• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Into the gap

Folding OS

  • Total voters
    42
P

Patriot

[H]ard|DCer of the Month - March 2011/June 2013/De
Joined
Dec 15, 2010
Messages
2,502
The gap between hard and evga now stands at 4m

Who is already using linux... who is not...

Who wants to but needs help...

Basically depending on the setup... we have seen 20k ppd gains for the a5 core running on opensuse vs windows 7
that is amd dual hexacores at 2.6

if our top 20 switches the dedicated boxes to linux I think we can trim down on that gap significantly...

betateam linux even more so.
 
I'm trying...and failing...

The second SR-2 isn't liking Ubuntu Server at all. I just backed down the clock and upped the voltage to hopefully allow it to get past 2 frames.

The first SR-2 is back up to 13 minute frames under Ubuntu Server - the same as Windows.

I am not sold on Linux being the best way to go yet. Right now, it is costing me more ppd than it is gaining me.
 
I have Linux experience and I kept the VMs on most of my systems in case I ever wanted to use them again. Problem is my systems are older than most people are running -bigadv on this team and the gains will not be substantial. For 1-2 min TPF drop I won't bother going through the hassle of restarting them. What are we looking at in terms of percentage of TPF reduction for a best case scenario?
 
musky said:
I am not sold on Linux being the best way to go yet. Right now, it is costing me more ppd than it is gaining me.
Click to expand...
This is the type of thing I'm worried about.

seems like Linux bigadv is still in flux to me. I'm still not sure if units are turning in correctly, or if certain distros are vastly superior to others, or what.




 
Tell yall what... those that dont mind the lost ppd... play around now...

I will try a few different distros and compare tpf differences...

Opensuse is on the list... though I will probably use sles so I have the drivers I need... ugh hate sles ...

will see about getting ubuntu on a box... we don't make drivers for debian.. but just hacked one into existence for a special case a few weeks back... might as well see if it works...
 
jebo_4jc said:
seems like Linux bigadv is still in flux to me. I'm still not sure if units are turning in correctly, or if certain distros are vastly superior to others, or what.
Click to expand...
I can't tell you about security of the new client in Linux, but there is a difference between distros. AAMOF, I believe I was the first to mention on this forum about distro differences and -bigadv back in '09. There was always a significant performance difference between distros specifically because they have different kernels installed. Even between the same distros there could be a significant performance gap if they have different kernels. Tons of variables in Linux compared to Windows.
 
Let's get some good data from Patriot, Mtnduey, etc. through the week - guys who know what they are doing - before we unleash a bunch of amateurs on Linux and turn this into a Linux support board. And yes, I am one of the amateurs...no disrespect meant to anyone.
 
APOLLO said:
I can't tell you about security of the new client in Linux, but there is a difference between distros. AAMOF, I believe I was the first to mention on this forum about distro differences and -bigadv back in '09. There was always a significant performance difference between distros specifically because they have different kernels installed.
Click to expand...

hopefully I can figure out which is the best kernel...
I mean you can install erratas on whatever distro you want if we can figure out which kernel is the performance one...
 
Patriot said:
hopefully I can figure out which is the best kernel...
I mean you can install erratas on whatever distro you want if we can figure out which kernel is the performance one...
Click to expand...
Unfortunately, my experience in Linux is almost exclusively folding related and therefore my knowledge is perfunctory at best. I'll just accept whatever your findings are. You appear to have great depth of knowledge about Linux that I and most here lack.

And musky, duly noted.
 
APOLLO said:
Unfortunately, my experience in Linux is almost exclusively folding related and therefore my knowledge is perfunctory at best. I'll just accept whatever your findings are. You appear to have great depth of knowledge about Linux that I and most here lack.

And musky, duly noted.
Click to expand...

I got my job because if it... was funny... a guy struck up a conversation with me at a social... started talking about his myth tv box and how he was messing with certain things...

well I had gotten into linux about 6mo prior having built my first pc and not wanting to pay the windows tax was coasting on xp 64 trials while trying madly to figure out linux...

started with fedora core 4... hellish place to begin...

anyhow... my knowledge was how to get stuff working for video and stuff... and that was what this guy was lacking... so me an amateur in linux unwittingly helped out a guy who had been working in enterprise linux for the past 15+ years... and unix before that...

I had the habit of picking out a new distro every weekend and messing with it... blowing it away and putting another on...

had messed with about 30 different distros from all the major branches... Know what I like and what I don't...

Anyhow... the guy asked for my email...then sent me a link to a job opening at hp... I have been an intern in the lab he manages for 3.6 years now...

first day of work... they brought me into the lab... showed me a few command line instructions... where to lftp the drivers to install from... and told me he would be if I needed help and left.

dropped in the deep end... and I learned to swim... learned bash, shell, perl and php at work and C++ in school that semester... my head was swimming... all nominally of course... I can code what I need in it... know enough to figure out how to learn what I don't...

of the past few years I have not messed with nearly as much other distros... mainly just RHEL (red hat enterprise linux) and SLES (suse linux enterprise server) ... for desktop I have liked ubuntu offshoots... due to the just mostly works nature... but they have gradually gotten more bloated and dummed down (read user friendly) ...
of course not to the point of castration (read OSX)

Its time for me to start dabbling again... back when I first started I was stumbling in the dark typing in whatever commands I read in the forums not knowing what the hell they did... now I actually am proficient...

I have never made my own distro... I know its not terribly hard but nor is it easy...

I figure if I can find the fastest distro... get that kernel from kernel.org...
take my favorite windows manager or distro with it... and recomplie with the speed kernel.... we have a HardFolding linux distro...
 
I am more than happy to run Linux on my 2 slaves, but I would need some serious support.

One more wrinkle into an already complex issue is the forthcoming changes in Gromacs/FAH for NUMA support, and/or 64Bit. These changes alone can likely obsolete an OS switch depending on how well applied across platforms they are.

I think this might be a bit of a moving target for a while - all the more reason for [H] to co-ordinate an effort so we aren't all reinventing the wheel.
 
may i be worthy said:
I am more than happy to run Linux on my 2 slaves, but I would need some serious support.

One more wrinkle into an already complex issue is the forthcoming changes in Gromacs/FAH for NUMA support, and/or 64Bit. These changes alone can likely obsolete an OS switch depending on how well applied across platforms they are.

I think this might be a bit of a moving target for a while - all the more reason for [H] to co-ordinate an effort so we aren't all reinventing the wheel.
Click to expand...



ou numa support... yeah that might mix things up... might completely screw the kernel speed differences over as well...
 
until i can play every game natively on linux i'll be sticking with windows.. i use to be a hardcore linux person during the RTCW/RTCW:ET days. but when ubuntu came out and basically destroyed the custom distro community i lost a lot of interest in linux after that.
 
I was the one with the 20K ppd increase on the dual hex core AMDs. Went from 32K to 52K just by switching it to OpenSuSE. tpf went from 30:22 to 22:02

Edit: Correction: Went from 37K to 59K

I'd been trying to tweak settings because Patriot was getting significantly better times than me, on slightly slower proc's. Tweaking NUMA settings shaved a couple of minutes off, switching from W2K8 R2 on a SSD to OpenSuSE on a laptop drive dropped my times by over 8 minutes per frame.

I've been trawling through the tpf's being posted over on foldingforum and there does seem to be an improvement when switching to Linux. I've been looking to match similar (identical when possible) hardware comparing 6900 to 6901.

Under Windows the times are pretty much identical, under Linux the times are significantly faster. ie A 3.8GHz i7-930 on linux gets around 38K with linux, vs a 4.1 i7-930 getting 31K under windows.

I've only three data points so far, but they are all consistent - A5 under linux outfolds Windows.

i7-2600K, @4.6GHz 28:11 Windows, @4.6Ghz 22:45 Linux, 41.1k ppd/56.7k ppd
i7 930, @4.1GHz 34:11 Windows, @3.8GHz 29:53 Linux, 30.8k ppd/37.7k ppd
Opteron 4180, @2.6GHz 30:22 Windows, @2.6GHz 22:02 Linux, 36.8k ppd/59.5k ppd

H.
 
Last edited:
yeah wine is almost always a dx version behind... I hear ya... but for dedicated folders ... if the performance is there....

but for some of you guys... it has to support gpu as well...
 
Patriot said:
ou numa support... yeah that might mix things up... might completely screw the kernel speed differences over as well...
Click to expand...

AFAIK the NUMA code has been submitted to Gromacs - not just fixing the deficit, but getting even better results with NUMA enabled. How far this is from shipping from Gromacs, and then making it into FAH is anyone's guess. ie, maybe 2 months, maybe a year.

But we are talking ballpark 20% better frame times on multi socket systems with the NUMA fixes, 64 bit, and other tweaks combined.

EDIT: Sounds to me like the 64bit linux might be the key here.

http://foldingforum.org/viewtopic.php?f=55&t=17695&start=15#p176152
 
I'd be willing to switch over my SR-2 rig to Linux, but would also need some pretty serious help doing it. I messed with some distros in VMs for folding, but it took me forever and a day of googling to get everything running properly. If ya'll can figure out the best distro and write a How-To guide, that'd be great. I'd love to get a boost in PPD.
 
Linux is not feasible for me. It's unlikely that I will ever start using it again on my boxes.
 
Whats the min tpf you need to meet the preferred deadline on bigadv? I'm willing to play around, but I'm 90% sure my 4122s won't make the preferred deadline.
 
Vediovus said:
Whats the min tpf you need to meet the preferred deadline on bigadv? I'm willing to play around, but I'm 90% sure my 4122s won't make the preferred deadline.
Click to expand...
57:36 is the threshold (the preferred deadline is four days, so 4*24*60/100).
 
Zero82z said:
57:36 is the threshold (the preferred deadline is four days, so 4*24*60/100).
Click to expand...

I have dual 2ghz no ht xeons make the time frame just fine... I see no reason why your dual 2.2amd quads couldn't
 
Alright, I'll give it a go in the morning then. its far too late for me to be thinking about compiling kernels.
 
APOLLO said:
Which Xeon socket and do you mean P690x WUs?
Click to expand...

1366... and even with 2684s they still make the cut... at a bloody slow 47min... better wu nets 10min faster...
 
Patriot said:
1366... and even with 2684s they still make the cut... at a bloody slow 47min... better wu nets 10min faster...
Click to expand...
Now someone tells me. I had several opportunities the past year of netting low-end S1366 non-HT Xeons that I could have slapped on a used, cheap server board. Oh well... :eek:
 
APOLLO said:
Now someone tells me. I had several opportunities the past year of netting low-end S1366 non-HT Xeons that I could have slapped on a used, cheap server board. Oh well... :eek:
Click to expand...
Well, based on those TPFs, you'd have gotten the same performance out of a single-socket LGA1366 system with an overclocked quad-core i7. I don't see much point in spending more money for and using more power with a dual quad-core system when you can get the same performance much more easily.
 
Zero82z said:
Well, based on those TPFs, you'd have gotten the same performance out of a single-socket LGA1366 system with an overclocked quad-core i7. I don't see much point in spending more money for and using more power with a dual quad-core system when you can get the same performance much more easily.
Click to expand...

2 60w chips...
ht gets you a good lil boost like the equive of 100mhz extra

and any good wu will have closer to a 37min tpf... I somehow think a 930 overclocked on the extreme would use more power... considering it starts at a higher tdp than the two xeons combined...
 
Zero82z said:
Well, based on those TPFs, you'd have gotten the same performance out of a single-socket LGA1366 system with an overclocked quad-core i7. I don't see much point in spending more money for and using more power with a dual quad-core system when you can get the same performance much more easily.
Click to expand...
Because I ONLY build multi-socket systems and the prospect of dropping in two hex-cores with the later possibility of upgrading to an SR-2, would have been very tempting even though I will mostly likely terminate F@H in the near future. That could have served as my main system seeing that I'm still using a 6 year old setup that is wearing long in the tooth. Now those opportunities are no longer there, and so is the incentive quite frankly.
 
Patriot said:
2 60w chips...
ht gets you a good lil boost like the equive of 100mhz extra

and any good wu will have closer to a 37min tpf... I somehow think a 930 overclocked on the extreme would use more power... considering it starts at a higher tdp than the two xeons combined...
Click to expand...
TDP doesn't mean much. It's pretty unlikely that those two CPUs use less power than a 920 or 930, even when overclocked. As far as TPFs go, my 920 at 4GHz does between 34 minutes and 36 minutes on non-2684 units, and that's with a GPU client running at the same time.
APOLLO said:
Because I ONLY build multi-socket systems
Click to expand...
Well, that's a personal choice, but not necessarily a good one as far as F@H is concerned.
 
Zero82z said:
Well, that's a personal choice, but not necessarily a good one as far as F@H is concerned.
Click to expand...
Only with the power usage concern and to that I agree for the majority. There were opportunities, several actually, that would have made it a win-win in my case. Even more so when looking at things retrospectively and the sequence of events that played out last year. The ES hex-cores were available several times in the FS section alone, and for relatively cheap. I could have had a dual hex-core system even if the board was a standard server offering. It could have totally changed my production and future outlook if I was better informed.
 
Zero82z said:
TDP doesn't mean much. It's pretty unlikely that those two CPUs use less power than a 920 or 930, even when overclocked. As far as TPFs go, my 920 at 4GHz does between 34 minutes and 36 minutes on non-2684 units, and that's with a GPU client running at the same time.

Well, that's a personal choice, but not necessarily a good one as far as F@H is concerned.
Click to expand...

I have my low end units deprived of ram bandwidth... I pack in the ram to the high end hexacores...


ouuu... I might pull out that 2.66ghz single hexa I have folding and trade it off for a 2.8ghz I had loaned a friend ... if he can get his setup stable I will see about taking the 920 off his hands and sell it so I can build a dp system... anyhow... trading cpus might get me matching 2.8ghz hexacores :)
 
I'll be sticking with Win 2008 R2 Server on my -bigadv boxen. I may switch a few clients over to Linux for A3's again in the future, but until I see drastic improvements in the Linux -Bigadv frame times it'll have to wait.

...and for the record, I'd say there are many at EVGA who will switch to Linux as well if the -bigadv is that good.
 
I am back to one box running Linux - switched the second one back to Windows this morning after a failed 6901 and three straight painfully slow regular A3 units.

For you Linux folks, one thing I have seen a couple of times now on both Ubuntu Desktop and Ubuntu Server 10.10 - the F@H core was not using 100% processor. top was showing right at 2300%, so i am losing 1 core. Looking at the GUI system monitor, it looks like that unused core also bounces around, which is worse yet. I remember seeing this on Ubuntu Desktop 9.10 back when i had I7 920s running. 9.04 would fully load the processor, but 9.10 would not. Whatever causes that is a deal-breaker - on the one machine I have running Ubuntu the difference is 11:40/frame when fully loaded and 13:15/frame when not.

Also, be careful stopping and restarting a client. I bounced my Linux box this morning to try to get it to load all the processors again (it did), and lost my checkpoint and had to start over on the current unit.

Edit: The Linux box just slammed on the brakes and is back to over 13 minutes/frame. No idea what is going on with it.
 
Last edited:
musky said:
I am back to one box running Linux - switched the second one back to Windows this morning after a failed 6901 and three straight painfully slow regular A3 units.

For you Linux folks, one thing I have seen a couple of times now on both Ubuntu Desktop and Ubuntu Server 10.10 - the F@H core was not using 100% processor. top was showing right at 2300%, so i am losing 1 core. Looking at the GUI system monitor, it looks like that unused core also bounces around, which is worse yet. I remember seeing this on Ubuntu Desktop 9.10 back when i had I7 920s running. 9.04 would fully load the processor, but 9.10 would not. Whatever causes that is a deal-breaker - on the one machine I have running Ubuntu the difference is 11:40/frame when fully loaded and 13:15/frame when not.

Also, be careful stopping and restarting a client. I bounced my Linux box this morning to try to get it to load all the processors again (it did), and lost my checkpoint and had to start over on the current unit.

Edit: The Linux box just slammed on the brakes and is back to over 13 minutes/frame. No idea what is going on with it.
Click to expand...

That sucks...I did a reboot on mine as well (i7-950) when I was making some clock changes and lost the 6901 I was crunching on, so I reverted to standard A3's again. I also noticed later on in the run that I had a random core that was fluxing, and not processing on the WU at hand. It would swap off with a different core from time to time (i.e. core 6 would re-engage but core 7 would start going idle, and vice versa). Strange, as I even specified "SMP 8" in the parameters.
 
I've added the bigadv flag to my linux box. Right now its running Ubuntu Desktop, but I'm going to throw together a bare bones gentoo install on another machine while that crunches away. When the WU finishes, I'll swap drives and see if I can get a lower tpf.
 
Vediovus said:
I've added the bigadv flag to my linux box. Right now its running Ubuntu Desktop, but I'm going to throw together a bare bones gentoo install on another machine while that crunches away. When the WU finishes, I'll swap drives and see if I can get a lower tpf.
Click to expand...

Version 10.10? If it is any version after 9.04, could you do me a favor and check you CPU usage? I have never been able to get 9.10+ to fully load all cores. Thanks
 
798-799% everything is loaded up like it should be for me, but then again, I'm finishing a 6053 right now. I'll check again when it gets a bigadv.

edit: Yeah Ubuntu 10.10, 6901, and cpu usage is the same, using all 8 cores. This is my first Bigadv, so I wasn't entirely sure the correct way to pass the flags to the client, so I started it with -smp 8 -bigadv
 
Last edited:
Vediovus said:
798-799% everything is loaded up like it should be for me, but then again, I'm finishing a 6053 right now. I'll check again when it gets a bigadv.

edit: Yeah Ubuntu 10.10, 6901, and cpu usage is the same, using all 8 cores. This is my first Bigadv, so I wasn't entirely sure the correct way to pass the flags to the client, so I started it with -smp 8 -bigadv
Click to expand...

Vediovus, what CPU and clock speed are you running thi9s -bigadv on, and what kind of frame times are you seeing?
 
You must log in or register to reply here.
Back
Top