• 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

The board hardware seems to be consistent, in that everything labeled CPU0 (slot, CPU jumpers, etc.) maps together. It also seems like my LCD is showing the real CPU0 temp (the cooler one, not the hotter one near the backplane) but I could be mistaken. Later today I'm going to get out the multimeter and investigate further.

My other SR-2 did not show this 10-degree difference between CPUs, with either the X5650s or L5640s. It sounds like the difference in temps is not unusual, but I'd like to understand why. I'm going to try swapping CPUs and reorienting the heatsinks to see what changes. I may add a second fan too.
 
A note on the modded A49 bios with adjustable uncore...also known as "uncore can kiss my ass!"

I ran two benchmarks on my faster machine before starting all of this - LinX 1024 Mb memory and the average of 5 frame times using MIBW's saved P2686. I then update one of the bioses on the board to the modded A49 bios. One thing that became very obvious is that this almost certainly the reason for the 200 bclk wall most of us are seeing. This particular machine was running 203 bclk, which is actually pretty good. With the new bios and 16x uncore, i was able to get bclk as high as 211 into Windows and running LinX 1024 Mb. The problem is that this required 1.42 Vcore, 1.45 vtt, and 1.45 IOH. Actual Vcore was around 1.41, actual Vtt was around 1.425, and actual IOH was around 1.425. At these voltages, stablity and performance were inconsistant. Easing back down allowed me lower voltages across the board and better stability. I thought I found the sweet spot at 206 bclk and 1.4V for Vcore, Vtt, and IOH. LinX times were 0.4 seconds faster that the reference numbers, which is a significant percentage improvement (times were under 11 seconds to start with.) The problem is that the folding performance was actually 7-8 seconds/frame slower on the reference 2686. All other settings showed similar result - better LinX times and worse folding times. Here is the real kicker - the A49 bios set up exactly like my original settings, assuming uncore is 20x and MCH Strap is DRAM speed (these two settings do not exist in the original A41 bios) yields slightly slower folding times than the reference times.

Long story short, after 12 hours or so of tweaking, I am back to the original A41 bios with my original settings. I can post them if anyone wants, but I don't think anyone else is running Kingston HyperX and memory is a big factor in what settings work.
 
Temps - to think I was actually running my mid-heatsink fans on the hotter cpu a fraction slower than the ones on the cooler CPU and I didn't notice! :eek: :mad:

BUt I have been telling people the wrong thing, and this makes me very glad I am running nekked on the second rig. Now I have to get my head around the hotter CPU being the one to the rear of my Lian li case. I hadnt worried about it, because I thought it was the cool one, but now I realise it is the one right above the GPUs, and furthest from the fresh air. Glad I made a ghetto mod fan box to blow cool air from floor level right into the intake area.

@musky- I found 18x or 19x the only uncores worth working on, as anything lower killed folding perfromance more than any possible gain in baseclock. Shame you could not find a combo that gave you a benefit in either perfromance or stability. My current combo is that on the x5660 rig I was able to go higher and remain stable, and on the x5650 I am taking a slight performance hit in order to improve 6701 stability.:mad:

But it is a hell of a lot of work for perhaps little gain.

I really need to update my settings posts to note the big discovery (to me as noob folder!) that 6701s uncover weaknesses that Linpack/Memtest/HCI memtest/Prime95 do not. :eek:

And with me simply adding vcore/vtt did not help, and lowering uncore did (cross fingers)


@10e - I think 14x uncore might kill you dead in the water for folding times... like back to where you before a genius whose name I forget suggested we disable NUMA...
 
Well this little 'quirk' explains why my CPU0's PWM was showing 20c higher temp than CPU1's PWM, even though I have a 120mm fan running on full sitting in front of CPU0's PWM. Weird.
 
@SazanEyes: I'll have to check my mobo again and ensure that he LED shows the CPU 0 temperature as well. It was late Saturday night/Sunday morning when I was checking out which CPU was which and I was a bit bleary eyed and sleepy. I've since added a second fan on the back CPU.

@MayIBworthy: I am not sure about the performance, because since the change to uncore adjusting A49, all I've gotten is 670x units :( but they seem consistent with what I was getting before, except with a slightly lower CPU temperature..

There IS something funny with LinX on these setups. It doesn't push the CPUs any harder than folding does it seems. Whereas with single CPU configs I'll get at least 3 or 4 celsius higher CPU temperatures, on the SR-2 it's the same if not a little bit lower temperature-wise with LinX vs. folding.

Another thing I've noticed is the With/No Vdroop function seemingly does nothing as well.

EDIT: Since I just got another p6701 unit on the SR-2 I "oneunit" flagged it on restart and will try the P2686 test unit once this is done and see what different unCore speeds give.
 
musky said:
A note on the modded A49 bios with adjustable uncore...also known as "uncore can kiss my ass!"
[snip]
Click to expand...

I hate to say this is precisely why I'm still running A47...it works. I want to disable NUMA but otherwise it works fine. I set 16x uncore frequency, that's what it shows up as. (3200mhz with my 200bclk).

Either CPU-Z is dead wrong or my board is an anomaly. :confused:

Can you bios back-flash? I thought that was a no-no. (or did you just switch over to bios 2 or 3?)
 
LinX always gives me at least 5C higher temps vs. folding.

I have stock BIOS (A43) on 1, A49 on 2, and A49 unlocked uncore on 3. I see no difference between A49 hacked and unhacked (without messing with uncore), so I'm going to try to flash A47 on 2. The new board definitely is not overclocking as well as the old one, so I'm trying to isolate the problem.

It seems like the problem is heat on second CPU so far, and the different RAM does not seem to be an issue (running at 1600 now, but haven't tried disabling NUMA). The older board runs both CPUs at almost exactly the same temps -- individual cores vary, but the peak core temp on each CPU is either the same or 1 degree off. When I had the X5650s in that board I noticed a little variation, so I think one of the CPUs may run a little hot. I'm going to try swapping them to get the "hot" CPU in the "cool" slot and see if that makes any difference.
 
zero2dash said:
I hate to say this is precisely why I'm still running A47...it works. I want to disable NUMA but otherwise it works fine. I set 16x uncore frequency, that's what it shows up as. (3200mhz with my 200bclk).

Either CPU-Z is dead wrong or my board is an anomaly. :confused:

Can you bios back-flash? I thought that was a no-no. (or did you just switch over to bios 2 or 3?)
Click to expand...

I tried A49 for a quick bit, and then flashed back to the A47 BIOS and that's where I've been ever since. As a side not, I do have NUMA disabled on mine...
 
I'm still a little confused several things on this board. CPU "0" runs 7-10c hotter than the other. I've tried switching the HSF's around, rotating them, changing the fan speed. Nothing helps. Its not a big problem even if it's a bit annoying.

I'm happy to say that I have it running at 200blck stable. I was worried about the PWM being too hot, but I got it under control by rearranging some cabling.

I updated from the A43 to A49 bios to check out "uncore" however I never found the option. Worse with the new bios I was having problems with my memory timings. There was a new "MCH strap" setting that messed with the timings and subtimings. Even after reverting to optimized defaults, I couldn't get it stable.
So I selected bios2 and started over with it. I have the VTT and IOH at 1.35, the VCORE is at 1.25. I couldn't see a difference with NUMA on or off, so I have it off now.
I turned the memory to 1.63 in the bios, because it was showing higher at 1.65. On occasion I can see in cpuz that the cpu's are at a 20x multi, then it switches back to 19 and stays. I'm not sure if these are 1x or 2x turbo (pretty sure it's just 1x). I know that it stays happily at 19x almost all the time.
I don't have an exact PPD figure yet, but it seems to be right at 30k ppd on some of the many 6701's I've picked up. I saw 49k on a 6025. I'm still waiting for a bigadv.

Hopefully next week I can do some more experimenting and get it to 210bclk. This is the A2 revision board. Although I don't think there are any physical differences.
 
Last edited:
nimbyfaygo said:
I'm still a little confused several things on this board. CPU "0" runs 7-10c hotter than the other. I've tried switching the HSF's around, rotating them, changing the fan speed. Nothing helps. Its not a big problem even if it's a bit annoying..
Click to expand...
Is this the first dual-socket machine you're running? Many of my builds have one CPU running warmer than the other. In some systems, the difference is so great it is very noticeably by touch. It is not always the case but often dual-socket machines will run one CPU significantly warmer than the other. Don't know how common it is with the SR-2 (don't have one) but it is not out of the ordinary with multiple sockets.
 
nimbyfaygo said:
I'm still a little confused several things on this board. CPU "0" runs 7-10c hotter than the other. I've tried switching the HSF's around, rotating them, changing the fan speed. Nothing helps. Its not a big problem even if it's a bit annoying.
Click to expand...

Well you just answered my reply...I was going to ask about rotating HSF if you had them in standard front to back blowing formation (like mine are). Most SR-2 pics I see, people put them horizontally but I didn't think that was a good idea since at least 1 of the cpus (depending on how long your gpu was) would intake heat from the back of the PCB of the topmost gpu....in my case (having a 260), that would be at least a few extra degrees when @ load. Because of that, I decided to run the HSF in standard vertical front to back airflow orientation.

I've noticed the difference in cpu temps, the thing I think that's funny is how it seems like cpu0 is the hotter one when it should probably be cpu1 that's hotter since it would intake some of the exhausting heat from cpu0....but I dunno if that's because they're mislabeled or something else is going on. Either way, I'm still well within acceptable specs (if not ridiculously low). I still see around 50c at load f@h and I'm more than happy with that since that just puts one in mid 40s or upper 50s at worst.
 
nimbyfaygo said:
I updated from the A43 to A49 bios to check out "uncore" however I never found the option. Worse with the new bios I was having problems with my memory timings. There was a new "MCH strap" setting that messed with the timings and subtimings. Even after reverting to optimized defaults, I couldn't get it stable.
So I selected bios2 and started over with it. I have the VTT and IOH at 1.35, the VCORE is at 1.25. I couldn't see a difference with NUMA on or off, so I have it off now.
I turned the memory to 1.63 in the bios, because it was showing higher at 1.65. On occasion I can see in cpuz that the cpu's are at a 20x multi, then it switches back to 19 and stays. I'm not sure if these are 1x or 2x turbo (pretty sure it's just 1x). I know that it stays happily at 19x almost all the time.
I don't have an exact PPD figure yet, but it seems to be right at 30k ppd on some of the many 6701's I've picked up. I saw 49k on a 6025. I'm still waiting for a bigadv.

Hopefully next week I can do some more experimenting and get it to 210bclk. This is the A2 revision board. Although I don't think there are any physical differences.
Click to expand...

Is everyone running A47? I flashed to A49 before attempting any overclocking and I haven't really understood how the MCH strap setting actually works.

If I set my timings manually, do the different MCH strap settings actually do anything? I've just left it at auto so far.
 
MCH strap - does effect performance, even with timings set manually - come to think of it I say that but there was always one mem setting left set to auto - the back to back cas delay - because I never knew what to set it.

1867 wont boot, 800 is too tight generally, and I have it nice and tight at 1067 on my 2:8 SR2 ans at 1333 on my 2:10 box. Setting it to 1600 at first while you get your OC stable, then lower it last to as low as you can while staying stable/bootable.

A49 (hacked) is stable at 4.3Ghz for me, so as far as I am concerned - update!

Linx temps are all over the place depending on settings.
original.jpg

Useful tip - Very high 32 thread on Intelburn2.5 test - the valleys in the graph, not the peaks, are equal to folding temps.

Now that I know which damn CPU is which :eek:, I did a lot of work on fans yesterday, and got my temps down a lot.
 
Well I'm seeing 47,000ppd on a 6069. LOL 1:47 frames are damn near neck and neck with my GPU

195x19 (3705mhz) Steady temps at 68c and 61c, that's in my overly cramped case with the board upside down.

Much pleased I am, pushing 9x higher ppd than the Q6600
 
may i be worthy said:
MCH strap - does effect performance, even with timings set manually - come to think of it I say that but there was always one mem setting left set to auto - the back to back cas delay - because I never knew what to set it.

1867 wont boot, 800 is too tight generally, and I have it nice and tight at 1067 on my 2:8 SR2 ans at 1333 on my 2:10 box. Setting it to 1600 at first while you get your OC stable, then lower it last to as low as you can while staying stable/bootable.

A49 (hacked) is stable at 4.3Ghz for me, so as far as I am concerned - update!

Linx temps are all over the place depending on settings.
original.jpg

Useful tip - Very high 32 thread on Intelburn2.5 test - the valleys in the graph, not the peaks, are equal to folding temps.

Now that I know which damn CPU is which :eek:, I did a lot of work on fans yesterday, and got my temps down a lot.
Click to expand...

Your IBT is what my temps look like when using LinX (the peaks and valleys). Have you updated you linpack_xeon32 / linpack_xeon64 files to the latest from Intel?
 
Aix. said:
Your IBT is what my temps look like when using LinX (the peaks and valleys). Have you updated you linpack_xeon32 / linpack_xeon64 files to the latest from Intel?
Click to expand...

Sorry - I was being sloppy - that screenie is on Intelburntest2.5, which I believe is a GUI for running intel linpack - dunno what version, never ran it otherwise.

Doe the latest version run more consistently temps wise?
 
OK, first experiment after knowing which CPU was which. I checked voltages with my Fluke and they seem correct, and match up well with E-LEET. The hot CPU is not getting extra voltage. I reversed the fans on the TRUEs, so that CPU1 (the hot one) is blowing onto CPU0 instead of the other way around. This had a definite effect on temps on both CPUs. CPU1 went from average 70C to 66C with the hottest core at 67C, and CPU0 went from 60C to 63C with the hottest core at 66C. In short, temps are not much cooler on the hot CPU, but they are much more uniform across CPUs now.

I'll try more tomorrow once the P2684 is complete. I want to reoganize some wires and remove the GTX 295 closest to the CPUs, since it generates a lot of heat even while idle.
 
Aix. said:
Is everyone running A47? I flashed to A49 before attempting any overclocking and I haven't really understood how the MCH strap setting actually works.

If I set my timings manually, do the different MCH strap settings actually do anything? I've just left it at auto so far.
Click to expand...

Same here.

may i be worthy said:
MCH strap - does effect performance, even with timings set manually
Click to expand...

I don't know what to set it to myself, that's why I left it on auto. :confused: Maybe that's why I can't boot after disabling NUMA? [double confused]
Running 200bclk, 2:10 divider.
 
Aix. said:
Have you updated you linpack_xeon32 / linpack_xeon64 files to the latest from Intel?
Click to expand...
Would you just replace the "stock" IBT\LinData\<x86\x64>\linpack<32\64>.exe with the associated linpack_xeon<32\64>.exe from the intel link you provided?
 
Hey guys, just wanted to say thanks for all the people who helped me with my questions about the SR-2 and Xeons ES chips.

I've got a brand new pair of x5680s for sale or trade for a pair of L5640s, got a special price just for [H] folders. PM me if interested :)
 
It appears that all it does is give you the unlocked uncore. Since I was actually slower with the hacked A49, I will be passing on this one.
 
I have always approached it by getting Vcore, Vtt, and Vmem the same for both CPUs. For the one I remember, that meant Vcore0 at 1.38125, Vcore1 at 1.400, Vtt (both) at 1.350, Vmem0 at 1.65, and Vmem1 at 1.64. That gives me a Vcore of around 1.375 for both, Vtt of 1.33 for both, and Vmem of 1.67 for both. My temps are pretty consistant between processors (<5C), so that is good enough for me. Temps have never been my problem with the L5640s, though.
 
I briefly toyed with trying 2 copies of IBT, but gave it aways as something I would do after I found my overclocks, then optimise.As we have seen, knowing just what settings are effecting which actual CPU can be a PITA. As it turned out, I spent most of my time on memory issues, so I didn't really need the extra hassle. But having worked on them so hard for so long, watching the buggers like, just fold is kinda boring ;)

Also as my comfort factor has grown, so have my volts and temps, so coming into summer here I could use some heat relief...assuming I don't immediately use it to try for more GHz :rolleyes:
 
That's a very useful graphic. It's interesting that not only are CPU temps reversed, but NUMA nodes are as well (unless you have disabled NUMA).
 
I gotta dig back through this thread this weekend and work on disabling NUMA and figure out what to adjust so my system boots when I do.

Oddly enough ~ I get a lower TPF (by about 30 seconds) running a 2686 with my 260's folding w/gpu2 at the same time vs. running 2686 only. :confused:
It's about 15:19 TPF w gpus vs 15:48 TPF without.
 
that's probably just random variation

I never got numa disabled either, but I chalked it up to my mis matched CPUs.
 
It holds steady though....I've let it run at least 10 frames without the cards, and 10 frames with the cards. :confused:

Ok, someone help me here....what exactly is NUMA tied into regarding ram settings?
(digs back to look at MIBW bios screenshots)
My settings are:
[[[a47 bios]]]
Code: 
(frequency/voltage control)
cpu freq 200
pcie freq 101
cpu multi 17
qpi freq 4.800gt
memory freq ddr-1333
mch strap auto (should i put it on something else?)
cpu uncore freq 2133mhz 16x

(memory config)
dram tCL 9
dram tRCD 9
dram tRP 9
dram tRAS 24
command rate 1
Back-to-Back CAS delay 8 (boots 9)
***everything else auto***

(voltage config)
vdroop --  without vdroop
bootup/eventual cpu0 vcore 1.20
bootup/eventual cpu1 vcore 1.20
bootup/eventual cpu0 vtt 1.30
bootup/eventual cpu1 vtt 1.30
cpu0 dimm voltage 1.65
cpu1 dimm voltage 1.65
cpu0 PLL voltage 1.65
cpu1 PLL voltage 1.65
IOH voltage 1.4
IOH PLL voltage 1.8
ICH I/O voltage 1.5
ICH voltage 1.125

(signal tweaks)
IOH QPI 0 signal -85
IOH QPI 1 signal -16
***everything else auto***

Is there something I'm missing that is the deal breaker for NUMA disable working? :(
 
Last edited:
The one thing about your setting is drastically different than my setting on my two machines with NUMA disabled is that your Vcore is significantly lower than mine, and your Vtt is quite a bit lower. Vtt may affect NUMA. Is your IOH voltage set to anything? Mine is at 1.4. I also do not have the signal tweaks set at all.
 
Yeah, I must have deleted my IOH setting...it's 1.4.
You may be onto something with Vtt being hand in hand with disabling NUMA since Vtt voltage probably has some bearing on the IMC. I'll try raising my Vtt a notch tomorrow and see if that fixes my "no boot with NUMA disabled" problem.

Thanks for the valuable input (as always :)), musky ;)
 
@zero2dash - concur with Mr Musky about some vtt, but also I would set your MCH to 1600, and/or set your timings completely manually.

I just think your memory is very high - 2000MHz is killer - if you can't boot I would loosen to 11-11-11-24 etc 2T, or for starters try auto all timings and MCH 1600, which should set things to 11's anyway.

If you are on A49 or earlier the lower Uncore you have set won't actually do anything help you, but if you are on the latest A50 or the hacked A49 then lowering uncore will help you boot. Don't go below 18x - not worth it for performance loss, might as well clock down a notch or two.

Once you have proved you can boot at all, then start tightening timings back.
 
I'll work on setting MCH strap.
I'm running 2000 rated ram. This. running 2 sets for 12gb total.

re: changing uncore to 18....I've always had more luck with uncore at 16 since it's supposed to be 2x ram divider. technically i could try to raise this to 20 now that i'm running 2:10 with this ram, I may try to do that as well. Regardless I'll see if I can do 17,18,19, or 20. I assume changing my uncore will definitely require more Vtt though and TBH I'm not going to push more than 1.35 through these chips so I'll go as high as I can on the juice I find ok. ;) normally I'd be ok with going a notch or two above but in this case I'm running no vdroop so....that's a no-go for me. :)

I already reported about my findings re: the uncore being "fake unlocked". I have it set as 16x, in cpu-z the frequency comes up as 3200mhz which is correct, so I don't think it's fake in my case. Maybe it's cpu dependent whether the setting actually works or not, but AFAIK it is working correctly for me.

edit here's 2 apps showing correct uncore frequency :confused: IDK maybe it works for some cpus and not others. works fine for me.
ss.png
 
Last edited:
How to find Bios Version??

This is what my splash screen says:
Bios Date: 05/05/10 09:15:13 Ver: 08.00.16

Just seeing if I need to update Bios.
If so, best version?
 
hfuser2009 said:
so the labeling on SR-2 board is incorrect?? for the CPU0 & CPU1.
Click to expand...

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.
 
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 :)
 
You must log in or register to reply here.
Back
Top