• 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.

SR-2 Optimization Thread

SazanEyes said:
No, the labeling on the board is correct. The voltages for CPU0 map to the socket labeled CPU0. The only things that are reversed are the software temperature sensors.
Click to expand...

And that the CPUs in taskmanager of 0 to 11 refer to socket labeled CPU-1, and 12 to 23 refer to socket labeled CPU-0.

Does. My. Head. In.

But last night I dropped voltage on CPU-1 (socket labeled) from 1.400v to 1.375v vcore - reducing temps on my hottest cpu by 5+ degrees. This is awesome.

So the end result is that the temps between CPUs are far closer, but the reported voltages are further apart!

CPU-0 socket labeled = 1.400v vcore = reported 1.40v = average 58c folding ("CPU-1")
CPU-1 socket labeled = 1.375v vcore = reported 1.35v = average 63c folding ("CPU-0")

Passed 5 hours Linpack, have moved on to non-bigadv to give it the real trial by fire - 6701s :cool:

Jmtyra said:
Asking the experts: Any [H]ard PPD numbers w/a pair of X5680s? Found a few posts here and there, but no actual hard PPD numbers. Trying to balance out the $$$/ppd/kWh ratio.

TIA :)
Click to expand...

Haven't got hard numbers for x5680s, but the best bang for buck (outside of special chips!) would be X5650s. (here they cost less than L5640s) I would recommend x5660s if you can afford them. I would not recommend x5680's except maybe second hand, if you can get them for new x5660 prices.

Running my two rigs, there is a small difference in speed, but for some reason the x5660s at 4.3 seem to draw a lot better units. I haven't run them long enough to call it for sure, but I can't help but think the slightly faster machine is being favoured with better units. So right now I recommend anything that can do 4.3Ghz and up. ;)
 
Jmtyra said:
Asking the experts: Any [H]ard PPD numbers w/a pair of X5680s? Found a few posts here and there, but no actual hard PPD numbers. Trying to balance out the $$$/ppd/kWh ratio.

TIA :)
Click to expand...

Here are Punchy's numbers on dual X5670s at stock speed:

http://hardforum.com/showpost.php?p=1036140100&postcount=72

If you are talking about running them in an SR-2, then it is going to depend on how high Westmeres can actually go, and how well you can cool them. I would guess they are similar to something like a 980x - 4.3 - 4.4 is about as good as you are going to get and MIBW's 160K+ ppd on his 5660s will be where you will end up. It brings up the point of how much processor do you really need when you figure in overclocking? i am going to guess that it is somewhere in the X5650-X5660 range, and anything over that is going to get too hot to run at the higher potential clock speeds.
 
MIBW/musky: thanks for the info -- I need to sell off some stuff to afford them now. :D
If they do become available I'll post some OC attempts. ^.^
 
Thanks to Safield, the CPU0/CPU1 confustion can end!

annotated2.jpg


Link to thread: http://www.evga.com/forums/tm.aspx?m=651731
 
Cool, so if CPU1 is showing inconsistent temps and needs to be reseated.
Then I need to reseat CPU0. Sweet and thanks for the headaches that were to come.
 
At the sake of sounding stupid, I have to admit that I can't find the uncore adjustment at all! I recently updated to A50 bios.
I'm running stable at 202 bclk, but I'm not able to get any higher with my current settings. For for the purpose of oc'ing I have the ram set to 2:6.

I'm really hoping I can get to 211. But we'll see
 
nimbyfaygo said:
At the sake of sounding stupid, I have to admit that I can't find the uncore adjustment at all! I recently updated to A50 bios.
I'm running stable at 202 bclk, but I'm not able to get any higher with my current settings. For for the purpose of oc'ing I have the ram set to 2:6.

I'm really hoping I can get to 211. But we'll see
Click to expand...

It should be right above the DRAM frequency setting, or at least it was in the modded A49.
 
Hey everyone. I just wanted to give a THANK YOU!!! for all of the help. I've got mine up and running and it is a sight to behold!! I've got NUMA turned off and used one of the BIOS templates and worked like a charm.

Of note, I did spend 4 hours trying to get past a BSOD 124 at default. I said screw it and plugged into the OC settings and haven't looked back. This thing just FLIES!!
 
Ugh, I thought I had a stable overclock(2xL5640 @3.6GHz) - it ran for a couple of weeks doing -bigadv and regular -smp WUs - but then the other day it failed at 60% of a 2686 unit and I received an "unstable_machine" error. Shit.

So I check out my settings, adjust and try a different memory strap (was: AUTO). I tried 1600 and LinX failed after 20 runs. I tried 1366 and LinX failed after about 10 mins. After changing to DRAM:Ratio, I was able to complete 50 runs at maximum ram (12GB) without error...

...so back to BigAdv, I crunch regular SMP including multiple 6701s and 6702s. Finally I snag a 2684...and the rig folds fine up until 60% on the nose when it shuts off again! At least I didn't get an unstable_machine error and lose the WU, but what the hell? I don't have time to bench since I'm working a lot this week, so my temp solution is to dial back to 190 BCLK (everything else the same) and hopefully it'll at least complete the WU.

Are my chips degrading or something? Or was my OC just not stable to begin with but I "got lucky" for a couple of weeks? Sorry for the long post.
 
sounds similar to my experience. I would be seemingly stable up around 200mhz bclk, but I would get an error after a couple days.

I had to back down to 190mhz also to maintain 100% stability.
 
Hmm, that's disappointing since I've become accustomed to this high standard of folding! :p Dialing back to 190 adds 2 minutes (ish) to my 2684 TPF, but it's better than losing the entire WU, as I did with the 2686 (115k down the drain...).

My settings are 1.35 vcore, 1.4 vtt, so perhaps I have some headroom left on my vcore...if nothing else, I still want to OC my RAM.
 
Jmtyra said:
MIBW/musky: thanks for the info -- I need to sell off some stuff to afford them now. :D
If they do become available I'll post some OC attempts. ^.^
Click to expand...

No worries. I will have some hard data on how x5680's perform soon. I did the bad thing again.:eek: SR2 number 3, under construction. :D

@capreppy - great you are up and running, enjoy it. I had to fire up my previous box, a Q6600 and yes, I forgot how slow it is. These babies do fly ;)

@aix - I wish I could help - once I am stable at memtest and IBT I now move on the running non -bigadv for a day or so to get as many 6701s as possible :)rolleyes:yes I know) - if I can get through 3 or 4 6701s I know I am 100% stable. I have never had such an intermittent failure as what you describe. How much stability testing did you do? eg did you pass 8+ hours IBT at any point? If you were 8hr + stable, and could do half a dozen 6701s without failing, then I am out of ideas.

Could be memory - one weak region of memory that is rarely used. => try a dozen instances of HCI memtest at 1/12 ram each.

How is your cooling on the motherboard?
Could be power supply issues? (Wild Ass Guess)

I am struggling to find absolute definitions of stable these days - my x5650 rig can get 12+hr IBT stable, but needs another few notches of vcore to be stable for 6701s, but my x5660 can be perfectly folding stable 2 notches of vcore below IBT stable! :confused:
 
Last edited:
:eek:

MIBW, you bring the addiction to a new level

If you're ever interested in selling a hex core CPU or two, LMK.
 
Last edited:
may i be worthy said:
No worries. I will have some hard data on how x5680's perform soon. I did the bad thing again.:eek: SR2 number 3, under construction. :D
Click to expand...

Holy S**t, a 3rd SR-2:eek::cool:

Very impressive, is this a work machine, a pure folder or both???

PS don't tell Kendrak - He'll get jealous:p
 
may i be worthy said:
@aix - I wish I could help - once I am stable at memtest and IBT I now move on the running non -bigadv for a day or so to get as many 6701s as possible :)rolleyes:yes I know) - if I can get through 3 or 4 6701s I know I am 100% stable. I have never had such an intermittent failure as what you describe. How much stability testing did you do? eg did you pass 8+ hours IBT at any point? If you were 8hr + stable, and could do half a dozen 6701s without failing, then I am out of ideas.

Could be memory - one weak region of memory that is rarely used. => try a dozen instances of HCI memtest at 1/12 ram each.

How is your cooling on the motherboard?
Could be power supply issues? (Wild Ass Guess)

I am struggling to find absolute definitions of stable these days - my x5650 rig can get 12+hr IBT stable, but needs another few notches of vcore to be stable for 6701s, but my x5660 can be perfectly folding stable 2 notches of vcore below IBT stable! :confused:
Click to expand...

I've been under the impression that LinX and IBT are the same thing...they both use Linpack, right? If that is the case, I passed a run of 50 at max RAM without error; not sure how long that took, exactly.

As I mentioned, I did already completed a 2684 WU as well as a whack of 6701s. I didn't actually witness the most recent shutdown on the 2684, but the system actually tanked instead of giving me an unstable_machine error, and there isn't a minidump file for it either. Maybe I just need more vcore?

I'll try testing the memory with HCI before changing any settings. Really weird that my errors both happened at 60% though. Thanks for your help once again!
 
gkanakis said:
So what is the "Safest" voltage on L5640 on the high side?
Click to expand...

I was up as high as 1.42 Vcore actual to get to 18 x 210. It was not as fast as it is at 203 x 18, so I am back down to 1.375 actual.

IOH - 1.450 on the setting was fine. 1.475 would cause me to have to let the board cool down after it crashed, so i would not recommend that.

Vtt - 1.425 was the highest i went, and it seemed OK.

Vmem - I only went up to around 1.7 actual, and it didn't help stability. I am sitting at 1.669 actual on both right now.
 
may i be worthy said:
No worries. I will have some hard data on how x5680's perform soon. I did the bad thing again.:eek: SR2 number 3, under construction. :D
Click to expand...

So that will be one dual X5650 machine, one dual X5660 machine, and one dual X5680 machine? All I can say is...WOW! Makes my two SR-2 L5640 machines look kinda puny...
 
musky said:
I was up as high as 1.42 Vcore actual to get to 18 x 210. It was not as fast as it is at 203 x 18, so I am back down to 1.375 actual.

IOH - 1.450 on the setting was fine. 1.475 would cause me to have to let the board cool down after it crashed, so i would not recommend that.

Vtt - 1.425 was the highest i went, and it seemed OK.

Vmem - I only went up to around 1.7 actual, and it didn't help stability. I am sitting at 1.669 actual on both right now.
Click to expand...

All of my setting are lower then these, so I'm doing ok!!

I guess I can push a little more tomorrow!
 
jebo_4jc said:
:eek:

MIBW, you bring the addiction to a new level

If you're ever interested in selling a hex core CPU or two, LMK.
Click to expand...

As my 2 year old is fond of saying: "NO dats MINES!" :p

Nathan_P said:
Holy S**t, a 3rd SR-2:eek::cool:

Very impressive, is this a work machine, a pure folder or both???

PS don't tell Kendrak - He'll get jealous:p
Click to expand...

Work machine by day, folder by night. :cool:

I don't want kendrak to be jealous... I want him and sazan and musky to get a third SR2 too. Right now the [H] Triple SR2 Club is a bit thin on members.:)

Aix. said:
I've been under the impression that LinX and IBT are the same thing...they both use Linpack, right? If that is the case, I passed a run of 50 at max RAM without error; not sure how long that took, exactly.

As I mentioned, I did already completed a 2684 WU as well as a whack of 6701s. I didn't actually witness the most recent shutdown on the 2684, but the system actually tanked instead of giving me an unstable_machine error, and there isn't a minidump file for it either. Maybe I just need more vcore?

I'll try testing the memory with HCI before changing any settings. Really weird that my errors both happened at 60% though. Thanks for your help once again!
Click to expand...

Yes, ibt is a front end for linx. Forget the bullshit that people say 20 passes = stable. Stable is 8 hrs plus. Folding runs for days! I have found very high to be the most strenuous. Ymmv.

The only time I ever pranged a bigadv was my test unit, and that was when mucking around at 4.4 ghz, an oc that clearly also failed ibt. So I have never had a unit fail, but I have had a ton of 6701 induced blue screens that did not kill the WU. But I thought bigadv were less stressful.

Hci is easy to use if you set up a shortcut with flags to run 1024mb rather than entering it manually... It is in the help, sorry I am away for a few days and not at my machine.

musky said:
So that will be one dual X5650 machine, one dual X5660 machine, and one dual X5680 machine? All I can say is...WOW! Makes my two SR-2 L5640 machines look kinda puny...
Click to expand...

Hardly puny scores they are racking up!

Well when you spell it out like that, it looks like I am just an x series collector!...looks like the x5670 is next. But remember your rigs are folding 24/7 which beats my part time show ponies;)

But my two sr2s were just too different from my six q6600 render boxes w 4gb ram to render well together, too lopsided, so I am retiring them for the third sr2. That is the business case... What a coincidence that it will be of use folding.:cool:
 
may i be worthy said:
But my two sr2s were just too different from my six q6600 render boxes w 4gb ram to render well together, too lopsided, so I am retiring them for the third sr2. That is the business case... What a coincidence that it will be of use folding.:cool:
Click to expand...

So when you replacing the other three Q6600's? :p
 
may i be worthy said:
I don't want kendrak to be jealous... I want him and sazan and musky to get a third SR2 too. Right now the [H] Triple SR2 Club is a bit thin on members.:)
Click to expand...

Well Musky already has the processors sitting in an Asus dually board so its less of an upgrade for him, but they all have one problem - getting the spend past the wife:eek::D
 
may i be worthy said:
I don't want kendrak to be jealous... I want him and sazan and musky to get a third SR2 too. Right now the [H] Triple SR2 Club is a bit thin on members.:)
Click to expand...

Yeah, I don't think that's going to happen on my end. I'm still trying to decide whether to replace the dead i7 920 mobo or just sell the proc. For some reason I suspect you won't be lonely for too long, though.
 
Nathan_P said:
Well Musky already has the processors sitting in an Asus dually board so its less of an upgrade for him, but they all have one problem - getting the spend past the wife:eek::D
Click to expand...

That is not my problem. I actually kind of like the idea of having something that fits in a case and uses relatively little juice to run. SR-2s are fun little toys, but sometimes simplicity in this game is a good thing.
 
musky said:
That is not my problem. I actually kind of like the idea of having something that fits in a case and uses relatively little juice to run. SR-2s are fun little toys, but sometimes simplicity in this game is a good thing.
Click to expand...
Truth
Posted via [H] Mobile Device
 
Hey all,

Haven't contributed in a while, but what results have you all seen for the QPI signal tweaks?

I was able to lower my QPI signal tweak to -93 for QPI 0 and (obviously) left QPI 1 at -16. This allowed me to run 195 BCLKx20 multi but with a lower VTT of 1.40 from 1.425 while leaving VCore at 1.35.

I left the unCore at stock x16 (2x RAM speed) and this allowed me to lower temps by a couple degrees celsius on both chips.

I did notice that LinX is showing 75 - 76 GFlops vs. 77 - 78 so there must be some small side effect if RAM timings and other speeds are identical to before?

I'm gonna also troll around EVGA's forums, but I figured I'd ask here first because everyone is so [H]ard core on their SR-2s :)
 
Good info. I'm not sure too many of us here have played with signal tweaks too much. Somebody chime in if I'm wrong.
 
I've read through this SR-2 thread (and evga ones) quite a bit, and the consensus seems to be to put your QPI as low as possible. shamino's SR-2 OC template has a post where he mentions keeping QPI 0 as low as you can get it and still boot "consistently".

Not sure, or haven't see, too much mentioned outside of that. Personally, I haven't had the OC patience to verify how much of a difference it does/does not make.
 
Every time I have touched the IOH QPI signal tweaks, it has lead to disappointment. I just don't think they do enough to warrant spending any time on them. I have never "gained" anything by setting them to anything besides Auto, but I have lost both performance and stability with certain settings. THe best I have accomplished is no difference over Auto, which leads me to my opinion of them not being worth messing with. I would love for someone to prove me wrong.
 
Thanks for the input guys,

Right now I bumped up the VCore to 1.35625 on both and taken VTT down to 1.375 and LinX 0.6.3 is stable after about an hour. I tend to try and go around 4 - 6 hours on LinX before I draw conclusions on stability, but again, temps seem around 2 - 3 degrees lower than I was getting with VTT at 1.40 and a good 6 - 7 degrees lower than with VTT at 1.425.

This seems to be pretty close to the limit for my chips. It seems even if I get stable at 4.0+ Ghz with or without turbo, it doesn't give me the associated folding@home performance increase I'd be expecting.

What's nice here is that unlike my Core i7 chips which have gone as long as 12+ hours to fail LinX, the SR-2 seems to fail earlier if at all.

Cheers!
 
musky said:
That is not my problem. I actually kind of like the idea of having something that fits in a case and uses relatively little juice to run. SR-2s are fun little toys, but sometimes simplicity in this game is a good thing.
Click to expand...

There is that, what kind of power draw are you seeing on the asus?
 
I notice now (after tweaking A47's uncore from 16x -> 18x) that it is indeed locked at 16x still.
Time to go get A50 I suppose.... :eek:

Running A50, if I put uncore to anything other than 16 it still reports 16.
CPU-Z and ELEET report NB freq as 3200, CPU-Tweaker reports 16x uncore, 3200mhz.
I've raised Vtt thinking that was the problem....up to 1.35 now, still 16x.
Is the L5640 locked to 16x or something? Is it because they're extra spicy and uncore has indigestion???

I think that's it.
 
Last edited:
You must log in or register to reply here.
Back
Top