SR-2 Optimization Thread

then run windows, and linux in a VM (your PPD will still be higher this way than running it within linux)

You wont get a unit that will take over 2 days within windows (atleast to my knowledge) since the real big units will be in linux only, you will be pretty much getting only -smp units.
 
-bigadv on win 7 have been scarce for months, partly due to the large number of i7's, i5's and x6's that have been running -bigadv. So far i have not seen anyone post that they are now getting -bigadv on win 7 and i reckon that won't change until new projects are released


Yea! That would be a week and a half ago, or in other words; coming up Real Soon Now!

;) :) :)

I just have to give it a shot, I'll set up a temp machine for everyday uses, and let the SR2 fold Linux. I also just noticed another reply, on a VM solution to my issues.

Thanks to all you guys, one day I'll have more experience on this -bigadv stuff.
Bruce
 
Quick question on memory setup. For when the blck and divider does not match up values for the MCH strap, is it better to:

  • tighter memory settings with lower strap
  • looser memory settings with high strap
 
then run windows, and linux in a VM (your PPD will still be higher this way than running it within linux)

You wont get a unit that will take over 2 days within windows (atleast to my knowledge) since the real big units will be in linux only, you will be pretty much getting only -smp units.

W.Feather,

Any who might help me to "have a clue", feel free to step in.

******
I now have Ubuntu 11.10 running nicely in a VMBox, (Oracle), and the client download went fine, but when I tried to run -configonly it shut me down and said I had less than 2 CPUs and should run standard client. Wrong!

How do I arrange for the client to see 24 cores(threads actually), with this arrangement?

Seems I'm close to folding except for that. BTW, I tried using -smp 24 in place of -smp in the config command, but no dice.

TIA,
Bruce
 
This is embarrassing! :eek:

How do I find that readout?

My Ubuntu experience so far is little more than copying instructions from the folks who understand what they are doing.

Bruce
 
This is embarrassing! :eek:

How do I find that readout?

My Ubuntu experience so far is little more than copying instructions from the folks who understand what they are doing.

Bruce

cat /proc/cpuinfo from a terminal.
 
Apparently, I only have CPU 0. It got the speed as 3300 instead of 3600.

Bruce

cpuinfo.jpg
 
Looks like the VirtualBox VM is not set up correctly or the version Ubuntu does not have a kernel with SMP support.
 
Thanks for the help, I'll see what I can learn at Oracle, probably a forum there on the 'setup', and with luck a folder there.

Bruce
 
Zat a 402 Chivvy SB? :)

Thanks for the guidance guys, I have just loaded core a5 and started on a 6904. I think that is the object of my search. Now to get a monitor on this new VM.

Using Teamviewer to re-install VMBox twice is such a cumbersome process, I lost track and forgot to set CPU in the 'settings' a third time, after I finally got the V-drive size right, I got preoccupied worrying about my having botched the linux command details.

Now to see what the production will be!

Bruce
 
Good to hear you are up and going.

On a side note, yes it is a bored and stroked chevy LS1.
 
Quick question on memory setup. For when the blck and divider does not match up values for the MCH strap, is it better to:

  • tighter memory settings with lower strap
  • looser memory settings with high strap

Upgraded to DDR3-1600 and right off hand it looks like a huge difference from the DDR3-1333 I was running. Memtest definitely looks to be running faster. Here are my current settings:

blck: 195
mem div: 8:1
uncore: 16x
mem timings: auto everything.

Effective memory speed of 1560Mhz.

mch @ 1333 = mem timings of 9-9-9-24
mch @ 1600 = mem timings of 11-11-11-29

Which mch value will result in better performance?

According to the Kingston spec, the ram should have timings of 9-9-9-27 at 1600 with 1.65V
 
Last edited:
Upgraded to DDR3-1600 and right off hand it looks like a huge difference from the DDR3-1333 I was running. Memtest definitely looks to be running faster. Here are my current settings:

blck: 195
mem div: 8:1
uncore: 16x
mem timings: auto everything.

Effective memory speed of 1560Mhz.

mch @ 1333 = mem timings of 9-9-9-24
mch @ 1600 = mem timings of 11-11-11-29

Which mch value will result in better performance?

According to the Kingston spec, the ram should have timings of 9-9-9-27 at 1600 with 1.65V

CAS 9 1333 (13.49ns) is about the same speed as CAS 11 1600 (13.75ns) access time wise. If you can tighten the timings on either one that would be the best.
 
Does this seem right TPF of 27:03 on a 6903

X5650's at 3.74Ghz clock=170
Linux 10.10 Musky Guided and all performance packages applied
memory at 7-7-7-20-1T on auto 6GB HyperX 2000
most recent bios
 
Does this seem right TPF of 27:03 on a 6903

X5650's at 3.74Ghz clock=170
Linux 10.10 Musky Guided and all performance packages applied
memory at 7-7-7-20-1T on auto 6GB HyperX 2000
most recent bios

I would expect something a bit quicker, at stock i can get 28:42 on 6903
 
Got a 6904, and it took about 3-2/3 days to complete, it slows when I use the computer for normal work, even doing e-mail, browsing, watching a YouTube vid or whatever.

The frame rate when undisturbed is about 51m30s. FahMon showed swings from 112K PPD to 117K, total points given, was about 418K.

E5645s at 3.6 GHz.

Thanks to all who responded for all the help along the way.

Bruce Hinton
 
Last edited:
Got a 6904, and it took about 3-2/3 days to complete, it slows when I use the computer for normal work, even doing e-mail, browsing, watching a YouTube vid or whatever.

The frame rate when undisturbed is about 51m30s. FahMon showed swings from 112K PPD to 117K, total points given, was about 418K.

E5645s at 3.6 GHz.

Thanks to all who responded for all the help along the way.

Bruce Hinton

Congrats, are you running that in a VM or native linux?
 
It's running in an Oracle VMBox. I need too many of my everyday Windoze programs to go native Linux. I'd be shutting down too often.

I loaded Ubuntu11.01 which offered to self-update to 11.10, to which I said, "why not?", and so it was done.
 
I have found running my 'day-to-day windows apps' in a virtualbox and running linux natively to be quite effective.
 
I have found running my 'day-to-day windows apps' in a virtualbox and running linux natively to be quite effective.

Are the Linux folding times similar?

Now that I have done the reverse of your VM, I thought about that, but the VMBox, at least the Oracle, is slow and small, and in the past, I've not been able to get all the programs into native Ubuntu.

Whether Kompozer, Gimp, and some others are usable in a window, and whether I can have 32 bit I.E. for surveillance camera monitoring of customer equipment, etc., etc., remains to be learned, but I'll try various combinations over time. The VMware may be totally different, but I've stayed away due to cost, now I'm told they have a version is freeware.

My usual 3:00 AM insomnia has lots of interesting new channels to explore now!!!

Bruce
 
Are the Linux folding times similar?

Not even close. The windoze scheduler is complete crap.

Over the last couple of days I have decided to boot up windows to play a game. I figure I can have the fah client running in the background with 20 cores out of 24. When the fah client is running as the only app, it does about 2m30s per frames on smp cores. When I start up the game( single threaded ) frames drop to about 8m45s :eek:

With linux, I run the fah client with 24 cores and it pulls about 1m45s per frame on smp cores. When I fire up VirtualBox( with 4 cores ) frames drop to about 1m55s. That is even when VirtualBox is running picasa( multi-threaded ) on a large directory of pictures.

I was completely surprised how crappy windows 'performed'.
 
Stroker,

This definitely needs looking into. Now I'm going to be in over my head again! Now I have to put the darn box into Ubuntu. The only way I like running Ubuntu is from its own drive as a bios boot selection upon starting computer. I load both O.S.s on their own drive with the other unplugged, hence no "dual boot". Security reasons...

Bruce

Not even close. The windoze scheduler is complete crap.

Over the last couple of days I have decided to boot up windows to play a game. I figure I can have the fah client running in the background with 20 cores out of 24. When the fah client is running as the only app, it does about 2m30s per frames on smp cores. When I start up the game( single threaded ) frames drop to about 8m45s :eek:

With linux, I run the fah client with 24 cores and it pulls about 1m45s per frame on smp cores. When I fire up VirtualBox( with 4 cores ) frames drop to about 1m55s. That is even when VirtualBox is running picasa( multi-threaded ) on a large directory of pictures.

I was completely surprised how crappy windows 'performed'.
 
It is pretty easy and common for installing each os on a separate drive. Main trick is to install windows first on its own drive, move it to a lower port, add the second drive for linux on a higher port, update the bios to boot the drive for linux, install linux on the second drive, and make sure the boot loader installs on the linux drive. Later on if you need to remove linux, just pull the linux drive out and update the bios boot order.

If you need more info or help, feel free to PM for it :)
 
Has anyone run the "CPU Frequency Scaling Monitor" on your SR-2 system?
I use it on my rigs ...

1- i7 970 on Giga UD7 ... it shows correct OVERCLOCK CPU speed

2- 2x L5640 on SR-2 ... it does NOT show overclock speed, but shows default cpu speed.
(currently crunching 6901 with tpf < 12) .
 
I just finished two -bigadv units at full rate, in the Ubuntu VMBox, inside Windoze 7 x 64. I have company and cannot run 24/7 so must stop that, with its wonderful point total of 415K every 3and 3/4 days. That and my electric bill, --but we at Kitguru made it to the sub 200 rank.

So I go back to the Windows only Tracker, with unchanged settings and all of a sudden, PPD is (23!!)-30K instead of 70-90K PPD in that scenario. (Edited to add); I let the Tracker SMP finish, and just started the 'command window' console version, same problem...

It's as if the VM software affected the normal Windows O.S. somehow. VM is just a 'run' type program, no service install that I can decipher, or process still running.

CPU-Z still shows 3.6 GHz, and PerfMon still shows 100% for 24 cores; temps indicate a full normal workload, and a quick CPU benchmark was normal. But the production is almost exactly what the 2600K did (at 4.2 GHz it ran 27K average), that I just delivered.

Has anyone got any idea on why this drop for my SR2 might have occurred?

I think can fix it with the little three position switch on the motherboard for BIOS selection. But I'm puzzled why my frame rates are so slow, 7minutes vs. well under 2 minutes before, but all performance indicators are fine, if not the F@H interface.

Bruce
 
That's why I gave a point range instead of one PPD rate to imply the variance due to WUs. This is one third of regular production, not small variance.

These several are all ordinary like 6097, don't remember more. Did several hundred before trying the VMUbuntu for -bigadv.

Bruce
 
So I finally got mine set up, 2 X5560's @ 180x21 (actually 180x23 since the board can force Turbo).
Problem is it likes to drop RAM; I can no longer get all 12GB to show up, and it wavers between 4 and 10GB detected.
I'm wondering if its because of the crappy modules I'm using (1.65V 9-9-9-24 DDR3-1333 OCZ Gold); can anyone confirm this before I go out and drop $120 on nicer memory?
 
So I finally got mine set up, 2 X5560's @ 180x21 (actually 180x23 since the board can force Turbo).
Problem is it likes to drop RAM; I can no longer get all 12GB to show up, and it wavers between 4 and 10GB detected.
I'm wondering if its because of the crappy modules I'm using (1.65V 9-9-9-24 DDR3-1333 OCZ Gold); can anyone confirm this before I go out and drop $120 on nicer memory?

Bump up the Vtt voltages for both CPUs. I had the same problem with OC settings.
 
Whats a good safe Vtt? I have it at 1.375 right now, and 1.4 didn't help.

I had to put mine put to 1.4. Prior to that, I was also getting some board flex which common. I loosened up the tension on the heatsinks a little and that cleared it up.... until I added four more sticks and had to bump up the Vtt voltage to 1.4.
 
Upgraded the memory to G.Skill Ripjaws and all is well.
It runs terribly hot (70C Folding), but seems to be rock solid at 180x23.
Next up, better cooling and and the quest for 4500 MHz!

UPDATE:
Got it up and folding -bigadv, getting about 107k PPD on a Virtualbox running Ubuntu 10.10 under Windows 7. Not bad for just 8 cores...

Relevant parameters:

BCLK: 180
Vcore: 1.41875
Vtt: 1.425
IOH: 1.40
VDIMM: 1.65

Everything else on auto.

Turbo on gives me a 23x loaded multiplier.
 
Last edited:
I backed down my OC to 4030 MHz and that let me bring down the Vcore by quite a bit.
I ordered some A0 X5650's off Ebay; does the SR-2 still need a special BIOS to run A0's?
 
I backed down my OC to 4030 MHz and that let me bring down the Vcore by quite a bit.
I ordered some A0 X5650's off Ebay; does the SR-2 still need a special BIOS to run A0's?

Niet
 
I think Wylullraven ment to say "&#1053;&#1077;&#1090;"

On topic, wasn't the modified BIOS to support different chips that weren't supported in any stepping, rather than A0 steppings generally?
 
I think Wylullraven ment to say "&#1053;&#1077;&#1090;"

On topic, wasn't the modified BIOS to support different chips that weren't supported in any stepping, rather than A0 steppings generally?

IIRC, the modified BIOS release were (and are) required for some of the 55XX nehalem based ES chips that don't like to run in anything. The 56XX series have much better compatability and are really the only thing you would want to be running in an SR-2 anyway.
 
I'm having problems with Turbo - if I enable Turbo I get a 21x multiplier under full load (expected) but I also get a 23x multiplier under light loads that crashes the system. Right now I have Turbo disabled and am running at 20x200 for a 4 GHz OC, but I was wondering if there is a way to disable the 22x/23x multipliers and have the processor stay at 21x.
 
Back
Top