F@H and Linux

[W.Feather]

Quick question....is there a client similar to HFM.Net for linux?

since schools starting up and I really shouldnt be gaming during the week, ill be keeping my box on linux for the week days(unless i need something in windows), so it will pretty much fold 24/5, and i would like an easy to to check its PPD and such.

 

[musky]

HFM might run under Wine. One of them (FahMon?) is cross platform. I was going to set up a Linux VM tonight, so I'll see what I can come up with.

 

[W.Feather]

cool, thanks man

 

[musky]

Alright, HFM would be a complete pain in the behind to get to work through Wine since it is a .net app. You need Mono, and I managed to get way over my head trying to get it to work pretty quickly. In theory, it is possible if you know more about Linux than I do.

FahMon, at least in Debian, is tricky as well. If I really needed to do it, I could probably figure it out. It is supported natively in Linux.

In other news, -bigadv may again be possible in Linux. More to come...

 

[W.Feather]

well, i'm running ubuntu, and well my dads main expertise is in Linux, pretty much every server/machine he has at work is linux/unix based, so ive got that covered......do you have a link for what all your talking about?


on a side note: Ive got a 9800GTX just sitting around wiating for me to get my lan setup going....i'm also not very concerned about power draw(its high for us no matter what....and weve got all desktops doing DC projects in downtime anyhow), what points would I expect? enough to warrent wiring up another two PCIe power connectors?(i post this here so i dont make another thread )

 

[ccityinstaller]

well, i'm running ubuntu, and well my dads main expertise is in Linux, pretty much every server/machine he has at work is linux/unix based, so ive got that covered......do you have a link for what all your talking about?


on a side note: Ive got a 9800GTX just sitting around wiating for me to get my lan setup going....i'm also not very concerned about power draw(its high for us no matter what....and weve got all desktops doing DC projects in downtime anyhow), what points would I expect? enough to warrent wiring up another two PCIe power connectors?(i post this here so i dont make another thread )

I am getting between 5-6K PPD out of a 8800GT (shaders @ 1854Mhz) with 112SPs, so you should get at least another 1K PPD since you have 128SPs, and should be able to hit a pretty high shader o/c..

 

[musky]

http://fahwiki.net/index.php/Third_Party_Contributions

There is an FahMon link on this page, along with all the others.

 

[W.Feather]

thanks musky, ccity ill figure out a good way to put it in there and let ya guys know the points i get....the shader did clock welel, was memory limitted on OC......does shader help more than the core speed?

 

[W.Feather]

question with Linux bits.....can it use two graphics cards to fold?

edit: also is just plugging in my 9800GTX all I need to do to get it to work in windows(no monitors will be attached as of now to it(maybe I can find myself a spare one some where....one can hope

 

[Zero82z]

question with Linux bits.....can it use two graphics cards to fold?

Yes, but you will need a monitor or dummy plug connected to each card.

 

[musky]

Yes, but you will need a monitor or dummy plug connected to each card.

I'm pretty sure that you don't need the dummy plug. I have two cards in a machine long ago and I never had a dummy plug.

Links:
http://foldingforum.org/viewtopic.php?f=54&t=6793&start=0
http://www.aoaforums.com/forum/nvidia-ati-gpu2-clients/48001-64-bit-linux-setup-guide-gpu2.html

Nothing about two GPU clients, though. I used to have a link to a nice guide, but the site appears to be gone now.

 

[sirmonkey1985]

I am getting between 5-6K PPD out of a 8800GT (shaders @ 1854Mhz) with 112SPs, so you should get at least another 1K PPD since you have 128SPs, and should be able to hit a pretty high shader o/c..

just to correct you.. your shaders are at 1836 not 1854 on the 8800GT (remeber g92 based cards shaders go up in 54mhz steps.. 8800GT starts at 1512mhz so you can do the math from there)


the 9800GTX though will do around 500-700 PPD more then the 8800GT clock for clock with the shaders.. 9800GTX should be able to do 1836 probably 1890 on the shaders.. just make sure the gpu is above 650mhz.. seems to be the sweet spot on g92 based cards when the shaders are over 1800mhz.. adds around 200-300 ppd..

 

[W.Feather]

would a monitor need to be plugged in all the time? i can get one on there for setup, if need be i can always find a converter(think a friends got a spare) and just hook up a small CRT ive got laying around

 

[sirmonkey1985]

would a monitor need to be plugged in all the time? i can get one on there for setup, if need be i can always find a converter(think a friends got a spare) and just hook up a small CRT ive got laying around

should be able to just plug a monitor in.. start the client and unplug it.. should keep working.. then just use the remote monitoring in hfm to keep track of it.. i know that worked for windows..

 

[Kendrak]

Why not just run windows and run a linux VM. Seems like a simpler solution.

 

[W.Feather]

ill be tempted to play games, which is why i'm wanting to stay in linux during the week.......

GPU Folding is doable in Linux right? if not, ill be win windows for sure, to get the most folding

 

[Zero82z]

I'm pretty sure that you don't need the dummy plug. I have two cards in a machine long ago and I never had a dummy plug.

Links:
http://foldingforum.org/viewtopic.php?f=54&t=6793&start=0
http://www.aoaforums.com/forum/nvidia-ati-gpu2-clients/48001-64-bit-linux-setup-guide-gpu2.html

Nothing about two GPU clients, though. I used to have a link to a nice guide, but the site appears to be gone now.

Yeah, you're right. I got mixed up for a second. What I had to do to get multi-GPU folding in Linux was not to use dummy plugs, but to fake having monitors attached by editing my xorg.conf file. You don't need dummy plugs because you can just do it in software. However, without either that trick or a monitor connected to each card, they would not fold. I don't know if anything has changed since then, but that''s how it was when I was running a triple-GPU box under Linux.

 

[W.Feather]

so GPU folding is supported in linux(not through wine, ill have to find my license to WINE, what use not using it if ive got a full copy?)

 

[musky]

GPU folding is not officially support in Linux. You can make it work using Wine. Wine is free, so no license is needed. It is somewhat complicated to set up multple GPUs in Linux, but if I can do it, anyone can. You do not need a dummy plug or a dummy monitor to run multiple GPUs in Linux. I think that covers it...

All of this because you can't just not play games during the week? To each his own I suppose...

 

[W.Feather]

WINE also has a licensed version(dont remember what else is done....was given out because the owner made a promise about something with Gas prices, got one when they had the free licenses)....

and ya, well for the first few weeks i want to keep temptation away, in High School I never studied....that wont work now that i'm in the "Big Leagues"

 

[leagle]

WINE also has a licensed version(dont remember what else is done....was given out because the owner made a promise about something with Gas prices, got one when they had the free licenses)....

and ya, well for the first few weeks i want to keep temptation away, in High School I never studied....that wont work now that i'm in the "Big Leagues"

<THREAD DERAIL>

Based on my experience, you're really just treating a symptom and not the cause of the problem. If you want to be distracted, you can always find something to do that is more interesting than school work. The temptation of playing video games will just be replaced by something else. You'd be better off learning how to ignore or manage the temptations rather than creating artificial barriers.

What worked for me was to actually schedule my study time and to always do it in the library, away from distracting roommates and their infinite time wasting skills. If you make studying a part of your daily routine, then there's no reason that you can't play some games after your work is done.

</THREAD DERAIL>

 

[W.Feather]

thats why i'm trying to keep off of games(easiest way since ill be using my gaming computer for work, will be in Linux) for a few week to get a routine in place, then I go back to windows, just dont want to lose 2-3 weeks of folding(21 days to generate habit if i remember psych correctly), so to remove main temptation(facebook/interwebz browsing will be there regardless of place of study), ill jump to linux for the week....

though I do see your point, i never had a true routine in high school, its time to start it.

 

[Mtnduey]

all of my 64bit hardware is running Linux (CentOS 5.4) but I also have a few 32 bit 8 core machines running windows SMP (win 2k3)

I'm a Unix/Linux guy by trade so it's pretty easy for me to manage, but I've tested all of my machines out on both windows and linux. Without a doubt my machines turn faster frame times on Linux than they do on Windows.

When Linux Bigadv was out I could get a WU complete with about 60% time remaining, yet the same number of steps in the WU under Windows Bigadv and I barely completed the WU in time to qualify for bigadv.

So I really think it comes down to how old/new your hardware is. Linux SMP vs Windows SMP on older hardware isn't even a contest from what I've seen. Linux SMP just destroys it.

Granted I'm seeing some really favorable times out of my new toy as mentioned in another thread here (again Linux SMP). Getting 1min 20sec frames on a 6056 WU, 3min 30sec frames on a 6701 WU, and 3min 59sec on a 6041 WU.

How does that compare to Windows? I have no idea, I dont have time to test that new toy with Windows to see how it performs.

 

[APOLLO]

all of my 64bit hardware is running Linux (CentOS 5.4) but I also have a few 32 bit 8 core machines running windows SMP (win 2k3)

I'm a Unix/Linux guy by trade so it's pretty easy for me to manage, but I've tested all of my machines out on both windows and linux. Without a doubt my machines turn faster frame times on Linux than they do on Windows.

Then you're the only one thus far that I have seen, but the qualifier here is Linux distro and/or kernel. I found this to be a major determinant for performance in folding. Whereas everyone has stated superior performance with A3 in Windows, my experience has been just about par with only a slight edge towards Windows. You're the first person who has reported inverse to the consensus, and lends credence to my supposition that Linux flavors can indeed make a difference.

When Linux Bigadv was out I could get a WU complete with about 60% time remaining, yet the same number of steps in the WU under Windows Bigadv and I barely completed the WU in time to qualify for bigadv.

Likewise, what hardware are you using that is equipped with 8 or more cores? We are probably running similar platforms judging from our parallel experience with original Linux -bigadv vs A3 -bigadv.

So I really think it comes down to how old/new your hardware is. Linux SMP vs Windows SMP on older hardware isn't even a contest from what I've seen. Linux SMP just destroys it.

Hate to sound like a broken record, but it is my conviction Stanford optimized A3, especially -bigadv A3 for Nehalem, namely the superior memory subsystem inherent thereof. If you are running pre-Nehalem or even AMD hardware, you're SOL like I am and anyone else that has non-Nehalem yet compatible hardware.

 

[W.Feather]

specs i'm running are in my sig


though i'm not concerned with the points, i just want to fold with the machine, its doesnt do anything else when i'm not here......also my question pertains more towards the graphics folding within linux rather than CPU, since I know the CPU will regardless, having 1-2 GFX sitting idle just chills my bones

 

[Zero82z]

Hate to sound like a broken record, but it is my conviction Stanford optimized A3, especially -bigadv A3 for Nehalem, namely the superior memory subsystem inherent thereof. If you are running pre-Nehalem or even AMD hardware, you're SOL like I am and anyone else that has non-Nehalem yet compatible hardware.

-bigadv perhaps, but as far as normal A3 units go, my Q9550 still performs very well with them. Now that it's not my main system any more, I see 8-10k PPD most of the time (excluding 6701/6702), which is not far off from a similarly-clocked i5 quad or i7 with HT disabled.

 

[APOLLO]

-bigadv perhaps, but as far as normal A3 units go, my Q9550 still performs very well with them. Now that it's not my main system any more, I see 8-10k PPD most of the time (excluding 6701/6702), which is not far off from a similarly-clocked i5 quad or i7 with HT disabled.

My systems also perform appreciably well with standard A3, I really can't complain. But, with 8 available cores, they're screaming for much more and therein lies my complaints with Stanford..

One thing that is different between 775 and 771 is the caching algorithm used in the Xeons apparently being different if my memory serves. I could be mistaken about the specifics (too lazy to check it now) but something is different between the two processor series besides multi-processor support and more stringent testing in the Xeons. That could explain some performance results.

 

[W.Feather]

well I opened a new thread to address all my questions, and since this one seems to be solved(just use WINE),

/closed

 

[Mtnduey]

Then you're the only one thus far that I have seen, but the qualifier here is Linux distro and/or kernel. I found this to be a major determinant for performance in folding. Whereas everyone has stated superior performance with A3 in Windows, my experience has been just about par with only a slight edge towards Windows. You're the first person who has reported inverse to the consensus, and lends credence to my supposition that Linux flavors can indeed make a difference.

Likewise, what hardware are you using that is equipped with 8 or more cores? We are probably running similar platforms judging from our parallel experience with original Linux -bigadv vs A3 -bigadv.

Hate to sound like a broken record, but it is my conviction Stanford optimized A3, especially -bigadv A3 for Nehalem, namely the superior memory subsystem inherent thereof. If you are running pre-Nehalem or even AMD hardware, you're SOL like I am and anyone else that has non-Nehalem yet compatible hardware.

all of my linux rigs run in console only mode (no gui), selinux/iptables off, anything non-essential turned off, etc...

I have several 4x dual core xeon 3.2ghz machines running so thats 8 cores and 16 threads on these boys, they don't turn out the best times in the world but they were free machines so I'll take it. They were linux bigadv capable which is where I started folding back in late April. When I tried installing w2k3 and 2k8 on these to see how they ran it was kinda shameful the frames they turned out. Win bigadv was 1hr per frame when Linux bigadv was about 30 on the same box. Tried -smp only on the same boxen and frames were again about double what I saw on linux.

So your thought that F@H A3 is optimzed for Nehalem isn't far off at all it seems. I'm still waiting for my dual 12 core 2.3ghz AMD rigs to show up so I can put them through their paces. Maybe then I can get a decent Intel vs AMD showdown

I also have a couple 3ghz 4 proc dual core machines w/o HT but they're 32bit hardware, thus running winders as 32bit linux isnt supported for F@H. I have a 2 proc low-energy Intel quad 2ghz (socket 771 era) that is running some decent frams in linux. Thats the bulk of my folding stack but I have a couple of newer nehalem based rigs that are crunching away under Linux. These are some special rigs with no usb ports or dvd/cd anything so Linux is my only option right now as I can only install the OS by PXE boot (kickstart). I'm going to see if I can crack one of these open and see if I can find any interal connectors to maybe hook up a sata dvd or something.

then there is the "toy" but I only have that a few days, might try installing win2008r2 tomorrow to see how it runs with a proper winOS just to give a side by side.

I've got some AIX, Solaris and misc Unix boxes laying about with no workload, I might try installing wine on them to see if I can get Win SMP to work on those. Would be interesting if it worked.

 

[APOLLO]

all of my linux rigs run in console only mode (no gui), selinux/iptables off, anything non-essential turned off, etc...

When I was running Linux I had a stripped down version of Ubuntu using Fluxbox. Very light distro, probably the lightest 64-bit Ubuntu-based distro with a GUI.

I have several 4x dual core xeon 3.2ghz machines running so thats 8 cores and 16 threads on these boys, they don't turn out the best times in the world but they were free machines so I'll take it. They were linux bigadv capable which is where I started folding back in late April. When I tried installing w2k3 and 2k8 on these to see how they ran it was kinda shameful the frames they turned out. Win bigadv was 1hr per frame when Linux bigadv was about 30 on the same box. Tried -smp only on the same boxen and frames were again about double what I saw on linux.

OK, I remember your thread about these machines a few months ago. So, Socket 604 servers and they succeeded at -bigadv? 30 minutes TPF was better than anything my dual S771 machines were able to achieve in Linux, but they have a frequency under 3GHz and only 8 threads. The discrepancy you're seeing between Linux A2 and Windows A3 is even larger than I am experiencing, much larger. The older the architecture, the greater the percentage in performance discrepancy and in your case it's a 100% difference, effectively doubling your TPF...

So your thought that F@H A3 is optimzed for Nehalem isn't far off at all it seems.

It's what I believe but have not researched it to confirm my suspicions. What else can explain my previous near equivalent performance with i7 machines of 800-900MHz faster frequencies in Linux A2, and now the same machines are outperforming my systems by ~10 minutes TPF and better in Windows A3?? It seems extremely odd to me. The only situation these two architectures are comparable is with the P2684, where we both are pushing or exceeding the 50 minute mark. Otherwise there's a vast difference... Stanford stated the new A3 -bigadv WUs were designed with a much, much smaller memory footprint and this is what we have all seen. Did they opt to trade memory usage for certain optimizations based on architecture particulars? I do not know.

To a certain extent, I could agree with client architecture optimization especially when there's a new processor series released with a good future ahead of it in terms of life. The more mature F@H becomes however, the more older architectures get sidelined, with some hardware prematurely so, affecting more and more people as time goes by. This is what I fear could be the norm in the future where we may be witnessing a quickening of 'obsoleted' architectures for the simplicity and ease of the coders. I hope that's not the case.

A way out that undesirable scenario is to have different client settings to optimize performance for given hardware. This has 'sort of' been done recently with the GPU3 client to allow the processing of different WUs for older video cards. I think this would be a great idea for SMP as well, and might be accomplished with flags or the client config, but there has to be an interest and a will. Such options would come in very handy in the case of AMD hardware.

I'm still waiting for my dual 12 core 2.3ghz AMD rigs to show up so I can put them through their paces. Maybe then I can get a decent Intel vs AMD showdown

You're purchasing a couple of MC servers??

I also have a couple 3ghz 4 proc dual core machines w/o HT but they're 32bit hardware, thus running winders as 32bit linux isnt supported for F@H.

Unfortunately, the best you can do with them is run Windows standard A3. Not sure how worth it that is...

I have a 2 proc low-energy Intel quad 2ghz (socket 771 era) that is running some decent frams in linux.

What specific processor model, L series? I assume you didn't/cannot OC this machine? 2GHz is too slow for -bigadv with that architecture, at least the P2684 wouldn't make it. If you can OC it to the mid-2GHz frequency level you might be able to process some -bigadv WUs, but that's a big if...

Thats the bulk of my folding stack but I have a couple of newer nehalem based rigs that are crunching away under Linux. These are some special rigs with no usb ports or dvd/cd anything so Linux is my only option right now as I can only install the OS by PXE boot (kickstart). I'm going to see if I can crack one of these open and see if I can find any interal connectors to maybe hook up a sata dvd or something.

Yeah, SATA or IDE, anything that can connect to a drive, any kind of bootable drive would be your ticket.

I've got some AIX, Solaris and misc Unix boxes laying about with no workload, I might try installing wine on them to see if I can get Win SMP to work on those. Would be interesting if it worked.

Yes, that would definitely be interesting. Let us know how that goes.

 

[musky]

I've got some AIX, Solaris and misc Unix boxes laying about with no workload, I might try installing wine on them to see if I can get Win SMP to work on those. Would be interesting if it worked.

It does definately work now that you don't have the mpich piece with the WinSMP client. I would really like to see someone do this with a bigadv capable machine today. If I can get home in time this afternoon, I'll set up one or more of my machines with Linux + Wine + WinSMP + bigadv. Mine is all newer hardware, but it should give me a good Windows to Linux + Wine comparison.

 

[Mtnduey]

When I was running Linux I had a stripped down version of Ubuntu using Fluxbox. Very light distro, probably the lightest 64-bit Ubuntu-based distro with a GUI.

OK, I remember your thread about these machines a few months ago. So, Socket 604 servers and they succeeded at -bigadv? 30 minutes TPF was better than anything my dual S771 machines were able to achieve in Linux, but they have a frequency under 3GHz and only 8 threads. The discrepancy you're seeing between Linux A2 and Windows A3 is even larger than I am experiencing, much larger. The older the architecture, the greater the percentage in performance discrepancy and in your case it's a 100% difference, effectively doubling your TPF...

It's what I believe but have not researched it to confirm my suspicions. What else can explain my previous near equivalent performance with i7 machines of 800-900MHz faster frequencies in Linux A2, and now the same machines are outperforming my systems by ~10 minutes TPF and better in Windows A3?? It seems extremely odd to me. The only situation these two architectures are comparable is with the P2684, where we both are pushing or exceeding the 50 minute mark. Otherwise there's a vast difference... Stanford stated the new A3 -bigadv WUs were designed with a much, much smaller memory footprint and this is what we have all seen. Did they opt to trade memory usage for certain optimizations based on architecture particulars? I do not know.

To a certain extent, I could agree with client architecture optimization especially when there's a new processor series released with a good future ahead of it in terms of life. The more mature F@H becomes however, the more older architectures get sidelined, with some hardware prematurely so, affecting more and more people as time goes by. This is what I fear could be the norm in the future where we may be witnessing a quickening of 'obsoleted' architectures for the simplicity and ease of the coders. I hope that's not the case.

A way out that undesirable scenario is to have different client settings to optimize performance for given hardware. This has 'sort of' been done recently with the GPU3 client to allow the processing of different WUs for older video cards. I think this would be a great idea for SMP as well, and might be accomplished with flags or the client config, but there has to be an interest and a will. Such options would come in very handy in the case of AMD hardware.

You're purchasing a couple of MC servers??

Unfortunately, the best you can do with them is run Windows standard A3. Not sure how worth it that is...

What specific processor model, L series? I assume you didn't/cannot OC this machine? 2GHz is too slow for -bigadv with that architecture, at least the P2684 wouldn't make it. If you can OC it to the mid-2GHz frequency level you might be able to process some -bigadv WUs, but that's a big if...

Yeah, SATA or IDE, anything that can connect to a drive, any kind of bootable drive would be your ticket.

Yes, that would definitely be interesting. Let us know how that goes.

stuck in meetings all day so I may not be able to try out Wine until tomorrow. If this works out well it should be very fun in the coming months with AIX VIO server I can create an lpar, give it the lowest priority to CPU power but allow it to claim all available cpu cores and let it go to town. Would be interesting to see how the Sun UltraSparc T2+ cpu's with the massive multi-threadding work on a highly multi-thread capable workload like F@H

my low-energy CPU's are dual L5335, total power consumption around 110-120 watts which is pretty amazing considering the frame times it's capable of. (Linux 6701 9min 30sec frames)

sadly I can't OC it, bios is locked down pretty hard, about all I can do is change prefetch and a few other minor settings

as for the 2x 12core AMD rigs, yeah we're going to evaluate a new standard for our SQL clusters and this one promises a really nice price/performance ratio. SQL is more computational than it is transactional and thats the one place the AMD 12 cores really differ from the Intel 6/8-core's. I'm supposed to be getting two 2x12 core AMD's soon, depending on how that test goes we may try out the 4x12 core AMD rig just to see what it buys us.

 

[Nathan_P]

I'm supposed to be getting two 2x12 core AMD's soon, depending on how that test goes we may try out the 4x12 core AMD rig just to see what it buys us.

in terms of PPD a s***load, quad 12 cores has got to be good for 200k PPD at least

 

[sirmonkey1985]

in terms of PPD a s***load, quad 12 cores has got to be good for 200k PPD at least

less.. more in the range of 150-160k and requires 2 clients to be run for 48 threads.. for some reason the -bigadv WU's dont scale worth a damn past 26 or 27 threads so your best option is running 2 24 thread clients..

2x12 core 2.2ghz on A2 bigadv do about 67k PPD i believe it was.. something in that range..

 

[APOLLO]

stuck in meetings all day so I may not be able to try out Wine until tomorrow. If this works out well it should be very fun in the coming months with AIX VIO server I can create an lpar, give it the lowest priority to CPU power but allow it to claim all available cpu cores and let it go to town. Would be interesting to see how the Sun UltraSparc T2+ cpu's with the massive multi-threadding work on a highly multi-thread capable workload like F@H

Wow, I wish I had access to that kind of hardware and wish I had your Linux experience. That's one thing I wanted to be for over a decade, a Linux geek. I love Linux but every time I work in it I feel like such dummy...

my low-energy CPU's are dual L5335, total power consumption around 110-120 watts which is pretty amazing considering the frame times it's capable of. (Linux 6701 9min 30sec frames)

sadly I can't OC it, bios is locked down pretty hard, about all I can do is change prefetch and a few other minor settings

It's too bad they are 1333MHz FSB processors, otherwise you could have performed a BSEL mod for a higher frequency. Is this a brand name server, BTW?

as for the 2x 12core AMD rigs, yeah we're going to evaluate a new standard for our SQL clusters and this one promises a really nice price/performance ratio. SQL is more computational than it is transactional and thats the one place the AMD 12 cores really differ from the Intel 6/8-core's. I'm supposed to be getting two 2x12 core AMD's soon, depending on how that test goes we may try out the 4x12 core AMD rig just to see what it buys us.

 

[Mtnduey]

Wow, I wish I had access to that kind of hardware and wish I had your Linux experience. That's one thing I wanted to be for over a decade, a Linux geek. I love Linux but every time I work in it I feel like such dummy...

It's too bad they are 1333MHz FSB processors, otherwise you could have performed a BSEL mod for a higher frequency. Is this a brand name server, BTW?

*Nix in general is where I have fun. My brain works well in a CLI world which is kind of funny as I started my IT career long ago supporting windows

The Low-E rig is a Dell 1950 III (L) and while they said 1333 was the default build, thats not what they shipped us, so I got to keep their mistake as a toy for my lab. Aside from F@H its also running Cacti/Nagios monitoring for the lab and so far I'm pretty happy with it.