SR-2 Optimization Thread

[Tobit]

Does the NIC problem exist @ Stock?

I am now benching a 2684 at stock clocks to see if the NICs still fail after a certain period of time so I can rule out a bad overclock. It's a very intermittent problem in that sometimes it occurs after 30 minutes but sometimes not for upwards of 3 hours so it's a PITA to troubleshoot.

 

[Tobit]

Have you hooked power to the little floppy power connector near the ports? I've heard that helps with flaky USB but I have no idea about the LAN.

Ya know, I was just looking at the big visual guide and it says this connector is for USB and Memory so it might be something to try anyway. My MM tray will be here today and I was planning on putting it in the case this afternoon. I'll go ahead and hook it up.

 

[Tobit]

Do we have an idea of what a safe 24x7 maximum voltage is for IOH? KONA over at EVGA forum wants me to try more IOH to try and cure my NIC issue at 210/212 BCLK but I am at 1.4V now.

 

[musky]

I can tell you that 1.475 is too high, as it overheats something to the point of having to let the system cool down before you can reboot. I haven't gone over 1.4V 24/7 personally. Keep in mind that this is just the setting - for me at least, actual is in the 1.38V range. One more tick higher would put you barely over 1.4, which is probably fine.

 

[Tobit]

I'll tell y'all one thing, A49 BIOS is much more forgiving on high BCLKs than A50. I think A49 is my go to BIOS at the moment.

 

[musky]

I'll tell y'all one thing, A49 BIOS is much more forgiving on high BCLKs than A50. I think A49 is my go to BIOS at the moment.

Since I came to the conclusion a while ago that A50 sucks and the original A41 these things originally came with is the best bios released for them, I may need to try the A49. Switching to A50 from a modded A47 (for CPU support) netted me slower average frame times in general and all kinds of fluctuation, using identical settings and the same hardware.

 

[nitrobass24]

Can someone zip and upload their IBT folder?
Everyone i have downloaded (5) doesnt work for me!! :x

 

[Chilly]

Do we have an idea of what a safe 24x7 maximum voltage is for IOH? KONA over at EVGA forum wants me to try more IOH to try and cure my NIC issue at 210/212 BCLK but I am at 1.4V now.

I was running at 1.45v no problem for months, I set it auto(by accident) and it auto sets it *stable* at 1.35v...

The silly thing is, if I set it to 1.4v manually, or 1.35v, I get lower than expected performance, but auto sets it at 1.35v and it runs perfectly. Go figure.

 

[nimbyfaygo]

I'm developing a new problem it seems, I've received the BSOD "System Service Exception" 0x0000003B twice yesterday. I scoured through msdn to try and find a solution. I came up with...

Bug Check 0x3B: SYSTEM_SERVICE_EXCEPTION
This error has been linked to excessive paged pool usage and may occur due to user-mode graphics drivers crossing over and passing bad data to the kernel code.

The system specs are 206bclk 1.4 vcore 1.375 VTT and 1.4 IOH. 75c on the hottest core. Average temps are about 71c.

I can't duplicate the error, it happened first time when the computer was folding. Only minimal processes running. 2nd time I was playing SC2.The system came back up perfectly after restart. I ran a max memory run of intel burn test to stress the system a bit.

I'm not able to do any any more testing since I'm at work, but I just want to post here and see if anyone has any ideas.

I'm wondering if it's the GPU, it could be about to kick the bucket. But it could be related to the overclock, or even ram related. I want to try running at a reduced clock and see if I can eliminate that variable.

I'll be speaking with a system architect at lunch and see if he has any ideas (maybe not the best person to ask, its slow on the weekends)

 

[jebo_4jc]

That's an odd one.

I'm no help.
Posted via [H] Mobile Device

 

[musky]

I'm developing a new problem it seems, I've received the BSOD "System Service Exception" 0x0000003B twice yesterday. I scoured through msdn to try and find a solution. I came up with...

Bug Check 0x3B: SYSTEM_SERVICE_EXCEPTION
This error has been linked to excessive paged pool usage and may occur due to user-mode graphics drivers crossing over and passing bad data to the kernel code.

The system specs are 206bclk 1.4 vcore 1.375 VTT and 1.4 IOH. 75c on the hottest core. Average temps are about 71c.

I can't duplicate the error, it happened first time when the computer was folding. Only minimal processes running. 2nd time I was playing SC2.The system came back up perfectly after restart. I ran a max memory run of intel burn test to stress the system a bit.

I'm not able to do any any more testing since I'm at work, but I just want to post here and see if anyone has any ideas.

I'm wondering if it's the GPU, it could be about to kick the bucket. But it could be related to the overclock, or even ram related. I want to try running at a reduced clock and see if I can eliminate that variable.

I'll be speaking with a system architect at lunch and see if he has any ideas (maybe not the best person to ask, its slow on the weekends)

If it happened while folding, assuming you are not folding on your GPUs, i doubt it is video card-related. Could be a suppose...something is displaying while folding...just seems unlikely. I'd guess it to be a general O/C issue - too much speed, not enough voltage, etc.

 

[nitrobass24]

Ok what all have i missed in terms of getting 200+ BLCK?

I have loosened my memory to 11-11-11-29 B2B disabled and cranked my vcore and VTT to 1.4 for both CPUs.

When i go to 202 BLCK and run IBT i get failures.

 

[may i be worthy]

Ok what all have i missed in terms of getting 200+ BLCK?

I have loosened my memory to 11-11-11-29 B2B disabled and cranked my vcore and VTT to 1.4 for both CPUs.

When i go to 202 BLCK and run IBT i get failures.

IOH at 1.4v? QPI signal tweak at ~ -90?

I wish I knew for sure. It certainly seems hit and miss over 200. SR2#3 is doing great at 205, whereas SR2#2 cannot do over 201.

Do you have BIOS of A49 hacked or A50 to allow uncore down to 17x or 18x? (previous versions had the setting but it did not work)

 

[nitrobass24]

IOH = 1.4

QPI Sig tweaks are as low as they go.

How do i know which BIOS i have?

 

[may i be worthy]

IOH = 1.4

QPI Sig tweaks are as low as they go.

How do i know which BIOS i have?

Pop into your bios - first option for Standard BIOS Features: System Overview, AMIBIOS ID: 1w555A49 etc

EDIT: OR run CPU-Z

 

[10e]

I've gotten this in the past with systems that were IBT or LinX stable, but as soon as the GPUs went hard core the system did this.

I've seen it a lot more on LGA-1156 systems and in those cases it was a VTT issue, but those chips have the PCI-E controller onboard.

With a Socket 1366 board it could be IOH or ICH voltage or even IOH/ICH IO....

That I7 overclock guide on EVGA is pretty good at showing some useful voltages but for x58 boards, but if your friends are no help try it out. There's a part about SLI, 3dMark and IOH/ICH voltages a little lower on the post:

http://www.evga.com/forums/tm.aspx?m=7528

I'm developing a new problem it seems, I've received the BSOD "System Service Exception" 0x0000003B twice yesterday. I scoured through msdn to try and find a solution. I came up with...

Bug Check 0x3B: SYSTEM_SERVICE_EXCEPTION
This error has been linked to excessive paged pool usage and may occur due to user-mode graphics drivers crossing over and passing bad data to the kernel code.

The system specs are 206bclk 1.4 vcore 1.375 VTT and 1.4 IOH. 75c on the hottest core. Average temps are about 71c.

I can't duplicate the error, it happened first time when the computer was folding. Only minimal processes running. 2nd time I was playing SC2.The system came back up perfectly after restart. I ran a max memory run of intel burn test to stress the system a bit.

I'm not able to do any any more testing since I'm at work, but I just want to post here and see if anyone has any ideas.

I'm wondering if it's the GPU, it could be about to kick the bucket. But it could be related to the overclock, or even ram related. I want to try running at a reduced clock and see if I can eliminate that variable.

I'll be speaking with a system architect at lunch and see if he has any ideas (maybe not the best person to ask, its slow on the weekends)

 

[Tobit]

I don't have the secret answer for BCLKs over 200 but I have discovered too much voltage can slow down overclocks. Going faster doesn't always mean increasing voltages, sometimes you have to lower them.

I discovered this when troubleshooting the onboard NIC problem at high BCLKs with my SR-2. Everyone kept telling me to increase IOH but that actually made my overall OC worse. I then read some posts over at XS where people were lowering certain voltages during high OCs so I decided to give it a try.

My 210 BCLK OC is now running at an IOH of ~1.34 volts and there are no longer any problems with NIC stability and the overall OC is much smoother and I'm getting excellent, consistent, frame times.

I'd share all my settings but y'all probably wouldn't believe it and call shens.

 

[nimbyfaygo]

I've gotten this in the past with systems that were IBT or LinX stable, but as soon as the GPUs went hard core the system did this.

I've seen it a lot more on LGA-1156 systems and in those cases it was a VTT issue, but those chips have the PCI-E controller onboard.

With a Socket 1366 board it could be IOH or ICH voltage or even IOH/ICH IO....

That I7 overclock guide on EVGA is pretty good at showing some useful voltages but for x58 boards, but if your friends are no help try it out. There's a part about SLI, 3dMark and IOH/ICH voltages a little lower on the post:

http://www.evga.com/forums/tm.aspx?m=7528

I think the problem was related to the IOH voltage and BCLK. I didn't think those were related, but it seems to have cured that BSOD. I dropped the BLCK to 203 and bumped to IOH 1.425. It seems like that is a little high. But the temps seem to be ok. Even stranger my TPF on the same WU actually improved by a couple seconds.

I put the most stress I can imagine on the system. I ran prime95 small tft, burn test on standard, 3dmark vantage, and furmark. Other than bogging down it's staying rock solid.

 

[nimbyfaygo]

I don't have the secret answer for BCLKs over 200 but I have discovered too much voltage can slow down overclocks. Going faster doesn't always mean increasing voltages, sometimes you have to lower them.

I discovered this when troubleshooting the onboard NIC problem at high BCLKs with my SR-2. Everyone kept telling me to increase IOH but that actually made my overall OC worse. I then read some posts over at XS where people were lowering certain voltages during high OCs so I decided to give it a try.

My 210 BCLK OC is now running at an IOH of ~1.34 volts and there are no longer any problems with NIC stability and the overall OC is much smoother and I'm getting excellent, consistent, frame times.

I'd share all my settings but y'all probably wouldn't believe it and call shens.

I'd love to drop some voltages, the only voltage I haven't played with much is the CPU voltages. I'm sure I could drop 1 cpu, but I just haven't messed with adjusting them independently.

 

[nitrobass24]

Tobit can you take some pics of the BIOS?

 

[Tobit]

Tobit can you take some pics of the BIOS?

and pause this lovely bigadv that is currently folding? Well, umm.. ok.. give me a few.

 

[musky]

Tobit can you take some pics of the BIOS?

FYI, I did try Tobit's bios version and actual settings with one of my machines this weekend with basically identical hardware, and I couldn't even boot at 203 bclk. Tobit apparently has two very good chips that do not require much voltage to run at >3.6 GHz. My two require a boatload (1.38 actual to be exact, at 3.618.) The other thing he has that is significantly different is IOH on Auto, which equates to 1.32 or something insanely low. He can run there, but I couldn't come close.

By all means, try his settings. Just don't be too surprised when they don't work for you. If they do, fantastic.

 

[zero2dash]

Just wanted to share this ~
If anyone has problems with sound stuttering (I thought of it more like sound being processed through a tin can on the high frequencies), KONAn came up with a signal tweak that fixes it (and I can confirm that it does work).
http://i963.photobucket.com/albums/ae111/BikeBarian/Clipboard03.jpg

 

[Tobit]

2 x L5640 @ 210 BCLK -- DDR3-2100
A49 BIOS

 

[Tobit]

The other thing he has that is significantly different is IOH on Auto, which equates to 1.32 or something insanely low.

1.34v actually. One thing I still have to try is lowering CPU vcore even more. I intentionally started high but I suspect I can go lower.

 

[nitrobass24]

Cool i have some new stuff to try!!!
Which BIOS are you using?

 

[Tobit]

Cool i have some new stuff to try!!!
Which BIOS are you using?

Sorry, A49 BIOS. A50 is flaky at high BCLKs.

 

[nimbyfaygo]

Very nice! I can't wait to try tweaking some settings.

 

[may i be worthy]

OMG Tobit - I am pissing myself looking at your screenies. "Set Auto on everything, crank to 11 and damn the torpedoes!". Looks like you have a golden board and golden CPUs. Fabulous.

What is your actual uncore being set by Auto? - look up NB freq in the eleet Memory tab and divide by baseclock. Auto always sets mine at 20x, and I know for a fact that uncore 19x and 20x will fail (make that explode) at 205 baseclock and 4.3Ghz

But the funny thing is SR2#2 - which cannot go over 201 is still running hacked A49, and SR2#3, which is just humming at 205 is using A50. So working on the Seinfeld/George Costanza theory of maybe lets try doing the opposite of what I would normally do, I might try A50 on SR2#2 - hoping that there is something tuned better in A50 for higher multi chips. Hurts you, helps me?

I might also try hacked A49 on SR2#3 with the inconsistent frames times. Or I might try Auto-practically-everything-Tobit-style and see what happens. Never actually tried that. Too much of a control freak.

Lets start over on all 3 rigs! Hmmm, are you actually hired by Musky and Flecom in an elaborate conspiracy to drop my ppd?

 

[Tobit]

What is your actual uncore being set by Auto? - look up NB freq in the eleet Memory tab and divide by baseclock. Auto always sets mine at 20x, and I know for a fact that uncore 19x and 20x will fail (make that explode) at 205 baseclock and 4.3Ghz

3360/210 = 16

Lets start over on all 3 rigs! Hmmm, are you actually hired by Musky and Flecom in an elaborate conspiracy to drop my ppd?

I plead the fifth!

 

[musky]

Uncore is locked on the L5640 ESs, so it will be 18x regardless of bios. A50 will make much more of a difference for those of you with chips that allow uncore adjustments.

I was as excited as the rest of you seeing Tobit's results with Auto everything...that is, until I actually tried it. I'm going ot say that if your L5640s will run 3.6 or better on 1.3V +/- Vcore, you have a fighting chance. It seems that high Vcore (which in turn appears to require high IOH) stops you dead at around 200 blck. If you can relax Vcore and IOH, you may be golden. That being said, SR-2 #1 of mine fits the bill (3654 @ 1.32 actual Vcore), so maybe there is hope...

 

[may i be worthy]

Uncore is locked on the L5640 ESs, so it will be 18x regardless of bios. A50 will make much more of a difference for those of you with chips that allow uncore adjustments.

Confused - Tobits is an Extra Spicy right? And his uncore seems to be at 16x?

I was as excited as the rest of you seeing Tobit's results with Auto everything...that is, until I actually tried it. I'm going ot say that if your L5640s will run 3.6 or better on 1.3V +/- Vcore, you have a fighting chance. It seems that high Vcore (which in turn appears to require high IOH) stops you dead at around 200 blck. If you can relax Vcore and IOH, you may be golden. That being said, SR-2 #1 of mine fits the bill (3654 @ 1.32 actual Vcore), so maybe there is hope...

Yeah my attempts this morning with more auto have not got past booting yet.(FU Tobit! )

In the last half hour I have updated SR2#2 to A50, and things were just as explodey at 205 - same freezes and STOP 124 - and immune to more vtt and less uncore.

Just tried upping IOH from 1.40 to 1.45 and IOH pll from Auto to 1.85 and bingo - suddenly stable for 3 passes of IBT - (never got through 1 pass before) Testing to see if it was the IOH or the IOH PLL or both doing the hard work... might be on to something here, but as I am changing a few other things at the same time it might pay to slow down a bit... more to follow.

 

[Tobit]

Testing to see if it was the IOH or the IOH PLL or both doing the hard work...

I think IOH PLL does the more hard work. With both set at auto, IOH is 1.34 and IOH PLL is 1.82 according to E-LEET. 1.82 is quite a bit higher than normal IIRC.

 

[may i be worthy]

I think IOH PLL does the more hard work. With both set at auto, IOH is 1.34 and IOH PLL is 1.82 according to E-LEET. 1.82 is quite a bit higher than normal IIRC.

Don't knock it - it might be part of your secret sauce! That is my theory I am testing - will know is a few mins... but SR2#3 @ 205 working nicely 1.8v set = 1.82v in eleet.

 

[may i be worthy]

Hmm, maybe not. It was IOH voltage needing to be 1.45 or better.

IOH - 1.450v, IOH PLL 1.850v = pass 3 - bench 6061: 1 min 01sec
IOH - 1.400v, IOH PLL 1.850v = STOP #124
IOH - 1.425v, IOH PLL 1.825v = STOP #124
IOH - 1.425v, IOH PLL 1.850v = STOP #124
IOH - 1.450v, IOH PLL 1.800v = pass3 - bench 6061: 1 min 03sec

Also changing the IOH PLL results in a failure to post after setting in the bios (06 - timing clock) Cold boot after and away you go.

I might bench it before I get much older...

EDIT: added bench times. Seems to like a bit more IOH PLL volts... maybe I should try 1.9v

 

[Tobit]

BTW, for the record.. I've never admitted to being IBT or Prime stable. Since this boxen is dedicated F@H, I have several bigadv and non-bigadv "bench units" saved off that I do all my OC testing with. I learned long ago that other stability testing doesn't necessarily mean jack shit.

 

[may i be worthy]

QFT.

IBT is great for getting an idea of how close you are, but 6071 is the evil bastard you really need to pass.

 

[jebo_4jc]

Just wanted to say I think I've tried nearly everything to get 203 bclk stable with no luck. 200 appears rock solid at 1.29v vcore. Nearly everything else is auto.
Posted via [H] Mobile Device

 

[may i be worthy]

Well fructose.

After posting bench times that were very very promising (2686 - 10:40) - it turns out the clock is getting messed up under load... I have seen this before - the benchmarks report 10mins 40, but my stopwatch shows 12 mins 13 - and the clock is running a few minutes low. Testing to see what is messed up...

EDIT: I cant get the system clock stable at 205, despite folding seeming to work fine. But every second boot will not post and things are wacky. So looks like just the one SR2 at 205 will have to do for now.

 

[nimbyfaygo]

I never was able to stay stable past 206. I think my limit is the CPU's they are the 45nm versions, and Vcore is seemingly quite important to them.