SR-2 Optimization Thread

IME gpu2 doesn't really cause much of a 'cpu cycle fight' between the smp2 client running bigadv vs the gpu2 client(s).

I think when I averaged it out I gain 10k PPD for running my 260s even after losing 1-2k to the smp2 client itself. Maybe things have changed since I checked this; admittedly, it's been awhile and there's been new WU's released for gpu2.....guess I'll have to investigate this myself. :eek:

I've always netted a gain of about 3000-4000ppd running gpu2 client. However I never did really like to run the gpu client in the SR-2 because it places a bit of a strain on the psu. I don't think the extra load and power draw is worth the minor ppd increase
 
Hmm, guess I'll -oneunit the gpus and see what happens. ;)
edit interesting....things have changed quite a bit since last time I checked.

wgpus.png

vs
nogpus.png


Considering the power draw and my ridiculously increasing power bill, they will now stay off permanently. ;)
I hope I pull another 6900 so I can see what kind of TPF/PPD I pull with the cards off for the entire length of the WU. :eek:
 
Last edited:
Hey guys, nearing that awesome 4.0 speed - running into a couple hitches though. Figured as this is an SR-2 thread that maybe I could get an idea from someone as to what's up? Permalink to my build log on OCN is below and explains all my settings so far.

Currently at 200 BCLK @ 3.80ghz!

http://www.overclock.net/intel-build-logs/864314-melchior-project-sr-2-build-oc-8.html#post12050983

To sum up at the bottom, I've had to try a lot of different settings now but changing the Uncore to 12x (1,600mhz) manually from the default auto suddenly made the game and testing lockups / hangs go away for now. Sound still stutters at max CPU load but I'll take that over insta-locks in a game or the desktop lol.

If I want to hit 4.0+ do I need more voltages or should I just set the Uncore higher? Also despite these being redline modules @ 1,600, how much higher should I take them before dropping to the 2:10 ratio? They're running at rated speed now at any rate which is cool as the rest of the system is quite OC'd a bit.
 
Last edited:
Hey guys, nearing that awesome 4.0 speed - running into a couple hitches though. Figured as this is an SR-2 thread that maybe I could get an idea from someone as to what's up? Permalink to my build log on OCN is below and explains all my settings so far.

Currently at 200 BCLK @ 3.80ghz!

http://www.overclock.net/intel-build-logs/864314-melchior-project-sr-2-build-oc-8.html#post12050983

To sum up at the bottom, I've had to try a lot of different settings now but changing the Uncore to 12x (1,600mhz) manually from the default auto suddenly made the game and testing lockups / hangs go away for now. Sound still stutters at max CPU load but I'll take that over insta-locks in a game or the desktop lol.

If I want to hit 4.0+ do I need more voltages or should I just set the Uncore higher? Also despite these being redline modules @ 1,600, how much higher should I take them before dropping to the 2:10 ratio? They're running at rated speed now at any rate which is cool as the rest of the system is quite OC'd a bit.

I'm not 100% sure about your 5620's since I have older cpus. But It would seem to me that you need more vcore. I have both of mine set somewhere around 1.4 in the bios, seeing 1.36 on cpu0 and 1.39 on cpu 1.
I'm at 203bclk. I've got the vtt's both at 1.375
Cant help you with the sound issue. I've never experienced it.
 
Some notes:

- 1.425 IOH is higher than you should need for 200 bclk.
- DRAM ratio - 2;10 (DDR-1333 is whatthe setting is called) is faster than what you should have it on now (DDR-1066 or 2:8.) Slower would be DDR-800/2:6.
- MCH Strap - All of mine are on Auto, with memory timing set manually on all but one.
- Your memory should not be your hold-up, even over 200 bclk. You could back your multiplier down to verify this. I can't imagine it not being able to do 1640 or so. If in doubt, back it down to DDR-800 to see how high you can get bclk.
- I have never had any luck with the signal tweaks. YMMV however.
- You are already in the same bclk range as all of mine, so I am not much help. Plus, i don't care about sound. Mine could sound like hell and I wouldn't know.
- I have fits with the onboard NICs going out in Windows, to the point that one of mine has a USB wireless NIC just to keep a network connection. You may start running into this, and I have no idea how to fix it.
- You are still pretty low on your Vcore, so you do have some room there.
- At some point, you are probably better off to quit messing with it. I think you could try forever for 1 more bclk and waste a ton of time. I know I have.
 
Some notes:

- 1.425 IOH is higher than you should need for 200 bclk.
- DRAM ratio - 2;10 (DDR-1333 is whatthe setting is called) is faster than what you should have it on now (DDR-1066 or 2:8.) Slower would be DDR-800/2:6.
- MCH Strap - All of mine are on Auto, with memory timing set manually on all but one.
- Your memory should not be your hold-up, even over 200 bclk. You could back your multiplier down to verify this. I can't imagine it not being able to do 1640 or so. If in doubt, back it down to DDR-800 to see how high you can get bclk.
- I have never had any luck with the signal tweaks. YMMV however.
- You are already in the same bclk range as all of mine, so I am not much help. Plus, i don't care about sound. Mine could sound like hell and I wouldn't know.
- I have fits with the onboard NICs going out in Windows, to the point that one of mine has a USB wireless NIC just to keep a network connection. You may start running into this, and I have no idea how to fix it.
- You are still pretty low on your Vcore, so you do have some room there.
- At some point, you are probably better off to quit messing with it. I think you could try forever for 1 more bclk and waste a ton of time. I know I have.

Hey Musky and Nimby, thanks for the replies.

Yeah right now the RAM is still at 2:8. Again I think the biggest thing was the Uncore had not adjusted on auto but on the 12x setting it's working for me at the moment. Ah whoops, guess I meant the slower 2:6 RAM setting then, my bad.

I had the IOH that high because several different posts stated that it's the northbridge which assists in the Realtek sound output and if it's being overclocked, it'd need more voltage to stabilize the sound.

Yeah I don't want to spend eternity on this by any means. If I can hit 4.0 though, I think I'll be happy. Even if not, this is still what I'd like to think, quite a decent OC. First time really doing it as well.
 
Basically follow what Shamino put for 200bclk in the SR-2 template thread over at greenland (EVGA). :D
I can +1 a few of his settings...I am using the signal tweaks he listed [-85 and -16], in addition to IOH Vcore at 1.4v. The rest is all cpu dependent and if you're doing isolated overclocking like most x58 guides recommend, you should have a fair idea what your Vtt needs to be for a certain bclk, what vdimm the ram needs to run at that speed, and then what Vcc you need to raise your multi back up to it's max. The rest I'll echo most of what musky said....I'm personally running 17x200 (18 w/Turbo), 1.2 boot/eventual Vcore, 1.3 boot/eventual Vtt, 1.4 IOH Vcore, 1.65 Vdimm, DDR-1333 (to run 2:10 with my 2000 9-9-9-24 Corsair), MCH Strap is set to Auto, most of the other stuff is set to auto other than what I've listed otherwise. ;) Check the SR-2 owners thread here in DC land if you want plenty more examples...everyone who's participated in that thread posted their bios settings. :)
 
Well guys, spent a few more hours testing and behold! I'm not 100% stable but I did pass a few basic tests so once I can make sure games and stuff actually let me play them fine, I'll probably do a longer endurance test of some sort.

1161556974_BiwiM-S.jpg
1161556988_zVM9u-S.jpg
1161557011_vefuY-S.jpg


http://dfullerton.smugmug.com/photos/1161556974_BiwiM-X3.jpg
http://dfullerton.smugmug.com/photos/1161556988_zVM9u-X3.jpg
http://dfullerton.smugmug.com/photos/1161557011_vefuY-X3.jpg

Even though I'm ever so slightly below the magic #, I've been able to successfully pass two runs of 10 calculations of IBT w/16 threads and more importantly IMO, one test of two calculations of IBT w/16 threads on Maximum! I know it's not the most in-depth test, but I'm happy to see my system get through that.

While I know IBT puts an abnormal load on the cores, I don't think I'll probably be going too much higher as the hottest cores are at 75-77C. I would imagine through that other programs wouldn't make them heat up quite as much.

Settings:
---Main
210 BCLK x19
MCH Strap = DRAM
DRAM Freq = 1066
QPI = Auto
Uncore = 12x (was fine from 200 to now @ 12, but since I locked up while typing the original post I'm giving it a go at 14x / 1,867 mhz). lol
Rest at Auto or defaults.
---Voltages
CPU VCores @ 1.35
CPU VTTs @ 1.35
IOH @ 1.425
IOH PLL @ 1.875
DRAM @ 1.65
Rest at Auto
---Signals
I can't recall their names so you get #s.
-96, -16, -31, -31, -31, -31, Auto, Auto, -31, Remaining 8 at Auto

While it may not help everyone, I'm finally happy to have hit my goal ^_^. Only real problem now (aside from making sure I'm stable for sure) is to fix the sound when CPUs are loaded. Don't want to raise the IOH more and the -31 tweaks didn't help. Oh well, at least things sound great when at idle.
 
Last edited:
Ok, updated a crap-ton of BIOS notes on my OCN thread. Basically, it was a re-attempt to play around at 210 BCLK.

http://www.overclock.net/intel-build-logs/864314-melchior-project-sr-2-build-oc-8.html#post12074287

The strangest thing and what I think made me go from kinda-stable-but-we-still-want-to-lock-up-on-you to woot we work great was the fact that my DDR-1066 settings forced the RAM to set to defaults very close to, but not exactly to, what my Redlines are rated for. The 4th value (tRAS) was on 20 when the kit is rated at 8-8-8-24 (all other values were correct on auto, mind you). I simply changed that and so far, I've been stable for a couple hours with mild folding / Rosetta, browsing, and successfully completing one giant run of 50 calculations of Standard IBT 16 threads. Tempted to see if I can repeat my test of 10 runs of Maximum IBT but it's finding when I can leave the system semi-unattended for an hour I can waste, haha. Also bumped Uncore from 12x to 13x (1,867 mhz), otherwise all other values are identical to what's above except I brought the IOH PLL back down to 1.80.

Which brings me to another question really. What tests do you guys do to consider yourselves "Stable"? I know there's stable enough to game, stable enough to fold / crunch, and stable enough to be a critical solid server as strong as a brick ****house. Would my runs of IBT be good enough you think? I honestly don't feel like killing days to do endless Prime95 runs.

To be honest, I want to finish these graphs for my 4.0Ghz spot and be done. Seeming that it took me 0 work (just a lot of testing #s slowly over a few weeks) to get from 133 BCLK up to 180, a little work from 180 to 200 (a couple tinkering days straight), and then a LOT of work from 200 to 210 (several days of tinkering straight), I can see that I'm really having to push and tweak to a hair to reliably go higher than 4.0Ghz. I think I am more than happy to call it a day here, presuming it proves to be stable both short term and long term 24/7.

This reminds me of tuning a sports car, despite me knowing squat about cars. But I think you guys get the idea. I'll of cours keep you updated as I still have another 580 coming in and a few more photos to do and such. Thank you all again for such a **wonderful** thread and help.

**Edit - Just made the OCN Core i7 Stability Club (unoffically for now). I mean, I know these procs are Xeons / servers and should be stable but either way I passed their Linx test.** Woot.
http://dfullerton.smugmug.com/photos/1162430560_vReR6-X3.jpg
 
Last edited:
I consider my build stable after I've used the system daily for a while without a crash (maybe a month or so). That includes running Bigadv non stop, photo editing, gaming, regular web browsing, pretty much anything I can think to throw at it.

My initial stability tests are always a few runs of IBT in maximum, and I find that running CIV 5 for a few minutes is another good way to incite a crash if it's going to.

Now the way my SR-2 has been, I would be confident enough to host some VM's or a web server from it at 200 bclk.
 
A few interesting things I've seen this morning over at E land...
The best OCing bios on the SR-2 is A47
The SR-2 guru himself, MIBW :) has posted in that thread and thrown some evidence to specify otherwise....I would say I agree with him, I haven't had 200bclk problems on A46, A47 or A49 myself, and I know (like MIBW said) A49 finally having Uncore adjustments is a big positive on that end. Just something to keep in mind though....I may scale back to A47 later in yet another attempt to disable NUMA. :eek:

And on that note...
2 HD6950, will not post w/ NUMA disable.
This has me wondering if having 2 card slots populated is why I can't disable NUMA either. :eek: I'm going to yank out a card when I get home this afternoon and give it another shot.
If I'm able to disable NUMA with only 1 card, then I will probably be switching to a 1 card gaming solution after we get our tax refund (and if I do, it will be an ATi card).
 
And on that note...
2 HD6950, will not post w/ NUMA disable.
This has me wondering if having 2 card slots populated is why I can't disable NUMA either. :eek: I'm going to yank out a card when I get home this afternoon and give it another shot.
If I'm able to disable NUMA with only 1 card, then I will probably be switching to a 1 card gaming solution after we get our tax refund (and if I do, it will be an ATi card).

Keep us posted on that one. It is true that I only have a single video card in all 4 of mine, so that would explain why I have not had issues.
 
I will say I have yanked a card and ran single gpu in trying to fix the gaming issues I've had, and that didn't fix that....but I never would have guessed about multi gpu setups not allowing you to disable NUMA.

I can't wait to get home to try that and if it works (disabled) with 1 gpu yanked I swear I'm gonna do a fucking backflip (or die trying). :D
 
I will say I have yanked a card and ran single gpu in trying to fix the gaming issues I've had, and that didn't fix that....but I never would have guessed about multi gpu setups not allowing you to disable NUMA.

I can't wait to get home to try that and if it works (disabled) with 1 gpu yanked I swear I'm gonna do a fucking backflip (or die trying). :D

really wish this will solve it for you. But could you run numa disabled at stock speed with a single ram module or it just never worked?
 
2 HD6950, will not post w/ NUMA disable.[/url]
This has me wondering if having 2 card slots populated is why I can't disable NUMA either. :eek: I'm going to yank out a card when I get home this afternoon and give it another shot.
If I'm able to disable NUMA with only 1 card, then I will probably be switching to a 1 card gaming solution after we get our tax refund (and if I do, it will be an ATi card).

That is me. I was hoping I would get more answers over here. :p

If I remove the second HD6950 it will post no problem w/ NUMA disabled. As soon throw the card back in it will not complete POST. If I enable NUMA it is perfectly stable w/ both cards installed. Although the hit in PPD is something I can't live with. :mad:
 
GUYS.....IT FUCKING WORKED!

1 260 AND I CAN POST NUMA DISABLED!
2 CARDS IT DOES NOT POST....I GET THE 0C -> 2A LOOP.
HOLY SHIT!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

(sorry for all caps but good God this has been my dream for so long now!!!!!!!!!!!!!!)


really wish this will solve it for you. But could you run numa disabled at stock speed with a single ram module or it just never worked?

No, it just never worked. Even with a cmos clear/bios reset, everything at stock settings/auto...I STILL could not disable NUMA.
I am fully booted right now, all stock, 1 gpu removed (1 being used)...NUMA disabled.

THIS IS IT...THIS IS THE SECRET FIX FOR ME (at least).
It's not even a case of not having all the power hooked up either....I have all the power slots hooked up on the SR-2, including the extra 6pin for the PCIe slots.
Obviously it's doubtful EVGA has a solution for why this happens or how to run multiple cards AND disable NUMA, but for the love of Baby Jesus now I finally know how to successfully get a boot (hell, even a POST), with NUMA disabled.
My ram is still running 2000 9-9-9-24 1T also, I didn't have to loosen timings or scale back to 2T either like MIBW has recommended me try to do.
 
<VBG> Congrats on finding a solution to this long standing issue zero. I grinned when reading your post cause you can just feel the excitement in what you were saying!

Another barrier dissolved, at lest with a slight workaround...
 
I'm wondering now though, Falcon_CMH posted that he's run 580 SLI with NUMA disabled.
I replied regarding slots? I've always had my cards in 1 and 5 so there was a bigger airflow gap; I don't think I've ever tried 1 and 3 although I don't see why the slot population/arrangement should matter.

I'm going to try 1 and 3 just to eliminate that from the equation though....it must be something depending on cards, either that or some of us just have boards with picky slots or something. :confused: I dunno, I'm just glad to be able to finally disable NUMA. :D
 
Great news zero, we have all been there and it just gnaws at you, so big smiles here too!

FYI I had no trouble with NUMA disabled with SLI of 2 GTX 460s in slots 5 and 7. (for airflow)
 
Well I tried 1+3, 1+5, and 1+7, and got 0C -> 2A -> reboot cycle again.
Just to test other slots, I put a single card in 3 and was able to disable NUMA and post successfully.
I also tried with and without the SLI bridge, and I even tried using a different card and booting with 2 different cards (260 SC in slot 1, 8400GS in slot 3). No dice.

Again, I'm clueless as to "why" but having a solution after all this time is definitely worth any "wondering". :)
Besides it'll give me more reason to upgrade to a single gpu solution for gaming. ;)
 
Or build a gaming 2600k system so the SR-2 can fold in peace ;)
 
Dude you don't even know how close I am to doing that. ;) (or at least how MUCH I want to do that :D)
I just need to try to convince my wife that Family Feud, Wheel of Fortune, and Puzzle Saga on Facebook need a 2500k or 2600k instead of a q9450. :D
 
I'm wondering now though, Falcon_CMH posted that he's run 580 SLI with NUMA disabled.
I replied regarding slots? I've always had my cards in 1 and 5 so there was a bigger airflow gap; I don't think I've ever tried 1 and 3 although I don't see why the slot population/arrangement should matter.

I'm going to try 1 and 3 just to eliminate that from the equation though....it must be something depending on cards, either that or some of us just have boards with picky slots or something. :confused: I dunno, I'm just glad to be able to finally disable NUMA. :D
I have tried every variation of slots I could imagine trying to get one to work and it definitely did not help. I am going to try both GPUs on my second SR-2 this weekend to see if it is board specific or it just doesn't like certain pairs of GPUs. If it is specific to just one of my SR-2s I will be an extremely happy man.
Or build a gaming 2600k system so the SR-2 can fold in peace ;)
If I can't find a resolution by the end of the weekend this is going to be my resolution.
 
Dude you don't even know how close I am to doing that. ;) (or at least how MUCH I want to do that :D)
I just need to try to convince my wife that Family Feud, Wheel of Fortune, and Puzzle Saga on Facebook need a 2500k or 2600k instead of a q9450. :D

Step 1: remove antivirus
Step 2: frolic in questionable corners of the interwebs; click on everything
Step 3: fix non functional system with much needed upgrades

Edit:
Step 4: sell "broken" parts on the internet to the fools who will buy anything
Step 5: profit
 
Last edited:
I have tried every variation of slots I could imagine trying to get one to work and it definitely did not help. I am going to try both GPUs on my second SR-2 this weekend to see if it is board specific or it just doesn't like certain pairs of GPUs. If it is specific to just one of my SR-2s I will be an extremely happy man.

Please let me know what you figure out.....I hope some of us didn't just get bad batches of the board or something. I only have the 1 SR-2 but I'd hate to be down for a couple weeks while waiting for a replacement.

Step 1: remove antivirus
Step 2: frolic in questionable corners of the interwebs; click on everything
Step 3: fix non functional system with much needed upgrades

Edit:
Step 4: sell "broken" parts on the internet to the fools who will buy anything
Step 5: profit

WIN :D
+1 awesome post sir, beers for you next time you're in the STL ;)
(even better with the edit ~ LMAO)

edit looks like with no more NUMA I reduced TPF by 30 seconds and upped PPD by 2500k

NUMA Enabled
nogpus.png


NUMA Disabled
nomorenuma.png
 
Last edited:
This is some useful information. I wasn't able to disable NUMA on my SR-2 with dual GTX 295s, but I was never too worried about it. Now I know the (probable) reason.
 
Yeah, definitely try it and see if it brings you success. :)
[crosses fingers for you] ;)
 
wow....BRB rebooting. cross your fingers for me.

Edit. unbelievable. Worked like a charm. Booted right to windows with numa disabled and no other adjustments.

Big thanks to everybody that contributed to figuring this one out. Now we need to get EVGA to fix it with a BIOS release.
 
Last edited:
Congrats jebo!! Enjoy!
143.gif

I'm so glad I saw lowfat's thread. Sad I can't figure out a fix for him, but glad that this solves the "can't disable NUMA" problem a lot of us have had. ;)
This has gotta be either a bios or card issue, that's for sure. I'm almost tempted to go and try A50 because I know A49 didn't allow me to disable it with 2 cards installed. Maybe A50 will?
Or maybe the fix will be in A51 or whatever's next. (if there is a fix)
 
Small problem, after a certain frequency of base clock, it seems that some USB ports start acting up, for instance, at 200x19 the light on my mouse turns on and off and the mouse pointer jumps about all the time, if I lower it a bit, then the mouse only acts funny while folding. Is there a particular voltage which I need to tweak to fix it? (IOH is at 1.4, other settings are according to shinamos 200 bkc template)
 
Hmm. That might explain some problems I've been having with usb keyboards and mice. I thought my hub was going bad.
 
I haven't had any problems @ 200bclk with USB devices (mice, keyboards, printers or scanners). :confused:
 
Small problem, after a certain frequency of base clock, it seems that some USB ports start acting up, for instance, at 200x19 the light on my mouse turns on and off and the mouse pointer jumps about all the time, if I lower it a bit, then the mouse only acts funny while folding. Is there a particular voltage which I need to tweak to fix it? (IOH is at 1.4, other settings are according to shinamos 200 bkc template)

IIRC isn't there a floppy power connector somewhere on the board to provide additional power to the USB ports when doing heavy overclocks??
 
oh there is! That's very thoughtful of them :)
 
IIRC isn't there a floppy power connector somewhere on the board to provide additional power to the USB ports when doing heavy overclocks??

Again....200bclk, no issues with USB here, and if there's a floppy power in jack, I'm not using it. :confused: DeFex, you sure your issue is bclk related and not voltage related?
 
I have not seen USB issues either...to the point that 2 of mine now use a wireless USB dongle for network connection. I am not using the extra power connection, either.
 
Yea but evga didn't put the floppy power connector on there for no reason. I think its possible.....

I just don't wanna reboot yet to test.
 
I have not seen USB issues either...to the point that 2 of mine now use a wireless USB dongle for network connection. I am not using the extra power connection, either.

Yea but evga didn't put the floppy power connector on there for no reason. I think its possible.....

I just don't wanna reboot yet to test.

If the NICs are powerd off the same loop that the USB (and thus this floppy power header) are on, it might fix the network drop issue.
 
Back
Top