SR-2 Optimization Thread

@ everyone - that 8 9 8 24 timing is due to my funky Kingston RAM - whose spd recommends 7 8 7 20- most ram would be 7 7 7 or 8 8 8 etc. The reason I bring this up is that I have seen on my RAM 7 8 7 20 and 8 9 8 24 boot up, but 8 8 8 24 (in between) will not. This means that if you could be seeing issues with copying my asymmetric ram timings that the Kingston likes, but not yours.

Yeah @nitro - keep manual mem settings, go to 2T command rate, loosen MCH to 1600, and even try 2:6 / 800 ddr speed as K suggested for a bit just to see if you can push past 200.

Remember I am only at 191 bclck - you guys will be pushing a lot harder, so keeping the MCH at 1066 might be ambitious. We will find out.

@Sazan - I am shocked that you can punch past the bclck wall your x5650s had. That really is surprising, and not in a fun way. I have x5650s and that so easily could have been me. I feel for you mate :(

@Axdrenalin See my post on the first page for wattage figures. Yeah, it is heartstopping the first time you see idle wattages of 300+ watts, but then you realise it is less that the idle of the combined 5 older machines it is replacing and you accept it! And in terms of folding, it is still better than any other option for watts per ppd. My SR2 is really only 5 times faster than my Q6600, but is deemed up to 23 times better at ppd due to bigadv.
 
Last edited:
If it isn't listed, it is most likely on Auto

SR-2 #1
198 x 18
Vcore 1.328 actual
Vtt 1.35 (1.339 actual)
IOH 1.4 (1.368 actual)
Memory 1.669 actual
Memory timings - 8-8-8-20-88-6-12-6-6-24-2T (newer bios, so the sequence doesn't match the old bios)
Memory speed DDR-1066
MCH - Auto
Memory brand - OCZ Gold PC1600 8-8-8-20

It is stable at 200 bclk and probably over with 1.399V Vcore actual. It is also no faster than it is at 198, or even 195. LinX stable at 10240Mb memory. P2692 times - 14:30/frame.


SR-2 #2
187 x 18
Vcore 1.336 actual
Vtt 1.35 (1.339 actual)
IOH 1.4 (1.368 actual)
Memory 1.669 actual
Memory timings 9-9-9-24-88-6-12-6-6-24-2T
Memory speed DDR-1066
MCH - Doesn't exist in this bios

This one is very tough to get to recognize all of the memory on boot, and fails LinX 10240Mb after 2 passes. I was not able to get 190 bclk to read all of the memery. This one obviously has a bad memory stick, but I am not going to mess with it any more tonight. It is LinX 10240Mb stable at this speed.

Tips
- Get a reference point for relative speed. Mine was the time it takes to complete a pass on LinX at 10240Mb memory. High clock speed does not necessarily mean a faster machine
- If a higher clock speed tests out stable but is slower in your reference test, you probably need more Vcore
- If any clock speed is choppy running LinX, you need more Vcore.
- If LinX stops with error, you have a memory issue. Either relax timings, add voltage, add Vtt/IOH
- If the system crashes, it could be either CPU or memory.
- Turn off the boot splash sceen
- Disable ESATA and the USB3 stuff if you don't need it. You'll boot a heck of a lot faster

Now as I was typing this, the F@H client on SR-2 #2 crashed. Back to LinX...
 
What does MCH strap do? All i can see is that is sets my timings, but what is it doing if i am specifying timings manually?

Secondly, how much does Timings and Mem speed affect F@H performance?
Is my machine at 3.6ghz w/ 1600mem @8-8-8-20 going to faster than 3.78 w/ 1260mhz @11-11-11-29?

Do we gain anything by cranking up the QPI from 4.8 to 5.6?

I havent tested any of these scenarios....im just curious.
 
These numbers below are all at stock frequencies and voltage, but the results are quite interesting.
memtest.jpg
 
Interesting article on the MCH strap. They pretty much concluded that on the X58+ the MCH does really nothing beside change mem timings.
http://www.techreaction.net/2009/06/20/evga-x58-sli-classified-mch-strap-performance/

I find this odd though because even if I manually set timings, and then lower my MCH or set it to auto then i cant boot. Why would MCH strap be causing me not to POST if i have manually set the timings?
 
Found a weird Tubo bug on my L5640's
It will randomly jump up to 21x multi.....now how can i keep it there?
weirdturbo.png
 
never seen that one...mine jumps to 19X occasionally when testing, but never 21x.
 
nitrobass24 said:
What does MCH strap do? All i can see is that is sets my timings, but what is it doing if i am specifying timings manually?

Secondly, how much does Timings and Mem speed affect F@H performance?
Is my machine at 3.6ghz w/ 1600mem @8-8-8-20 going to faster than 3.78 w/ 1260mhz @11-11-11-29?

Do we gain anything by cranking up the QPI from 4.8 to 5.6?

I havent tested any of these scenarios....im just curious.
Click to expand...

MCH - see below, as for my testings with timings, it seems that the timings you really have to worry about (in order of priority) are your divider speed, then 1T vs 2T timing, and then the rest of the timings. Going from 7 8 7 20 to 8 9 8 24 did not hurt me much, but 1T => 2T will cost you about 3% frame times and running 800MHz 2:6 you may as well go home.

Never cranked the QPI up, I presume it would not boot, fry your memory controller, and cause an explosion like the death star going up (special edition, with obligatory shockwave ring). Why don't you try for us :D

nitrobass24 said:
Interesting article on the MCH strap. They pretty much concluded that on the X58+ the MCH does really nothing beside change mem timings.
http://www.techreaction.net/2009/06/20/evga-x58-sli-classified-mch-strap-performance/

I find this odd though because even if I manually set timings, and then lower my MCH or set it to auto then i cant boot. Why would MCH strap be causing me not to POST if i have manually set the timings?
Click to expand...

Even with mem timings all manually set It IS doing something, and it DOES effect performance (slightly), and it certainly has a big impact on stability and being able to boot:

MCH 1867 - won't boot
MCH 1600 - 13 mins 24sec
MCH 1333- 13 mins 23sec
MCH 1066 - 13 mins 01sec
MCH 800 - 13 mins 13sec

Now these are subtle differences, that you need to run benchmarks of longer than a few seconds to see. And bigadv is super sensitive to memory speed.


nitrobass24 said:
Found a weird Tubo bug on my L5640's
It will randomly jump up to 21x multi.....now how can i keep it there?
Click to expand...

I think that is just CPUz - I get that too, usually underreporting. Try graphing with argus monitor. http://www.argusmonitor.com/en/

See the turbo graph down the bottom here:

original.jpg


And a screenie from their site:
turboboost1_eng.png
 
I noticed after flashing to A47 that Uncore can be picked now. ;)

I should have some prelim SR-2 settings tomorrow after work...have lots of work to do (with the SR-2 of course). Man I wish I could take a sick day. :D
 
Well those screens i posted earlier were the best i can get at this point.
I have tried every combo i could think of.

2:10 mem divider just wont work because my memory cant handle 2000+mhz.
And no matter how much vcore or VTT i give it (tried up to 1.5 on both) or how loose i set the MCH and timings i cant even POST at 201blck.
 
nitrobass24 said:
Well those screens i posted earlier were the best i can get at this point.
I have tried every combo i could think of.

2:10 mem divider just wont work because my memory cant handle 2000+mhz.
And no matter how much vcore or VTT i give it (tried up to 1.5 on both) or how loose i set the MCH and timings i cant even POST at 201blck.
Click to expand...

Did you try 2:6 or 2T?

How low did you try QPI signal tweak? I think you need lower than -85 for over 200 bclck - your screeny showed only -72
 
On 2:6 if i run mem timings looser than 8-8-8-20 i cant post even at stock everything else.
I tried 11-11-11-29-2t on 2:8 @ all MCH values, cant post over w/ over 200blck. I never took the QPI signal lower than 82. I can try that tomorrow, but im done for tonight gonna watch some TV and enjoy my 14min frames on bigadv. :D

Maybe the whiskey is getting to me. Well try again tomorrow.
 
nitrobass24 said:
On 2:6 if i run mem timings looser than 8-8-8-20 i cant post even at stock everything else.
I tried 11-11-11-29-2t on 2:8 @ all MCH values, cant post over w/ over 200blck. I never took the QPI signal lower than 82. I can try that tomorrow, but im done for tonight gonna watch some TV and enjoy my 14min frames on bigadv. :D

Maybe the whiskey is getting to me. Well try again tomorrow.
Click to expand...

Hey that is good news, I Shaminos template calls for -85 at 200 bclck. (as low as you can go and reliably boot)

But take a break. A tired overclocker is a bad overclocker I say.;)
 
may i be worthy said:
Turbo is not dependent on C-State, I run without C-State, but have tried it. It has great savings at idle, but I do not like the lag/freezes as it kicks back in. And besides, I have discovered folding... what idle?!

My 4.0GHz overclock with Turbo and speedstate has passed over 48 hours of Prime95, Linpack/IntelBurn and memtests. Solid as a rock.
Click to expand...

QFT
With the SR-2 you have to turn on speed state to be able to enable or disable turbo.
C State settings do not affect either. (this is A47 bios at least)

Highest stable bclk was 203 with 1.3 Vtt, 1.4 IOH and signal tweaks as recommended by Shamino at the SR-2/EVGA forums.
Started working on raising multi before I went to bed late this morning; currently running 17 (18 w/turbo) x 160=2880 on stock volts (1.1125 Vcc). LinX stable. Will have more info later. ;)
 
Last edited:
Thanks to Musky for starting this wonderful thread and providing a lot of valuable insights.

I got her up and running with the following hardwares:

EVGA SR-2
2 x Xeon x5650
1 Prolimatech MegaShadow, 1 Prolimatech Megahalems B each with CM red LED 120mm 2000RPM (90CFM) fan on it
12GB G.Skill F3-12800CL9D-4GBRL RAM (3 kits) with one DIMM from each kit installed as the third DIMM in each bank o' RAM
Lian Li PC-P80 case
Seagate 300GB 3.5" Sata Drive
Samsung DVD-R/W

So far I've found two stable settings for my SR-2 + dual Xeon X5650 combination.

Setting 1, 3.917Ghz, 178 BCLK x 22 multiplier (with Turbo and SpeedStep enabled) ORIGINAL shipped BIOS (4/4/2010 date, I think A16)

VDroop: WITHOUT VDroop
VCore: Auto
VTT: 1.375
DRAM Voltage: 1.60V > 1.61 in HWMon
DRAM: 2:8 / 1066mhz > 1424Mhz, 9-9-9-24-86-2T
IOH: 1.375
QPI: 4.8GT/s
Uncore: ?
MCH Strap: 1600mhz

Signal Tweaks: QPI0 -70, QPI1 -16

All other set for auto except ACPI T > Disabled, Speedstep and Turbo enabled

This was stable for 4 hours of folding, P2684 @ 18:21 TPF for around 78K PPD (IIRC)

Newer Setting:

3.8Ghz 190BCLK x 20 multiplier, Turbo and SpeedStep disabled, newer A47 BIOS

VDroop: WITHOUT Vdroop
VCore: Auto
VTT: 1.375
DRAM Voltage: 1.60V
DRAM: 2:8 / 1066mhz > 1560Mhz, 9-9-9-24-86-2T
IOH: 1.375
QPI: 4.8GT/s
Uncore: x12 (1600mhz)
MCH Strap: 1600mhz

Stable for 3.5 hours of Linx using 5120MB of Memory.

Folding the P2684 is showing 18:35 TPF at this speed after one full frame for around 75K PPD.

One thing I've found strange is that with VCore at Auto I've had greater success with CPU speeds. Trying the above settings with 1.3250V Vcore yielded me errors in LinX or BSODs during stress testing, but with the VCore at Auto it's fully stable, AND the voltage is only showing up as 1.31Volts in HW monitor and 1.294V for each CPU in CPU-Z. With the VCore set

Alternatively I've found that not enough VTT for each CPU will manifest itself in a problem with Win7 only showing 6,8, or 10GB of usable RAM at higher BCLKs even if 12280MB shows up in BIOS. This is typical of Core I7s too, so the Oc'ing experience doesn't seem to be very different, except it's a bit tougher getting voltages exact for two CPUs versus one.

For example, at 190BCLK with 1.325 or 1.35 Volts VTT Win7 x64 showed 8, and 10GB respectively, without any other voltage or speed adjustments.

Additionally may i be worthy you might find it interesting that in some situations, even if all 12GB was showing up in BIOS and Windows boot, and Linx showed 10.7GB available at start, LinX failed with a "not enough memory" error at some point after starting stress testing (usually when trying to post the first "run), meaning that due to marginal VTT starvation it probably somehow failed to run all the RAM through the IMCs (integrated memory controller) on the chips. So as soon as the chip is stressed, the IMC somehow errors out a bank of RAM, or starts failing, so LinX or IntelBurnTest are definitely your best friends here.

Additionally, I'm noticing that with or without VDroop doesn't seem to make much of a difference showing up in HWMonitor or CPU-Z. Does any one else find this?

With other EVGA boards I've seen as much as a .035 volt difference (x58 SLI LE and P55 FTW), where "without VDroop" had volts go up by that much on load. Weird.

Also, using the above example, I think under load, there may be a tiny bit of VTT droop when the CPUs go full blast stress testing. Just a theory, but I've seen this on some x58 boards in the past, like Asus P6T Deluxe. This may explain the stress-test memory drop-off. So it might be that the VTT is marginal, but when it droops a bit, it goes into the "red zone" and the IMC doesn't have the juice.

I'm going to try going up to 3.9Ghz (195BCLK x 20) tonight and see if stability is good. Temps are very low on each CPU at 48 - 60 celsius on the cores. Need to re-apply TIM methinks. Arctic Cooling MX-3 sucks IMHO, I prefer MX-2 much more for spread applications using credit cards.

If I get speed up any further (ie. 3.9Ghz or 4.0Ghz) I'll update. I'm happy with 3.8 or 3.9 though right now. This thing rips through work units like a chainsaw.

Musky's list in the above post under "tips" explains most of this.

I'd only add that if you raise speed/BCLK without changing RAM or CPU dividers/multis, and you see lower performance in either folding or LinX/IntelBurnTest, your VCore is too low.

For RAM problems (ie. not all showing up) on known good hardware (RAM/mobo) VTT is your friend :)
 
nitrobass24 said:
Well those screens i posted earlier were the best i can get at this point.
I have tried every combo i could think of.

2:10 mem divider just wont work because my memory cant handle 2000+mhz.
And no matter how much vcore or VTT i give it (tried up to 1.5 on both) or how loose i set the MCH and timings i cant even POST at 201blck.
Click to expand...

There was a specific item in one of the reviews I read on the SR-2 where they said certain multipliers don't work on certain chips. They contrasted an x5680 and x5650 where the x5680 supported the 2:10 multi and the x5650 didn't.

I don't know if this was because of the original shipping BIOS or a later ver.

It was Bit-tech:

http://www.bit-tech.net/hardware/motherboards/2010/08/17/evga-classified-sr-2-review/4

Second last paragraph here:

"As not all Xeons support the same memory straps the Xeon 5650s would not POST when we set the RAM divider to 1,333MHz (10x) in the BIOS, only at 1,066MHz (8x). Because of this, when we overclocked the QPI to 190MHz the memory was limited to running at 1.52GHz. In contrast, the higher spec Xeon X5680s do support the 1,333MHz (10x) memory strap, so were able to run the memory at 1.9GHz."

Hope that helps.
 
I can post using a 10x divider just not with a decent OC.

I posted a chart a while back with my results.
 
Good info 10e. It looks like your X5650s are similar to mine, in that they are not great overclockers. However, the benefit as you found is that you don't have to run with a high Vcore and therefore have low temps and power usage.
 
nitrobass24 said:
I can post using a 10x divider just not with a decent OC.

I posted a chart a while back with my results.
Click to expand...

Yes, I saw your very thorough and excellent spread sheet.

I would think going beyond 133mhz BCLK with a 10x divider will not work if you are still on those same 1333mhz DDR3 G.Skill kits.

I don't like the SR-2 BIOS displaying RAM speeds incorrectly. If RAM speeds correlate to BCLK, a 10x divider and 166mhz BCLK means 1660mhz RAM speeds. But the SR-2 will show you 1333mhz. That's just plain wrong.

Or if I'm wrong, someone pls correct me here.

I would try the 2:6 divider and crank up BCLK as much as possible. With triple channel RAM hex-Core I7 and Xeons should have sufficient RAM bandwidth for all cores, unlike the Magny-Cours Opterons as they are really a 2+2 quad channel setup and really require 4 DIMMs per bank for full RAM speed.

I'm also going to try and dig up what info I can find about NUMA vs. UMA memory access on these boards. NUMA allows CPUs to access each others' RAM via QPI (or HT) but can slow things down, while UMA means each CPU only uses its' own RAM bank.

I'll see what I can find. I would theorize these are UMA access boards based on how frickin' fast they fold overclocked.

EDIT: That's probably the case that UMA is used, as the IMCs are separate from Qpi bus and I don't think QPI can carry RAM data "cross CPU" like HT on AMD can.

SazanEyes said:
Good info 10e. It looks like your X5650s are similar to mine, in that they are not great overclockers. However, the benefit as you found is that you don't have to run with a high Vcore and therefore have low temps and power usage.
Click to expand...

I'm going to try and see what I can get. Folding at 3.917Ghz (178x22) is actually a tiny bit faster than 3.8Ghz (190x20) even though the RAM is faster with the 3.8Ghz. It's only a 20 second TPF difference, but I am going to see if I can go 182x22 for an even 4.0Ghz.

With these boards, two Cpus can insert a lot more variables into the equation.
 
Yea i can post at 190blck with a 2:6 mem divider but i have to leave my timings really tight and i cant go higher than that without getting an F4 post code.

So i settled on 2:8 which gives me the most flexibility, i just seem to be memory limited at this point :(
Someone want to trade me for some DDR3-2000?
 
Any of you guys getting -bigadv units and processing them successfully on the SR-2? I can process A3 units all day long without issue, but as soon as I get a -bigadv unit F@H takes a dump and crashes. The log files points to a C0000005 error, which points to possible memory issues. This is the same RAM I used processing -bigadv units in my GB P55-UD3R with an i7-860 all day long, so I'm thinking it has to do specifically with the memory timings / settings and not the memory chips themselves.

CoreStatus = C0000005 (-1073741819)
Client-core communications error: ERROR 0xc0000005
Deleting current work unit & continuing...
This is a known Windows memory error, while running the v5.x GUI client with the GUI open while finishing and uploading a work unit. Workarounds include updating the video driver (doesn't always help), keeping the GUI closed near the end of a work unit, or switching to the console client and using a 3rd party utility to see the pretty pictures and monitor the client's progress.

It can also be caused by faulty memory or a bad memory controller, so you should consider both possibilities.
Click to expand...

My specific RAM that I'm using is Corsair 1600 9-9-9-24 1.65v DDR3, two matched sets of 2x2 sticks for a total of 8 GB (although I'm in Server 2008 32 bit, so it can only use 4 GB). Any suggestions on timings that I can implement to get this little issue resolved? I could also take that RAM out and slip in a 6 GB kit that I used with on of my x58A-UD3R boards, but I'd have to use 2x in with one CPU and a single in the other.

Thanks - Ax
 
I haven't had that specific issue, but did you leave your timings on "auto" in the BIOS or did you set them manually?

Several of us had better results once we manually set our memory timings.
 
jebo_4jc said:
I haven't had that specific issue, but did you leave your timings on "auto" in the BIOS or did you set them manually?

Several of us had better results once we manually set our memory timings.
Click to expand...

Yep, noticed that a few days ago and have them set manually. basically set them to what they were detected as (other than the 9-9-9-24) and I made sure to loosen it up using 2T instead of 1T.
 
Well, I've tried three different sets of RAM on this, and get the same results every time I hit a 2685. Throws -bigadv for a loop every time....I've reset the BIOS several times, and will have to play with this more when I've got time....

Wondering if I shouldn't pull the CPU's off seperately and try them on my Gigabyte UD3R board, and see if that doesn't make a difference....maybe it's something to do with the on-chip memory controller. Still doesn't amke sense that I can run A3 units all day long and never miss a beat. :confused:
 
Axdrenalin said:
Well, I've tried three different sets of RAM on this, and get the same results every time I hit a 2685. Throws -bigadv for a loop every time....I've reset the BIOS several times, and will have to play with this more when I've got time....

Wondering if I shouldn't pull the CPU's off seperately and try them on my Gigabyte UD3R board, and see if that doesn't make a difference....maybe it's something to do with the on-chip memory controller. Still doesn't amke sense that I can run A3 units all day long and never miss a beat. :confused:
Click to expand...

Have you stress tested it into the memory a little further? The main difference between bigadv and regular smp is a little more memory use. Regular SMPs may run on the first two chips, where bigadv hits the second two maybe? I don't know. I have the same chips x 2 and the same board x 2, and I am bigadv stable on both. This was after they were stable with LinX at max memory setting (10 Gb in my case.) When i had a couple of bad memory chips in one, the FAH executable kept crashing. I didn't lose units, but it usually took a few tries to get the client to start without crashing.
 
Axdrenalin said:
Well, I've tried three different sets of RAM on this, and get the same results every time I hit a 2685. Throws -bigadv for a loop every time....I've reset the BIOS several times, and will have to play with this more when I've got time....

Wondering if I shouldn't pull the CPU's off seperately and try them on my Gigabyte UD3R board, and see if that doesn't make a difference....maybe it's something to do with the on-chip memory controller. Still doesn't amke sense that I can run A3 units all day long and never miss a beat. :confused:
Click to expand...

Hi Ax,

I highly, highly, highly HIGHLY HIGHLY :p recommend using a dozen instances of HCI memtest in windows to test memory stability. Use the /t512 flag to limit each to a small amount of RAM.

Quickest method by far to find problems. Bigadv is so much more of a memory hog, it can't run on a wing and a prayer...
 
My settings (settled on 200; hit 203bclk, didn't really feel like farting around adding more Vcc to get 203 stable vs. 200 which is stable now).

ES L5640's
17x (18x w/turbo) x 200
PCIe 101
QPI 4.800GT (18x, runs at 3600)
DDR-1066 (2:8, 1333 ram runs at 1600)
Uncore 2133 (16x)
Ram 9-9-9-27 1T
Without Vdroop
Vcc boot/eventual 1.20
Vtt boot/eventual 1.30
DIMM 1.65
CPU PLL 1.65
IOH 1.4
IOH PLL 1.8
ICH I/O 1.5
ICH 1.125
IOH QPI 0 Signal -85
IOH QPI 1 Signal -16

LinX 8192mb, 10 passes stable, both bclk and multi testing. (good enough for me) I've done 20x in the past.....I'm happy enough with a clear on 10x, I doubt I'll have any instability.
A few settings were borrowed using either Shamino's SR-2 thread or linuxrouter's Max bclk strategy guide thread over at the EVGA forums. The rest was doing bclk -> ram (which I already knew the maxes) -> multi.
 
musky said:
Have you stress tested it into the memory a little further? The main difference between bigadv and regular smp is a little more memory use. Regular SMPs may run on the first two chips, where bigadv hits the second two maybe? I don't know. I have the same chips x 2 and the same board x 2, and I am bigadv stable on both. This was after they were stable with LinX at max memory setting (10 Gb in my case.) When i had a couple of bad memory chips in one, the FAH executable kept crashing. I didn't lose units, but it usually took a few tries to get the client to start without crashing.
Click to expand...

Keep in mind this is the exact same memory that I was running in my GB boards running -bigadv units with. I'm not sure why it would make a difference on the SR-2. Also, as soon as a -bigadv gets loaded, it crashes right after 0% and never makes it to even the first percent. I was running with a four 2 gig chips initially, then broke them down to just 2x2Gb to try to isolate any memory issues, then when that didn't work I loaded up my other corsair 8-8-8-24 chips with the same results...Strange, I tell ya...

may i be worthy said:
Hi Ax,

I highly, highly, highly HIGHLY HIGHLY :p recommend using a dozen instances of HCI memtest in windows to test memory stability. Use the /t512 flag to limit each to a small amount of RAM.

Quickest method by far to find problems. Bigadv is so much more of a memory hog, it can't run on a wing and a prayer...
Click to expand...

I will do this when I get home from work today. I left it running straight A3's last night again, since I was unable to make any real progress with the -bigadv units.

Thanks guys for all the suggestions and help here. Zero, I'll try to use your settings that you've posted once I run them through the stress test that MIBW mentioned above. I'm not used to everything being quite *this* finicky, especially when I know that this memory has worked so well in other boxen.
 
I find using LinX with maximum memory to work as well as running multiples of MemTest Windows.

It also stresses the CPU more.

It may be that you need to also bump VTT a bit, Axdrenalin, but see how it goes. The 32-bit version of Windows shouldn't be an issue, as the Folding@home binary is 32-bit anyways.

There are occasionally funny Bigadv and A3 units. I just stopped a BigAdv on my highly reliable Core I7 980x at 41% by shutting down the service. When I started it back up it got a new unit and started at 0%, so strange things can happen.

Not happy about losing 13 hours of work, but so far it's been mostly good.
 
Axdrenalin said:
Zero, I'll try to use your settings that you've posted once I run them through the stress test that MIBW mentioned above.
Click to expand...

The biggest factor I've found on i7 based oc'ing is uncore frequency.....until I found out what it was and what I should lower it to, I always left it on auto and I couldn't get higher bclks stable no matter how much voltage I pushed. Once I lowered uncore to 16 (which is technically where it should be with ram running at 8), it's like the clouds went away and I was able to see the sun again. :) The other common factor with lowering uncore is then it takes less juice to get stable at higher bclks and multipliers. I see people running 190 and pushing more Vcore and Vtt than they probably need, only because they're running a higher uncore frequency (or worse, left it on auto).

The one thing I didn't bother with on my UD3R's was IOH and that's because on those I didn't have to. It really seems like though on the SR-2, it likes IOH and needs it to go higher. Could be because of the dual QPI link, it could be because it's a dual processor board...I dunno. :eek: It seems like initially I was having trouble posting higher bclk on the SR-2 but once I raised the IOH, that problem went away. I'm not really familiar with what precisely IOH is (seems more like probably something to do with memory/bclk than just one or the other) but the SR-2 definitely likes IOH juice. :D

Anyway GL with your memory, hopefully it tests out ok.
 
You dont have to run 2x Uncore on Westmere. You should be able to lower it to 1.5 :)
 
My IOH PLL has been dropping a bit below 1.8v recently and been crashing my system with a 101 BSOD. My settings which were FULLY STABLE for 1million points(or about 14 days) with no issues

the IOH PLL dropping below 1.8v only began a few days ago and now my system crashes with a 101 error after about 3-4hours full load, when it never did before. I have shutdown my folding client(after finishing the unit with oneunit) till I can figure this out and get it stable once more.

Changing the voltage on the IOH PLL doesn't seem to work/take any effect.
 
Just wanted to update on this. I used both LinX and Memtest as suggested above, and the memory that I've been using is fine. I did toss in the settings that Zero2dash recommended from his own rig, and must say that things are working a lot smoother now. I stopped testing and put in back in folding mode this morning under these settings, and instantly got an A3 to process. Once that was completed, down came a 2684 -bigadv and it didn't throw any errors at all. I just checked it after arriving home, and it's up past the 10% mark already without issue. Makes me a happy man! :D

So, thanks for all of the advice and suggestions, and a big THANK YOU to Zero2Dash for posting your setup for me to help get my mess straightened out!

Fold [Hard]!!!
 
Glad it worked out for you Ax. I can't see how it works at that low of a Vcore and Vtt, but I'll definately try it next time I tinker with either of mine.
 
musky said:
Glad it worked out for you Ax. I can't see how it works at that low of a Vcore and Vtt, but I'll definately try it next time I tinker with either of mine.
Click to expand...

Not sure either, but its dropped my temps down about 9-10C, as well as taking my wattage use down to about 410 watts (from 513). Seems like my frame times are up a bit from where they were earlier, but I'll have to to cross reference some frames with you guys to see.

Right now on this 2684 I'm seeing about 25:23 minute frame times, was hoping for something in the low 20's / high teens - what are you guys seeing on your SR-2's?
 
You are way high. Here are my two (believer is 18 x 186, cowgirl is 18 x 198, believer had a few crashes due to memory issues):

Project ID: 2684
Core: GRO-A3
Credit: 8955
Frames: 100


Name: believer
Path: \\BELIEVER\fah\
Number of Frames Observed: 300

Min. Time / Frame : 00:20:08 - 68,173.2 PPD
Avg. Time / Frame : 00:26:40 - 44,723.3 PPD
Cur. Time / Frame : 00:20:34 - 58,227.1 PPD
R3F. Time / Frame : 00:20:39 - 57,992.1 PPD
All Time / Frame : 00:20:41 - 57,898.7 PPD
Eff. Time / Frame : 00:27:09 - 44,108.2 PPD



Name: cowgirl
Path: \\COWGIRL\fah\
Number of Frames Observed: 300

Min. Time / Frame : 00:19:19 - 72,541.9 PPD
Avg. Time / Frame : 00:19:55 - 69,288.6 PPD
 
You must log in or register to reply here.
Back
Top