only do about a 1000-1200ppd on smp. and a 8500gt does about 700ppd...
I think your best bet would be an 8800 for that machine, no cpu client.. 4-5000ppd.
but you can run smp on it, I'm running smp on 2 x2's (x2 4600 and x2 6000).

on your quad, you can run dual smp's with affinity changer and it'll net a little more ppd (probably about +1000ppd from what you're seeing now with one smp)...

all you have to do is install AC http://distributed.org.ua/forum/index.php?showtopic=1149
1-install another copy of smp to a separate folder.
2-create desktop shortcuts to both clients install.bats and fah.exe's (labeled 1 and 2 accordingly)
3-add a "-local" flag to the clients fah.exe shortcuts. (the end of the path inside properties of the fah.exe shortcuts).
for example: "C:\fold\fah.exe -local"
-when you launch the 2nd client for the first time and config it, answer yes to advanced config and change machine id to 2.
(the default machine id is 1, so client 1 is already good to go)

(now everytime you launch the clients, run both install.bat shortcuts, and both fah.exe shortcuts (they need to be launched with the shortcut to use the -local flag, otherwise the second client wont run)


that's about as short as I can make it on the dual smp setup...


the amd's don't ike dual smp's for whatever reason. it's best on intel quads.


anyway, I'm running 8 smp's, no gpu2's currently..
and look at my averages. these new gpu2's are monsters, 3 good gpu2's will beat my 8 smp's. so imo, it's best to try and go for the gpu2 when you can if you're thinking about spending money on hardware upgrades.
http://folding.extremeoverclocking.com/sigs/sigimage.php?un=MikeB12&t=77826

I plan on adding some 8800gt or gts g92's to my folders (the g92's are a great ppd value imo-$140-180=4-5000ppd)
gts g92's http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2010380048%201069633099%20106791921%201067924922&bop=And&Order=PRICE
gt g92's http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2010380048+1069633099+106791921+1067932704&name=GeForce+8800+GT&Order=PRICE
but I reserved a beach house for an end of the summer vacation last week of septemeber, so my pc hobby budget is shot until oct..
and who knows ati's code may be optimized by then. I'll see when oct gets here... doesn't really matter now, since I have a beach house to pay off for the next 3 months...

[[XC] Synthetickiller]

You're only asking MikeB12? Ok I'll shut up then :)

*whistles and walks away*

My question was targeted to anyone willing to give me some advice. The GPU client has me VERY interested in folding now, especially w/ a multiple GPU client on the way.

I'd buy another card for my quad, but that's a 4x electrical slot (p35), so I'm sol. I'd get the 8800gts, but wouldn't it be bottle necked by the 3800+? If not, i'd invest in it.

I want to save as much cash as possible for my next build as well as a VERY expensive car improvement job I have in mind. $120 won't be too much to spend. I could always upgrade to an nvidia board, but I'd rather put that money towards a chiller and a nehalem in about 2 years. I'll probably sell my quad at that point to make some cash for a multi gpu set up either for SLI & folding or just a good one card solution w/ other cards for folding on.

I might also consider selling my 3800+ set up on the cheap, and sell my 8800gts if I can get a 9800gx2 or gtx260 (whichever is better) and keep that till I upgrade again. All depends on the GPU client & if I can find buyers as well as if my WU & ppd output would double using the GX2.

[MikeB12]

I'm sure Marvin would have typed something more if he felt like it or had time...
he's not a shy alien, like most. ;)
http://img47.imageshack.us/img47/2333/beammeupvb9.gif


I'm not sure if a x2 3800 would bottleneck a 8800 or not. the nvidia client is not real cpu dependent as you could see from your result above, it only needed 8% of your quad. also someone else reported using a 8800 on one of these threads on a 939 setup and got like 5000ppd... so imo I think it would be fine on a 3800, especially if it were oc'ed some..

[Marvin_The_Martian]

It was a bad joke, sorry for that. I hope you'll be with us for a long time, happy folding and again the joke was a bad one ;)

:up::wtf::up:

@Mike, wel actually I was wondering if really just adressed you since he called you by name ;) And I wouldn't be surprised if he thought my spelling errors gave away a lack of knowledge about the subject at hand :D

[[XC] Synthetickiller]

Actually, I'm looking forward to staying on the team. My next computer will probably have 3 or 4 gpus, 1 for gaming and the others for full time folding.

I also got a friend on my folding on PS3 to join our team, he has about 10,000 points, so that should help.

Marvin the Martian, don't worry about it, trust me. There's not much you guys can do to tick me off. I'm just happy there's people out there who care enough to fold so that the outcome may benefit others.

Honestly, you guys know more about this than me, at least the computer side of it. I've OC'd for years, but finally got hardcore w/ it last year which will benefit you guys luckily. I'm good with hardware, but software has been my enemy, so thanks for clearing up the problems I've had.

[Marvin_The_Martian]

Actually, I'm looking forward to staying on the team. My next computer will probably have 3 or 4 gpus, 1 for gaming and the others for full time folding.

I also got a friend on my folding on PS3 to join our team, he has about 10,000 points, so that should help.

Marvin the Martian, don't worry about it, trust me. There's not much you guys can do to tick me off. I'm just happy there's people out there who care enough to fold so that the outcome may benefit others.

Honestly, you guys know more about this than me, at least the computer side of it. I've OC'd for years, but finally got hardcore w/ it last year which will benefit you guys luckily. I'm good with hardware, but software has been my enemy, so thanks for clearing up the problems I've had.

Looking forward.. like we would let you leave.. This is like the mob, once you're in you're in for life :D

I like your reasoning btw, down to earth. Think you'll make a great addition to the team :up:

One thing, you don't bring with you your points when you change teams, you start from scratch. I got almost a million points if you add them up, but only the recent half has been for XS. Not that it matters, we gladly accept all new members production numbers don't mean much to me.

[MikeB12]

I'm getting 7694.75ppd.

don't be too disappointed with those numbers. I would have killed to get 7k ppd from one pc a month ago... up until recently the best we could get from one pc was about 4000ppd... so 7k from one machine is quite an accomplishment. :up:

[Kingcarcas]

One thing, you don't bring with you your points when you change teams, you start from scratch. I got almost a million points if you add them up, but only the recent half has been for XS. Not that it matters, we gladly accept all new members production numbers don't mean much to me.
I printed out my certificate from my last team and hung it on my wall. :up:
don't be too disappointed with those numbers. I would have killed to get 7k ppd from one pc a month ago... up until recently the best we could get from one pc was about 4000ppd... so 7k from one machine is quite an accomplishment. :up:
Yup, i was happy with 2K just a while ago, now i'm thinking 4850/70 and a quad later down the road, 10K from one box is what i'm hoping for. :D

[Marvin_The_Martian]

I printed out my certificate from my last team and hung it on my wall. :up:

Under which name did you fold before? I didn't print anything out, points don't mean that much to me knowing I got a limited budget and this is an expansive habit, but I did always fold under the same name.

Yup, i was happy with 2K just a while ago, now i'm thinking 4850/70 and a quad later down the road, 10K from one box is what i'm hoping for. :D

Take a q6600 and two 8800gt's and you're done. I wouldn't put my money on it yet, maybe Ati's help will deliver a means to get all those sp's to work but atm it isn't looking like that's right around the corner. Atm the Ati client seems a resource hog with no stamina compared to even an entry level Nvidia card. Offcourse gaming will be allot beter, but then again 8800gt's aren't exactly bad either :up:

It's never been so cheap to get points and help the cause.

:up::wtf::up:

[MikeB12]

I'm looking to overhaul one of my dedicated folders....

curently *Q6600@3.4ghz | GA-P35-DS3L | 2x1GB Ballistix | 8400GS | TT 430w | Raidimax 921WB | Scythe Mugen

The overhaul parts...

$175
Foxconn X38A LGA 775 Intel X38
http://www.newegg.com/Product/Product.aspx?Item=N82E16813186134

3 x $170 (- a $30 mir)
ZOTAC ZT-88TES3P-FCP GeForce 8800 GT 512MB AMP
http://www.newegg.com/Product/Product.aspx?Item=N82E16814500006

$150
Antec TPQ-850 850W
http://www.newegg.com/Product/Product.aspx?Item=N82E16817371009


total overhaul cost $800....



then I'll just have one dedicated GPU2 folder for about 15k ppd...
and the rest of my smp machines, leave alone.. 8k ppd...
I don't want to run gpu2 on my main box because of tv recording.
and if I put a high end video in any of the others, they'll need a new psu..

anyway, that would be a nice overhaul flip for that one folder...

[[XC] Synthetickiller]

Save some cash and fold MORE!

http://fxvideocards.com/ZOTAC-GeForce-8800-GTS-G92-512MB-ZT-88SES2P-FSP-256-bit-GDDR3-PCI-Express-2.0-x16-HDCP-Ready-SLI-ZOTAC-Video-Card-p-16280.html

8800gts g92 512mb. $150 w/ MIR

I don't know how you could fit a 3rd card on the bottom pci-e slot though.

That's $450 vs $510 and you still get better video cards. I don't know if that PSU can handle that though. At least get 2 of those cards, they run cooler and kick the heat out of the case.

[[XC] Synthetickiller]

I need to ask you guys something. I've been kind of busy so I haven't be able to mess with the SMP as much as I would have liked.

When running SMP, how do you force it to save so you can shut down and resume later on? I can't leave my computer on all night. I'll never sleep.

[Marvin_The_Martian]

Close it with ctrl+c, it should write the checkpoint before closing. But will you make the deadline if you close it? If you don't want to run 24/7, you can always use the normal cpu version instead of the smp. It's not about the points, it's about helping a cause. Offcourse, they award points toward scientific value, so getting allot of points should mean you're doing the most good.

Their is only one exception, if you don't make the deadlines. Every unit which isn't in on the final deadline will be given out again to have some other cpu run it, you will still get your points so maybe you think you're not hurting the science, the cause why most people fold, but you are. Because you fold that unit for points, and not for science, I have to fold that same work unit which means you've not contributed in any way, you just delayed the project.

[SparkyJJO]

I just shut it off normally, just go to shut down in the start menu. Or hit Ctrl+C first. Either always works for me. I have the -forceasm flag on it too, seems to help be more stable starting back up.

Shutting down overnight shouldn't make you miss any deadlines.

[MikeB12]

yeah, +1 Sparky...

if you do use ctrl+c, then make sure you give it a couple min to write, it's slow to write.

so I usually leave them running and let windows close them during start menu - shutdown...... that's actually the best method I've found.
sometimes if I use ctrl-c, I don't allow enough time for the files to write and I lose the wu on restart.

but using windows to close them never fails... it seems to force the client to write immediately.

[[XC] Synthetickiller]

Since that's my main concern, making deadlines, b/c as you said, whats the point of folding if I am not helping the study progress?

How can i install the graphical client on windows (regular F@H 5.03) and run it w/ the gpuv2 client?

[MikeB12]

most of your smp wu's are 2665's now. which have a preferred deadline of 6 days. you can run the client 25% of the time and make the deadline on your Q6600.. so even if you only run the client 4-6hrs a day, you'll make it.

I'm running 2 smp's at the same time and they take about 30hrs per client per wu. and that's with 2 running on a 3.4ghz Q6600...
so I turn in 2 every 30hrs of cpu time.. all I have to do is complete that 30hrs in 6 days from dl, and I clear both deadlines. but I let them run 24/7 just because I want the extra points.


I wouldn't bother with the single core client, it's not worth messing with for the little tiny points you get.
if you can manage to run the client 6 hrs a day, run the smp and you'll get about 2000 points for every wu you turn in.

[[XC] Synthetickiller]

Thanks Mikeb12

I'll use the save option about 10 minutes before shutting down to ensure everything goes right. I leave my PC on 12 hours a day or more, so a little over 2 days to complete a unit.

I'm running it now and i'll see how it works.

[Kingcarcas]

What do you guys think about an X2@2.4ghz running 24/7, will it make preferred deadlines?
I need to ask you guys something. I've been kind of busy so I haven't be able to mess with the SMP as much as I would have liked.

When running SMP, how do you force it to save so you can shut down and resume later on? I can't leave my computer on all night. I'll never sleep.
I'm the opposite, i can't sleep if it's off, too quiet :D Plus i have one fan with LEDs, it works great as a night light............yes i'm afraid of the dark a little bit, i probably watch too many ghost shows :shrug:

[Marvin_The_Martian]

I wouldn't bother with the single core client, it's not worth messing with for the little tiny points you get.
if you can manage to run the client 6 hrs a day, run the smp and you'll get about 2000 points for every wu you turn in.

I got something to say about that..

It's not all about the points yada yada yada, yes points in fah have always treid to be a representation on the scientific value, but even then that single core client runs diffrent wu's. You can even run p3's on do work that the other clients this way don't have to do.

So yes, it is worth messing with a single core client.

:up:


What do you guys think about an X2@2.4ghz running 24/7, will it make preferred deadlines?

I'm the opposite, i can't sleep if it's off, too quiet :D Plus i have one fan with LEDs, it works great as a night light............yes i'm afraid of the dark a little bit, i probably watch too many ghost shows :shrug:

Idk haven't ran the smp on any amd cpu but if the deadline is 6 days you should be good. I think when it was 3 days you could have been on the wrong end of the equation.

I'm like you, I can hear my pc's in every room in the house, it even serves as an anti dog screen, not that it keeps actual dogs away but I seem to hear my neighbours dog allot less :up:

Btw,

:wtf: everyone that team graph is looking incredible!

[MikeB12]

hey king, yeah a 2.4ghz amd will make the deadline easy on one smp client running 24/7..
I'm running smp on a x2 4600@2580mhz.
it turns a little better points in linux, currently I'm on mepis with a single smp. and it turns 1125ppd on a 2605.
I used to have it on xp and it turned about 1000ppd on a 2653. (but I haven't had xp since they switched up the wu's).

the key here is dual cores bench different than quads so they get easier wu's imo... at least in linux. not sure about xp.

in any case my x2 4600 linux folder cranks out a 2605 every 1.6 days, or 38hrs. and the deadlines are 3 days preferred, 4 days final for that wu.

[SparkyJJO]

Most dual cores should have no problem. A stock X2 3800+ 1.8GHz might have an issue, perhaps, but seriously, who runs stock? :D

[Kingcarcas]

Ok thanks, we'll see how i go about this, it'll depend on what i get for my b-day :D I was thinking 4850/70, move my 3870 to the X2, but then i realized if i could just run SMP and get the same points i won't need the added power consumption of the 3870 :up: Ps3 when i decide to give a PC a "break".

Or i could go with a Q6600, run SMP and GPU2 on my rig since we still don't know how optimized the 48*0 will be :shrug: I might just wait and see who has the bang for the buck + performance per watt by the end of the year in GPU folding. Everything plays fine with my 3870@16*10.

P.S. PC Lights are great when you have to get up and go to the bathroom in the middle of the night, don't want to step on a cat or something :rofl:

[coo-coo-clocker]

Or i could go with a Q6600, run SMP and GPU2 on my rig since we still don't know how optimized the 48*0 will be

Its a good idea King. The really high points for NV GPU2 won't last. I suspect Stanford is assigning high points as a part of a deal with Nvidia. They only have to leave the points high for long enough for NV to sell x number of cards... seems like a smart strategy. Just conjecture on my part.

But I am not rushing to any conclusions. I didn't go running out and standing up a quad-core smp farm, and I am not going to do the same for NV GPUs. Just keep moderately upgrading and seeing how it all pans out.

:wtf:

[[XC] Synthetickiller]

I just hope the point will increase w/ ati once they get the ati stuff working. I plan on getting that 8800gt, maybe 2. I'm also waiting for newegg to drop the prices on AM2 cpus so I can get a 3.0ghz x2 for 80 bucks. Add in an 8800gt and thats a decent single folding machine for smp & gpuv2. Add in 2 gpu's on my main computer plus an old 1.4ghz mobile pentium m and i'm at least cranking out a little extra ppd and wu's.

[SparkyJJO]

Synthetic, I recently bought an X2 4400+ brisbane and overclocked to 3GHz easy. It isn't anymore because the board I ended up pairing it with doesn't support voltage adjustments :rolleyes:
If you got the 5000+ black edition 3GHz is very common.

[[XC] Synthetickiller]

Ideal set ups IMO

1. AMD

Quad w/ four 9800gx2's for 8 GPUv2 and SMP

Or

2. Quad q6600 w/ four 8800gt's or if you can fit it on a board, four 9800gx2's

Both ideal set ups are expensive.

For the best price / output

Q6600 w/ four 8800gt's .
$209 for the CPU
$100x4 for the 8800gt's
$____ for the mobo.

If you need a new computer, its an option, although many would need a new PSU.

[[XC] Synthetickiller]

Synthetic, I recently bought an X2 4400+ brisbane and overclocked to 3GHz easy. It isn't anymore because the board I ended up pairing it with doesn't support voltage adjustments :rolleyes:
If you got the 5000+ black edition 3GHz is very common.

I have a crappy foxconn mobo from a emachines computer. Sadly, no bios options. Otherwise, I would OC it. I'm stuck with stock which sucks. That's the only reason I'm looking at a higher clocked chip.

[SparkyJJO]

I have a crappy foxconn mobo from a emachines computer. Sadly, no bios options. Otherwise, I would OC it. I'm stuck with stock which sucks. That's the only reason I'm looking at a higher clocked chip.

Oh ok, that sucks. Sorry about that.

[Marvin_The_Martian]

Its a good idea King. The really high points for NV GPU2 won't last. I suspect Stanford is assigning high points as a part of a deal with Nvidia. They only have to leave the points high for long enough for NV to sell x number of cards... seems like a smart strategy. Just conjecture on my part.

But I am not rushing to any conclusions. I didn't go running out and standing up a quad-core smp farm, and I am not going to do the same for NV GPUs. Just keep moderately upgrading and seeing how it all pans out.

:wtf:

:shakes:

I just woke up but I'm ready to :slapass:

Don't spread this kind off FUD around ok?

The reason the points for nvidia are high, is because the benchmark machine the wu's are tested on, hasn't changed. With other words, all wu's are tested on an ati 1900 series card ( iirc 1900 could be a tad newer but I'm pretty sure ). The Nvidia's just do more work then the Ati cards and as such they get more points.

No deals, no nothing.

So please let this be the last time I see this crap here.

Folding at Home isn't in the bargains section if you know what I mean.

[MikeB12]

the 19xx cards went away with gpu1 in the beginning of june. I think the gpu2 bench on the ati side is a 3xxx.

anyway, ev1 knows why ati is not stacking upto nvidia ppd wise right now; it can't utilize more than 320sp's on the current wu's with the current client code..

this might help...
http://foldingforum.org/viewtopic.php?f=43&t=3319&st=0&sk=t&sd=a
I can understand gtx 2xx series folding brilliantly, but where does g80/92 power comes from? For ages ATI was murdering nvidia in any scientific or rendering test, and suddenly nvidia somehow finds all that unknown folding power. So is ATI coders so pathetic, that they couldn't achieve in few years what nvidia achieved in few months? That is looking at PPD of 8600GT (~2100PPD) and 3850 (~1900-2000PPD). Can 8600GT really do 250ns?

It's really not so simple. Note that all of this is done with FAH programmers architecting the calculations, getting help from NVIDIA and ATI for tuning/porting to their hardware. The NVIDIA guys (i.e. Scott & Mark on the FAH side) got to see the ATI code as a reference, so while we spent 3 years perfecting the ATI code, the NVIDIA project got to start from that perfected code, so the NVIDIA code didn't just come together from scratch in a few months, we did it building on 3 years worth of work.

There's also some other complexities here. It looks like the small proteins map reasonably well to the NVIDIA hardware, much less well to the ATI hardware (especially on the high end). The relative speeds will likely change as we change proteins (especially on the high end hardware), so this will get more complex. We're getting help on the ATI side to tune performance, so this will continue to be interesting as time goes on.

With that said, Scott (@NVIDIA) has done a fantastic job porting and tuning our GPU code to CUDA and he deserves a lot of credit -- and I bet there may still be more performance optimizations to come.

[MikeB12]

Hey Synthetic.. here's a brisbane x2 5400 (2.8ghz) for $87. http://www.newegg.com/Product/Product.aspx?Item=N82E16819103235
I have 2 x2's on cheap mobo's with no bios oc control.
both mobo's are nvidia nforce chipsets, and I use ntune and get about +12-15fsb on stock voltage before it becomes unstable.
http://www.nvidia.com/object/ntune_5.05.54.00.html
just save the ntune profile and it boots to the oc every time.
my x2 6000 will do 212x15*3180mhz on stock vcore with ntune.
my x2 4600 will do 215x12*2580mhz on stock vcore with ntune.

so a stock x2 5400 @ 200x14*2800mhz would run very close to 3 ghz on stock vcore. 213x14*2982mhz (assuming you'll get somewhere bt 212-215 on the stock vcore fsb oc, using my 2 as reference-the lower multi's seem to go a little higher)
anyway, that would get you to 3ghz ish on a $87 AM2 cpu. with a cheap oem mobo as long as it is an nvidia chpset.


just some food for digestion in case you need it.

[Marvin_The_Martian]

Yeah maybe I should have pointed to the hw details but I just got this red flag waving before my eyes, sorry.

I can't let someone make accusations about these kind of things. Stanford is an University not a buisiness, Vijah is a professor, not a buisinesman. This project is about finding cures not about making deals with manufacturers.

It's just a red flag with me.

And DOH, gpu2 doesn't run on x1xxx series so yeah I think it probably would be a 3870 seeing the previous one was the 1900xtx ( also the top model of that generation ).

[coo-coo-clocker]

I can't let someone make accusations about these kind of things. Stanford is an University not a buisiness, Vijah is a professor, not a buisinesman. This project is about finding cures not about making deals with manufacturers.

It's just a red flag with me.

woah, easy there Marvin.
I am stating an opinion - and one that I couched as conjecture on my part, since I have no facts whatsoever to back it up. I appreciate your willingness to believe in the pure non-profit orientation of the school and the project, but you and I disagree on that point. Universities (especially private schools like Stanford) are run like businesses (http://www.bizjournals.com/sacramento/stories/2007/12/31/daily10.html)... they have CFOs, manage multi-billion dollar balance sheets and are highly motivated by profit and revenue. There was no sinister intent in my remarks - I think it's smart to create partnerships with large manufacturers like Sony, ATI, NV, etc. It's naive to think there aren't profit motives within a huge global project like F@H. Kudos to VP for using that leverage to advance to goals of the project.

[Marvin_The_Martian]

Ok I see your point and it's valid. VP has to work with both Nvidia and Ati, when the Ati cards where the only one's capable of folding the Nvidia camp had rumours about it being a setup, while the unified shader architecture of the x1xxx cards is what made it technicly possible in the first place. I know, people who are more interested in the project know better then to listen to those rumours, but just as the Futuremark and Nvidia claim, once it's made it's impossible to get that notion swept aside doesn't matter if all the facts point to the other side.

Just as with this.

You say you don't agree, but please, take the quote from Vijah and tell me which part isn't true then? Isn't the ppd derived from a benchmarking machine? Isn't that an Ati card, so that when the wu's are assigned to an Nvidia card which happens at this point to be that much faster, they are turned in that much faster and thus generate x amount off points?

Which part isn't true? :shrug:

If you asked me, does money play a role, yes offcourse. But is the Pande Group using politics to gain from it? Because basicly what you're saying is; Nvidia, who donated Whoopass btw ( or only MacFolden ), got our gratitude for that so we decided to make it look like their hw is more efficient at our cause? Isn't that what you said? If it's not, I apologize again this time for misunderstanding you. But if you accused them of such a thing, sorry coo-coo-clocker but I think you're really ready for the coo-coo house ;)

Think about it, Ati has done more work for Pande then Nvidia but they decide to favour Nvidia?

I know what you mean by saying you give Kudo's to VP, and I think I been saying throughout the forum that I think Nvidia should play this out even more publicly, since I like any attention to the project, but also since the ggpu calculations can speed up finding a cure so incredibly much it will be a waiste to not inform everyone about this.

[MikeB12]

Marvin likes a good debate... :eek:
http://img366.imageshack.us/img366/8049/x1ls8.gif

I don't think he meant anything by the original fah comment, Marvin. it was just a offhanded conjecture comment... at least that's how I took it.. :)

[Marvin_The_Martian]

Lol!

If it was a conjecture comment, I would like to know where it conjectured from, so I asked to be more specific. Sure, Stanford is a private school, but the Pande Group proud themself on a certain aspect of their effort namely that their purely doing this for science. The results of Folding At Home, are public in opposition to some other projects, which rules out any motive of gain. Why would they be so open and forward about this being important, to go make deals with hardware manufacturers to supposedly help sales?

If this isn't a point off debate, and I know for most of us it's not, I would like him to retract the statement, and if he stands behind it I would very much like him to rebute my reasoning. Offcourse, I don't carry any weight around here so he can just ignore me if he whishes but for me personally these remarks do harm to the project and I will not stand by and let that happen, atleast not without trying to point out any flaws I see in their line of thought. So far, his only points have been, hey they get more points and schools are run like a buisiness because it's dependant on funds. Their both blatantly and for me atleast, obviously, wrong.

[SparkyJJO]

Whoa easy guys, nobody attacked your sister ;)

I was under the impression that it was higher points because of more work done. Like back in the 7900/x1900 days the 7900 sucked for FAH really, there was a chart where it was around a P4 capability IIRC and the coding for it was extremely difficult. Now, the nvidia cards are more capable and coding isn't as hard, and at least right now it seems they do more than the 3800 and 4800 series cards.

I do think that stanford may boost the points temporarily on newer clients, to get more people interested so they can have more input on getting the betas worked on, but I don't think it has anything to do with paying nvidia off or anything like that. Intel, AMD/ATI didn't get any payoff on their previous clients so why should nvidia? :shrug:

I'll just leave it at that.

Please don't get all bent out of shape over this stuff people, it really isn't worth arguing about about :)

[Marvin_The_Martian]

No problems here :)

Sparky I send him a pm to explain why this gets the hairs in the back off my neck standing and hoping he will reply to it.

I enjoy reading his posts and I am thankfull for his contribution so don't think I got anything personal against him of his believes, he's entiteld to anything he whishes to believe. But if you make such a public statement, and conjecture thought does not invalidate this, you have got to be prepared to back it up since a mentioning of foul play if left unchallanged can turn into indisputable proof of something in the long run if you know what I mean.

And Sparky you're right, they always boost the ppd for beta clients but then again that's because they expect those who run it to monitor it more closely and be actively involved in reporting bugs. That's the side of the points awarded we will loose when the client comes out off beta but I don't think it will be that much. The only way for the Ati camp to gain, is to get more efficient and since their currently only running on 320 sp's instead of 800 I think we can assume that will happen. I don't know if they'll be able to double the output, I think there are more bottlenecks then just stream processor count. If the Nvidia's do just as well with the larger wu's as the current smaller one's I think Nvidia will definetly keep the lead in this process, but that infact is just a conjecture thought :up:

I would just like to end with a smilley mkay? :wtf: :buddies:

[[XC] Synthetickiller]

Hey Synthetic.. here's a brisbane x2 5400 (2.8ghz) for $87. http://www.newegg.com/Product/Product.aspx?Item=N82E16819103235
I have 2 x2's on cheap mobo's with no bios oc control.
both mobo's are nvidia nforce chipsets, and I use ntune and get about +12-15fsb on stock voltage before it becomes unstable.
http://www.nvidia.com/object/ntune_5.05.54.00.html
just save the ntune profile and it boots to the oc every time.
my x2 6000 will do 212x15*3180mhz on stock vcore with ntune.
my x2 4600 will do 215x12*2580mhz on stock vcore with ntune.

so a stock x2 5400 @ 200x14*2800mhz would run very close to 3 ghz on stock vcore. 213x14*2982mhz (assuming you'll get somewhere bt 212-215 on the stock vcore fsb oc, using my 2 as reference-the lower multi's seem to go a little higher)
anyway, that would get you to 3ghz ish on a $87 AM2 cpu. with a cheap oem mobo as long as it is an nvidia chpset.


just some food for digestion in case you need it.

My 3800+ i sitting at 2600+mhz. Not huge, but that should make help. Luckily, I had an nvidia chipset.:D

I'm just waiting to see if its stable.

Thanks buddy! :up:

Edit: 217x12 is stable so far. I might try for 220x12 later. If I can creep it up to 225x12, I'd be at 2700mhz.

Edit 2: Broke the 2.7ghz barrier. 228x12 (2.736ghz) didn't create any instability or BSODs yet. I'll let this run a while and see if I can push it higher. Maybe I have a great chip?

[MikeB12]

wow! I got my x2 4600 to 218 @ 100% before the bsod's kicked in... my x2 6000 wouldn't go above 215. once I backed em down I ended up with 215 and 212...

jeez, that's awesome...


actually, I used that little 20 min auto tune ntune feature on one of the tabs to find the maxes. it'll oc the pcie bus too though, so make sure your turn that back down if you use the ntune builtin auto tune. that's all ntune 20min tune does is find the max and back it down a couple on reboot of bsod.

[Marvin_The_Martian]

:rolleyes:@you guys

Come on, this is XtremeSystems even if we're a folding team we don't use ntune for overclocking...

Though, I think I'm the one who should have started a thread here when I joined and asked the old boys to help me maximize my torture practises ( while keeping em alive, to be tortured more offcourse ) since yeah this is XS and I'm still just a newb compared to most.

@Synthetic

Nice results :up:

That's going from the above poster as my current lineup does not include a brisbane nor did I ever own one. Last Amd was a venice 3200, but I do think the people on xs should atleast know enough to not rely on ntune ;) Last time I used an auto oc tool was... ehm... never?

[[XC] Synthetickiller]

:rolleyes:@you guys

Come on, this is XtremeSystems even if we're a folding team we don't use ntune for overclocking...

Though, I think I'm the one who should have started a thread here when I joined and asked the old boys to help me maximize my torture practises ( while keeping em alive, to be tortured more offcourse ) since yeah this is XS and I'm still just a newb compared to most.

@Synthetic

Nice results :up:

That's going from the above poster as my current lineup does not include a brisbane nor did I ever own one. Last Amd was a venice 3200, but I do think the people on xs should atleast know enough to not rely on ntune ;) Last time I used an auto oc tool was... ehm... never?

Marvin,

The mobo has NO oc options in the bios at all. I don't have other options :p:. If you look @ my sig, I clearly don't want to use software over bios, but I'll do what I can with what I have.

MikeB,

Its not an x2, so maybe that's why I can push it much higher than you can?
If my OC is stable, I will try for 234x12 to break 2.8ghz. Not bad for a POS machine w/ a crappy chip eh?

Update:

Won't Oc past 230 or 231, so I'm leaving it there. 2.76ghz isn't bad, although I might be selling it tonight. If so, i'll get that 8800gt for 70 after MIR. I wish I could get a 750i, I'd get another 8800gts (the zotac for 120) and SLI those bad boys.

[Marvin_The_Martian]

:poke:

Can't take a joke?

Btw, don't reference to your signature, take the time to tell me you're a dedicated builder who chooses his parts carefully ( though.. the ram ;) ) it's more polite and you won't get any smart remarks from a Martian about his own signature ( the part where it says your signature does not describe you, by telling you my specs don't describe me ). I have allot of respect for someone who takes the time to talk instead of reference. Just a quirk I got. Same goes, obviously, for postcount. Though that might change when I get more posts myself. If no one get's it, it was a joke lighten up

[[XC] Synthetickiller]

:poke:

Can't take a joke?

Btw, don't reference to your signature, take the time to tell me you're a dedicated builder who chooses his parts carefully ( though.. the ram ;) ) it's more polite and you won't get any smart remarks from a Martian about his own signature ( the part where it says your signature does not describe you, by telling you my specs don't describe me ). I have allot of respect for someone who takes the time to talk instead of reference. Just a quirk I got. Same goes, obviously, for postcount. Though that might change when I get more posts myself. If no one get's it, it was a joke lighten up

Ha, usually I do a run by posting, so I don't always take the time to read everything and take it all in. :p:

Yeah, my ram kids suck. My evo kit is a POS just like the G Skill. I just leave them at ddr2 800 speeds. My AMD rig has one stick of crucial ballistx, the remnants from two 1gig x2 kits that I blew out. Damn micro d9 stuff is so sensitive.

I am really considering that $70 8800gt, since my crunching would double. I'm almost at 50k points (43k right now). I'm just trying to break top 100 and then eventually top 25. IN THE NAME OF SCIENCE (and a little ego :p:)

[Marvin_The_Martian]

Ha, usually I do a run by posting, so I don't always take the time to read everything and take it all in. :p:

Yeah, my ram kids suck. My evo kit is a POS just like the G Skill. I just leave them at ddr2 800 speeds. My AMD rig has one stick of crucial ballistx, the remnants from two 1gig x2 kits that I blew out. Damn micro d9 stuff is so sensitive.

I am really considering that $70 8800gt, since my crunching would double. I'm almost at 50k points (43k right now). I'm just trying to break top 100 and then eventually top 25. IN THE NAME OF SCIENCE (and a little ego :p:)

I would wait till the client comes OUT off beta to decide what camp to choose for points. Like I said allot off times, in principal the Ati cards should do better if their able to utilize it fully. Though mhouston on the folding forums told me ' it's not the schedulers per se, it's more about getting the code to issue things wide enough' ;)

Atm I am voting for Nvidia for folding but I can't recommend anyone a buying decission based on my assumptions. (Gut feeling: get some cheap nvidia's. Other gut feeling, wait till their even cheaper.)

The fah people are saying the same ( and with them I mean the mods at the folding forums ) and I can't say more then point to them atm.

edit:
7im wrote:The Villin WU folding now is only 500 molecules. Double the number of molecules, and the complexity goes up exponentially, and so does the need more shaders, etc.

When a new client is released, Pande Group starts out with well known proteins that are small and quick to finish. That way it is easy to see if the client is working well or not, and they get back lots of feedback very quickly. Starting with a small WU does both of those.

Besides, these were benchmarked on a ATI card. The NV cards perform very well on small WUs, ATI not so much.

All these performance numbers are VERY preliminary. Give it a month for the optimizations to settle in, drivers to get tweaked, an for some real (larger) work units to get folded, and then let's see where we are.

I sure wouldn't be making a buying decision based on the performance numbers for a client that is less than a week old.

It's getting a lil dated already with all the updates ect times seems to fly, and I think the newer projects are already larger then the first one's. Idk, follow my gut and wait ( for them to become even cheaper??)?

[Kingcarcas]

From my understanding the reason Nvidia cards are getting so many points is because they're getting small WUs. When Stanford starts releasing larger ones we should see the PPD come down and i'm hoping once they utilize more than 320SPs on the 4850/70 we should see their PPD skyrocket :up:

[SparkyJJO]

I'm sitting back and watching the whole GPU crunching pan out. One good thing about not having too much dough to throw around is all I can do is watch and not be too tempted to jump into anything :lol:

[Marvin_The_Martian]

Kingcarcas, your card isn't using 320 stream processors. What mhouston has been saying is, they can adress four off the simd's which have 80 sp's in total making a MAX of 320sp. If you follow the thread on the folding forum, and if he answers my questions we'll soon know more.

Sparky yeah well you're missing out on allot off fun to, and if you don't step it up 'm going to overtake you in a year ( not that I think it will happen, I think you'll crank that ppd allot higher ). :poke:

Idk I understand you, but a single 9600gt does 4K atm, almost double the ppd from your entire farm :shock:

Up to you :2cents:

[[XC] Synthetickiller]

I would have to get nvidia b/c i already have one nvidia card in my pc now. Can't run that and ati. Thats why I think the 8800gt is good, but I'll wait for it to come out of beta first.

[Marvin_The_Martian]

http://foldingforum.org/viewtopic.php?f=38&t=3376&st=0&sk=t&sd=a

That's a thread about which card is actually the best for us based on current reported ppd and power claims. Though I plan on doing my own calculations when the time comes and everything is out in the open, just to be sure :)

[SparkyJJO]

Kingcarcas, your card isn't using 320 stream processors. What mhouston has been saying is, they can adress four off the simd's which have 80 sp's in total making a MAX of 320sp. If you follow the thread on the folding forum, and if he answers my questions we'll soon know more.

Sparky yeah well you're missing out on allot off fun to, and if you don't step it up 'm going to overtake you in a year ( not that I think it will happen, I think you'll crank that ppd allot higher ). :poke:

Idk I understand you, but a single 9600gt does 4K atm, almost double the ppd from your entire farm :shock:

Up to you :2cents:

Well the stats counter is still adjusting for the new Q6600 and in a couple weeks when my ram for fold8 comes back from loan I'll be adding another 700-1300 PPD (depending on WU) from that. But yeah, I know. Tempting to grab one of those cheap cards and throw it in fold6 and turn the old single core s939 system into a points monster heh. But I must...watch...budget... :( Plus I don't think the old 450W PSU in there is up to the task, it isn't a very modern PSU and doesn't have a huge amount of amps on the 12V, let alone any PCIe connectors.

[Marvin_The_Martian]

Ow yeah... seems I got the slowest q6600 on the team ( someone sponsor me a nice P35 ) :D

[SparkyJJO]

Ow yeah... seems I got the slowest q6600 on the team ( someone sponsor me a nice P35 ) :D

What speed is it running at?

I'm quite happy with my abit P35-E, so maybe look for one of those. They aren't too expensive. Though it needs extra cooling on the mosfets and NB, even the SB gets a tad warm. When I'm back home next I'm going to dig through my old heatsinks and see if I can create something for the mosfets, and I'm pretty sure I have a larger heatsink that'll fit on the SB. I have the HR-05 on the NB already.

[Marvin_The_Martian]

333x9 on a p5w dh. I can do 3373 fah stable but that requires so much volts on my nb it will eue a wu if I don't keep the dust out off my 40mm fan which I attached on the nb ( where I cut the top copper plate off to get a normal hs which I could fan ). Same board will take my c2d over 4g spi 1m and 3800 smp stable, but can't mess with Vgtl on the board and no ( easy ) mod for it.

Those abit's seem nice, but I don't think I can afford unless I sell my p5w dh which costed about 3 times as much iirc. I was actually considering keeping this setup till nehalemm arrives, maybe only adding another nvidia card for folding, or two 4850's later on if they pan out to be ok since it's a CF mobo.

[SparkyJJO]

Wow only 3GHz. I'm on stock NB voltage on this IP35 even. What is the Vgtl?

[Marvin_The_Martian]

vgtl is the thing which got adjusted in the second batch off 680i boards, hope that tell's you something as they did it to improve quad core fsb overclocking. FCG had some nice articles about it at the time, also he was the one I thought to get a vgtl mod out for this board I got but he never finished it.

Btw, gpu server seems down.. I'm dropping like a stone everytime that happens, so please guys, let me get the wu's first eh :D :up:

[MikeB12]

I am really considering that $70 8800gt, since my crunching would double.

yeah, I got suckered into that deal too.. Mine gets here Tues according to ups track# eta...

it'll be a nice little addition, I'll probably stick it in my X5 box below and continue running dual smps along side it. I wanted more, but the $30 mir is only good for 1. currently I'm all smp, no highend gpu2's, so even a single gpu2 4500ppd addition will be nice
http://img381.imageshack.us/img381/9682/35620642iq4.jpg


some people say the 256mb ram hinders it, but for folding I don't see how you'd ever need more than 256mb on a single fahcore_a11. even if the number of atoms doubled from 576 now to 1152. we'll see if that ever happens.
I mean fah has said bigger wu's would be out shortly for the ati gpu2 when it came out earlier this year... and that promise is still sitting on the sideline. so who knows when or if this bigger wu/code opt day will arrive. :shrug:
because even the old ati gpu2's from the first gpu2 release earlier this year are still chugging along at the 500ish atom level. sure they increased the 98 pointer values, but that didn't change the atom size or the complexity.
http://img352.imageshack.us/img352/418/marvinthemartianxz8.jpg

[Marvin_The_Martian]

Idk about your last sentence, it's not just atom size it's the number of folds so even a 550 atom wu can take longer then a 800 atom wu if the number of folds is just a tad higher.

[MikeB12]

actually that's what they did to the 479point nvidia wu's recently.

it still doesn't affect the ram usage or complexity of thread calculations, it just increases frame time... ppd remains the same. and additional sp's are not utilized, so it has no impact other than longer wu's, not bigger/complex wu's..

I'm sure the 4xxx ati gpu2 guys would love a short length 800atom wu.. that would be manna from heaven for them. who cares how short the folds were, it would be a better wu for the 4xxx's.

[Marvin_The_Martian]

Ok thanks I thought the complexity increased with each fold to ( as each fold represents multiple possibilities ).

[[XC] Synthetickiller]

I have a question for you guys....

Since I have been having problems getting SMP to save w/ ctrl+c, any way to install F@H5.03 w/ the GPU client? I can't run both. It would be a little simpler for me and I'd still get out more wu's. Any install tips?

[MikeB12]

smp---just reboot (start-menu-restart) to kill the client. every time I say this people come out of the woodwork saying no no no no... but try it, trust me it works... I don't lose wu's anymore, and I used to lose them all the time with ctrl+c or x... especially in vista.. so if anyone wants to debate the issue, you're gonna debate alone, beacause you haven't tried it. plus what's quicker---> ctrl+c and waiting 10 min for the processes to end, or rebooting and being back up in 60-90 seconds.

there is no trick to the console client... just click the exe and go. no setup needed. http://www.stanford.edu/group/pandegroup/release/FAH504-Console.exe
but honestly your forfeiting major ppd...

[SparkyJJO]

What's odd is both restarting the PC or hitting ctrl+c works for me, and ctrl+c has never taken more than a couple seconds to finish..... :shrug:

I found a long time ago when SMP first came out that I was bombing out units from shutting the client off every so often, so I put -forceasm in the command line and that seemed to help. Appeared (to me at least) it wouldn't shut down "properly" half the time and adding that flag made it not check to see if it was a "proper" shutdown and would just continue where it left off. Since then, I've had zero issues with nuking a WU at restart. This is across 4 different systems I've done this too, old opteron, the phenom rig, fold8 in my sig, and now my Q6600. Worth a try anyway :shrug:

[[XC] Synthetickiller]

smp---just reboot (start-menu-restart) to kill the client. every time I say this people come out of the woodwork saying no no no no... but try it, trust me it works... I don't lose wu's anymore, and I used to lose them all the time with ctrl+c or x... especially in vista.. so if anyone wants to debate the issue, you're gonna debate alone, beacause you haven't tried it. plus what's quicker---> ctrl+c and waiting 10 min for the processes to end, or rebooting and being back up in 60-90 seconds.

there is no trick to the console client... just click the exe and go. no setup needed. http://www.stanford.edu/group/pandegroup/release/FAH504-Console.exe
but honestly your forfeiting major ppd...

I have driver issues (my luck) and I almost had to reformat my vista partition. SMP doesn't seem to like my computer, but honestly, I need the cpu power for other things.

OCing my 2.4ghz amd to 2.7ghz has really helped oddly. I was getting 88ppd before. Now I am getting 400+ ppd.

[Kingcarcas]

Hmm..........does anyone have SMP on their gaming rig? How do you pause it so you don't lose a WU?

[MikeB12]

smp is according to stanford a baby sit client, so it is finicky..

when I want to do something that requires stopping the client, I reboot the pc to stop the client, then do what I want.... and start the client when done.

like sparky said xp is alot more resilient. but still you have to find a method that works for you and not lose wu's...

there is no pause for smp. and if you complain about it fah forum, they tell you to run the regular client if you can't handle smp's quirks.
so it's a catch 22 type thing... dammed if you do, dammed if you don't.... I've seen quite few users get po'ed at fah for smp beta junk. and their respsonse is always the same... if you don't like it, run the regular cient and turn less points...

smp has been in beta for over a year, I doubt it's coming out.

[SparkyJJO]

Hmm..........does anyone have SMP on their gaming rig? How do you pause it so you don't lose a WU?

I don't pause it, I just run it. Never really noticed a slowdown even on my old dual core opteron. Well, the client slows down automatically, you'll notice that in the logs that it took longer between percentages, but the games don't seem to be affected by it running.

[Marvin_The_Martian]

Same here the smp doesn't slow my crysis down ( it stops the gpu client offcourse, if I don't pause it myself as it sometimes treis to start during level loads I think and it makes it very choppy the first few seconds off every level.

Since you can run it as a service there has to be a command line option to pause it to but doing a "fah.exe -pause" doesn't work and I don't really have much issues with the smp atm so I stopped thinking about it.

[[XC] Synthetickiller]

Hey guys, sorry to bring this thread up again, but how do i tag each instance of SMP? I have one running and it saved when I shut down windows :up:, but now i need MORE! I can install more instances, but can't run them and the walk-throughs are not helping me (They aren't clear to me and I'm still a n00b).

Plus how many instances should I run in total, 4? Also.... set each to 25% cpu usage?

[MikeB12]

simple answer is: use the -local flag to run 2 clients.. and set machine id=2 on the 2nd client in the adv config questions.

but it gets cloudy really fast...

you also need affinity changer with 2 smp's to maximize fahcore_a1 thread efficiency http://distributed.org.ua/forum/index.php?showtopic=1149

and it gets trickier. if you're also using the gpu2 client, then it only works in Vista.


In XP you run 1 smp and 1 gpu, and use this app to feed the gpu one core..
http://www.geocities.com/edgemeal_software/Files/Set_Affinity_II_1.035.zip

and by using this program to lock affinities, you can no longer us affinity changer... so you have to drop to one smp.
so xp sucks for smp/gpu folding..
http://www.xtremesystems.org/forums/showpost.php?p=3140035&postcount=14



refrsh my memory here:
what os are you on and what are you trying foldon hardware wise. I'll try to give you a better answer..

[[XC] Synthetickiller]

Thanks MikeB,

I'm on vista 64, set up is my sig (q6600, 8800gts, 4 gigs ram, took other 4 gigs out), so I'm set for multiple SMP & GPU :D

So i would rename the core "fahcore_a1 -local flag1" or something like that? I'm a little confused, sorry.

This is the only hardware I'm concerned with right now.

[SparkyJJO]

No you don't rename anything.

In your run box, put in "C:\FAH SMP2\fah.exe" -configonly (or where ever your second SMP client is installed) this brings up the initial config menu. You can just hit enter through it all until at the end it asks for machine ID. Set it to 2.
then, in the shortcut that starts the SMP client, put the same thing as above, except change the -configonly to -local and go ahead and add -forceasm as well (it helps with shutting down and resuming the clients, at least for me). That should get you set with the second SMP client. Then go to the shortcut to the first one, and put the -local (and -forceasm if you don't have it) on the end.
So now, the path for each should look like
"C:\FAH SMP1\fah.exe" -local -forceasm
"C:\FAH SMP2\fah.exe" -local -forceasm

again, the FAH SMP1/SMP2 part would be where you have them installed.

[[XC] Synthetickiller]

No you don't rename anything.

In your run box, put in "C:\FAH SMP2\fah.exe" -configonly (or where ever your second SMP client is installed) this brings up the initial config menu. You can just hit enter through it all until at the end it asks for machine ID. Set it to 2.
then, in the shortcut that starts the SMP client, put the same thing as above, except change the -configonly to -local and go ahead and add -forceasm as well (it helps with shutting down and resuming the clients, at least for me). That should get you set with the second SMP client. Then go to the shortcut to the first one, and put the -local (and -forceasm if you don't have it) on the end.
So now, the path for each should look like
"C:\FAH SMP1\fah.exe" -local -forceasm
"C:\FAH SMP2\fah.exe" -local -forceasm

again, the FAH SMP1/SMP2 part would be where you have them installed.

Both are running now. Thanks. I only had to run w/ "configonly" and its working now. I'll know later if it saves after restarting.

Thanks! :up:



All you guys w/ quads, you're running 4 instances, right?

[SparkyJJO]

I think you want only 2, because it takes a minimum of 2 cores to run each client. I'm currently just running one SMP, I suppose I could do two..

[[XC] Synthetickiller]

I noticed when running 2 it takes MUCH longer to finish than one. One client takes 12 minutes per percent, so 1200 minutes. It was more than double w/ 2, so I'll stick with this set up. 1 smp, 1 GPU. Thanks for the info though!

[MikeB12]

sounds like you are missing ac...

dl affinity changer and install it http://scootplace.net/fah/FahSmpAffinityChangerSetup_x64_1.0.4.zip
it's a service that runs on it's own... and optimizes the threads and frame times..


on a Q6600@3.4-3.6ghz with affinity changer running you should be getting better frame times closer to
Project : 2665
-- X4-smp2-Vista-Q6600-3420mhz -- Avg. Time / Frame : 16mn 08s - 1713.72 ppd
-- X4-smp1-Vista-Q6600-3420mhz -- Avg. Time / Frame : 16mn 10s - 1710.19 ppd

I'm turning about 8000ppd on a Q6600@3.4ghz and a 8800gt 700/1700/750 in Vista 64. >>>http://www.xtremesystems.org/forums/showpost.php?p=3140035&postcount=14

[[XC] Synthetickiller]

My SMP client right now is doing 2185 PPD.

I just DL'd the dl affinity charger. Should it help a single client or just when you run 2 clients?

[MikeB12]

run 2 smps on a quad, one smp on a dual core. it only helps intel. not optimized for amd...

you'll add about 1000ppd if you run 2 smps/with AC on a quad. (over just one smp)