Linux versus Windows - bigadv folding results

I'd stick with the enterprise-level CentOS or openSUSE distros... Leave Ubuntu for the desktop pc's (Not to bash Debian as I love it as a server OS too, just not for folding). :)
 
Some more TPF info

my slowest rigs running old school quad Intel(R) Xeon(TM) CPU 3.20GHz (like seriously old school, just barely makes the 3 day window)

Project 6901 on 3x above rigs:
42 min 33 sec
43 min 15 sec
43 min 20 sec

I'll prob switch these back to normal smp after their bigadv posts
 
I'm getting about 31 min 15 sec frames now on my old dual e5450's (3.0 GHz no HT, 12x512MB memory) on CentOS 5.5. Well within the preferred time and much better PPD @ ~35k as compared to the ~18k I was getting with SMP.

[10:13:03] Project: 6901 (Run 22, Clone 14, Gen 1)
...
Starting 8 threads
...
[10:13:13] Completed 0 out of 250000 steps (0%)
[10:44:29] Completed 2500 out of 250000 steps (1%)
[11:15:43] Completed 5000 out of 250000 steps (2%)
[11:46:58] Completed 7500 out of 250000 steps (3%)
 
Last edited:
For the AMD doubters around here - and those with either cash to burn or a system like 10e's. Ths is the true power of 48 cores of opteron folding goodness.:eek:

tpf of 06:40 per frame

You really do not want to know what the PPD is on that beast:eek::D:cool:
 
DAMMMM thats a nice TPF


post above it caught my eye....1090T @3.9 saying 28:56 TPF for a 6901.....
 
Nathan_P said:
For the AMD doubters around here - and those with either cash to burn or a system like 10e's. Ths is the true power of 48 cores of opteron folding goodness.:eek:

tpf of 06:40 per frame

You really do not want to know what the PPD is on that beast:eek::D:cool:
Click to expand...
You're right, I probably don't...

W.Feather said:
post above it caught my eye....1090T @3.9 saying 28:56 TPF for a 6901.....
Click to expand...
Does anyone have an idea why AMD architecture would benefit so greatly under Linux and the new client? This seems to me like a huge performance boost rabbit out of a hat, unless I was completely unaware of advances in that direction lately being stuck on Intel -bigadv all this time. :confused:
 
W.Feather said:
from the post i refer to:



....a AMD Phenom II X6 1090T @3.9 Ghz, getting 28:56 TPF on a 6901 ;) in linux
Click to expand...
yeah I know. sorry I wasn't clear that the post you referred to confused me. Not your post.
 
Nathan_P said:
For the AMD doubters around here - and those with either cash to burn or a system like 10e's. Ths is the true power of 48 cores of opteron folding goodness.:eek:

tpf of 06:40 per frame

You really do not want to know what the PPD is on that beast:eek::D:cool:
Click to expand...


The 4 socket x 8 cores +HT all at 2 GHz runs in the 8's which I thought was pretty good but that is impressive.
 
jebo_4jc said:
yeah I know. sorry I wasn't clear that the post you referred to confused me. Not your post.
Click to expand...

my question is, what did he do to get big adv to run on linux with a 1090T ;)

.....if its allowed fully maybe i need to speed up moving to linux as a primary OS soon
 
W.Feather said:
my question is, what did he do to get big adv to run on linux with a 1090T ;)

.....if its allowed fully maybe i need to speed up moving to linux as a primary OS soon
Click to expand...

No its not allowed - there have been several threads around the boards where people have tricked Linux into reporting 8 CPU threads to the client so as to enable -bigadv, but Kasson and Vijay have said that they would prefer people not to do this.
 
APOLLO said:
You're right, I probably don't...

Does anyone have an idea why AMD architecture would benefit so greatly under Linux and the new client? This seems to me like a huge performance boost rabbit out of a hat, unless I was completely unaware of advances in that direction lately being stuck on Intel -bigadv all this time. :confused:
Click to expand...

I think AMD is scaling as well as Intel - those frame times look proportionally better just like intel systems are seeing with moving to Linux64. I think we are seeing the benefit of 64bit more than anything else.

Due to the compiler used with FAH - AMD does not use SIMD instructions when folding, so AMD are not getting their full potential.
 
Hi, in the interest of the science and cordiality between teams (I really never under why there's animosity sometimes), I'm popping over from evga to share my results after MIBW posted in my thread. Right know it's purely speculative but disabling these guys:

Bluetooth
Hardware drivers checks
disk notifications
evolution alarm
power manager
print queue
remote desktop
Ubuntu One
Update Notifier

seemed to help somewhat, but right now what's working, (about 50 consistent frames so far) is setting the FAHcore_a5.exe set to a nice value of -5, having the Console and the system monitor (on the Resources page) open and on the active desktop, and MOST importantly NOT TOUCHING anything. If I just let it sit there it seems very content to fold at 100% load.


Anyway, Hi. And if I get anything conclusive I'll post here and at evga

Edit: I'm running ubuntu 10.1, a 980X @ 4.0, Ram is @ 1600 8-8-8-21, and the only thing I installed on the system were drivers for my graphics card.
 
Last edited:
MIBW,

You might want to look at OpenSuSE - I've got it on a 12 core machine, its an easy install and it just works. My Frame times over 2 WU's have been +- 5 seconds of one another.

Or you could send me one of those SR-2's to test with. ;)

H.
 
hspeirs said:
MIBW,

You might want to look at OpenSuSE - I've got it on a 12 core machine, its an easy install and it just works. My Frame times over 2 WU's have been +- 5 seconds of one another.

Or you could send me one of those SR-2's to test with. ;)

H.
Click to expand...

Sure - just the one? :D

Thanks for the tip - Trying to install openSUSE-11.3-GNOME-LiveCD-x86_64.iso to a USB stick as we speak, and UNetbootin is making a hash of it...
 
Well, if you want to send more than one ...... ;)

Never tried the live CD from a USB, I normally just throw in a scratch HD, install a minimal install (with a GUI desktop so I have a browser), init 3 to text mode, and let it rip.

H.
 
Did I mention the $10,000 postage and special handling charge per box? Perhaps I forgot.

I gave up on the USB and am installing to an old HDD now. PITA.
 
Nathan_P said:
For the AMD doubters around here - and those with either cash to burn or a system like 10e's. Ths is the true power of 48 cores of opteron folding goodness.:eek:

tpf of 06:40 per frame

You really do not want to know what the PPD is on that beast:eek::D:cool:
Click to expand...

CentOS is being installed tonight.

If I extrapolate Tear's numbers to my slower CPUs I will still see 7.71 minute or 7:43 TPFs on p6901s which still translates to some ungodly PPD number (which I will reserve listing due to my fellow Canadian's disgust :) ) He shall go un-named :)

Anyway I'm gonna try this on SkankBox 48 © first and then the SR2 to see what's going on in terms of performance upgrades for BiGadv. I am hoping to eek out something and see if NUMA on/off makes any differences. I saw Tear on the FF thread explain potential improvement based on the use of alternative schedulers.

If I see success on the SR2 I will definitely explain exactly how I installed CentOS or Gentoo on it for others to utilize to their advantage for team 33, including the FTFDU. Then until the CC2011 I'll be providing some D for 33. I was gonna do that anyways but the increase in performance on the 48-core might just drive me back sooner ;)

What's scary with this scenario is based on the performance improvements on AMD and potential improvements on the SR2 I could see 400K PPD from two boxes up from about 320.....

Yowza.

I'll be back to list what I'm seeing. Hopefully my old Solaris skills can contribute to my Linux knowledge in this arena.
 
10e said:
CentOS is being installed tonight.

If I extrapolate Tear's numbers to my slower CPUs I will still see 7.71 minute or 7:43 TPFs on p6901s which still translates to some ungodly PPD number (which I will reserve listing due to my fellow Canadian's disgust :) ) He shall go un-named :)

Anyway I'm gonna try this on SkankBox 48 © first and then the SR2 to see what's going on in terms of performance upgrades for BiGadv. I am hoping to eek out something and see if NUMA on/off makes any differences. I saw Tear on the FF thread explain potential improvement based on the use of alternative schedulers.

If I see success on the SR2 I will definitely explain exactly how I installed CentOS or Gentoo on it for others to utilize to their advantage for team 33, including the FTFDU.

What's scary with this scenario is based on the performance improvements on AMD and potential improvements on the SR2 I could see 400K PPD from two boxes up from about 320.....

Yowza.

I'll be back to list what I'm seeing. Hopefully my old Solaris skills can contribute to my Linux knowledge in this arena.
Click to expand...

That gives you about 308k if you are running 1.9ghz chips:D
 
now if HP could just fix my bloody dual 12 core servers. They're brand new and only recognizing a single CPU...grrr
 
Nathan_P said:
in your case PPD should be ~138k:)
Click to expand...


Hmm have been getting ~105-110k on a3 ...

but I did just reboot and turn numa off lol...

need to scavenge to get quad channel up and running... currently 4x4gb sticks...
 
Mtnduey said:
now if HP could just fix my bloody dual 12 core servers. They're brand new and only recognizing a single CPU...grrr
Click to expand...

What server?

I have 385s and 165s with dual 12 cores

ram in the right slots?
reset bios on add?
 
Patriot said:
Hmm have been getting ~105-110k on a3 ...

but I did just reboot and turn numa off lol...

need to scavenge to get quad channel up and running... currently 4x4gb sticks...
Click to expand...

Well, i can only compare apples to apples, the way my calculator works is it takes a baseline machine with a certain PPD and can then calculate for different core count/Freq. It takes account of nothing else so if the baseline machine is a linux box on A5 you get Linux A5 PPD, same for win A3 etc.
 
BL465c

They're brand new out of the box and it's only seeing one cpu in bios. We've reset everything, they've even replaced the CPU's with no luck, next they're going to try memory and system board.

Brand...new.... dual 6176's.....dieing to see what the 2.3ghz 24 cores can do

*sigh*
 
Mtnduey said:
BL465c

They're brand new out of the box and it's only seeing one cpu in bios. We've reset everything, they've even replaced the CPU's with no luck, next they're going to try memory and system board.

Brand...new.... dual 6176's.....dieing to see what the 2.3ghz 24 cores can do

*sigh*
Click to expand...

I have a pilot of one of those... I will play with it tomorrow...
no reason why that wouldn't work...

power capping in the enclosure?

Fans / power supplies in the enclosure?
 
Mtnduey said:
BL465c

They're brand new out of the box and it's only seeing one cpu in bios. We've reset everything, they've even replaced the CPU's with no luck, next they're going to try memory and system board.

Brand...new.... dual 6176's.....dieing to see what the 2.3ghz 24 cores can do

*sigh*
Click to expand...

If ur running linux i could tell you and really spoil ur fun. :p
(no I don't have the hardware b4 you all ask:eek:)
 
power isnt an issue. We have dual 3phase drops on each c-7000

we went through and did the ilo updates on the test enclosure to make sure that wasnt an issue, still no joy.

HP tech is stumped. Just sucks as we have 5 of these and only one of them works
 
Mtnduey said:
power isnt an issue. We have dual 3phase drops on each c-7000

we went through and did the ilo updates on the test enclosure to make sure that wasnt an issue, still no joy.

HP tech is stumped. Just sucks as we have 5 of these and only one of them works
Click to expand...


no... im talking about the dynamic power capping in the onboard administrator console...

you can limit the power draw... regardless of what you have available...

I doubt thats the issue but...
 
Tried OpenSuSE last night - same poor results, but different! About half the frames were ok, but a quarter were wildly worse.

Watching the taskmanager, at least there were short periods where CPU was actually at 100%.

(Dual socket SR2 - 12 cores -24 threads)
openSUSE-11.3-GNOME-LiveCD-x86_64.iso , disabled a ton of services... tried NUMA on/disabled, enabled c-states...

6701 frame times:
original.jpg
 
I have seen a fluctuation in frame times as well on rhel5u6 ...
 
may i be worthy said:
Tried OpenSuSE last night - same poor results, but different! About half the frames were ok, but a quarter were wildly worse.

Watching the taskmanager, at least there were short periods where CPU was actually at 100%.

(Dual socket SR2 - 12 cores -24 threads)
openSUSE-11.3-GNOME-LiveCD-x86_64.iso , disabled a ton of services... tried NUMA on/disabled, enabled c-states...

6701 frame times:
original.jpg
Click to expand...

Damn...I downloaded that before I left for work this morning and was going to play with it when I got home.

I guess Ubuntu 9.04 Desktop will be my next try - folded with I7 920's @ 100% load for several months with that release, and it picked up something that 8.043 LTS did not on the X58 boards (although I think it was something silly like a wireless card that I didn't need anyway.)
 
I will try a rhel5u4 tomorrow as I already have machines running 5u5 and 5u6
 
You must log in or register to reply here.
Back
Top