SR-2 Optimization Thread

M

musky

[H]ard|DCer of the Year 2012
Joined
Dec 14, 2009
Messages
3,154
This thread is for SR-2 owners to compare bios settings and hopefully help each other get the most out of our systems. Here's mine:

Basic system
SR-2, custom A47 revision bios to support processors
2 x E5530 ES Xeons, B0 stepping
2 x Cooler Master Hyper N520 HSFs, 60+ CFM fans x 4
6 x 2Gb OCZ Gold PC1600, rated 8-8-8-20
Corsair TX750 PSU
80Gb SATA HDD

Currently @ 3744MHz (18 x 208), memory @ 1664MHz triple channel
73K - 75K ppd depending on unit, 43K or so on P2684s
http://www.avid-edge.com/pc/sr2.png

Bios settings

Frquency/Voltage Control - main screen
CPU Frequency - 208
PCIE Frequency - 100
CPU Multiplier - 18
QPI Frequency - Auto (4.800GT)
Memory Frequency - DDR-1066
MCH Strap - 1867

CPU Configuration screen
Max CPUID Limit - Disabled
Coherency Support - Disabled
A20M - Disabled
Speedstep - Disabled
C-State - Disabled
Everything else is Enabled

Memory screen
tCL - 8
tRCD - 8
tRP - 8
tRAS - 20
tWR - Auto (14)
tRFC - Auto (98)
tWTR - Auto (7)
tRRD - Auto (7)
tRTP - Auto (7)
tFAW - Auto (28)
Command Rate - 2T
Everythign else is Auto

Voltage
VDroop (both) - Without VDroop
Vcore (all) - Auto
Vtt (all) - 1.3750
Memory CPU0 - 1.63
Memory CPU1 - 1.64
IOH - 1.400
Everything else is Auto

Signal Tweaks screen
Everything is Auto

Notes:
1. CPU voltage is 1.293 for CPU0 and 1.281 for CPU1 with Auto settings. Manually setting these did not make a difference. This is probably because with only an 18x multiplier, the CPU is probably not my limiting factor.
2. Memory is at 1.669 for both CPUs with the current setting. That is reason for the difference in the two CPU channels.
3. Vtt is around 1.35 with the setting at 1.3750
4. IOH is around 1.36 with the setting at 1.400
5. Memory timing were stopping me initially. I had them relaxed at 9-9-9-24, which caused two modules to fail at anything over 175MHz bclk. Leaving everything on Auto kept triple channel up to 190 bclk. Changing to 2T command rate got me to over 200 bclk. Auto would be fine at 208 bclk. I manually set them for some reason and never set them back to Auto.
6. Watch your actual voltages. Mine are quite a bit off from what the settings says. Memory was stopping me mainly. 1.65 on the setting ended up with 1.69V one one CPU and 1.68 on the other. These modules do not run well with that much voltage.
7. Mine pulls 590W, which is quite a bit more than others.
 
Thanks for this Musky, I am messing around with my ram based on your experience and have removed half of them, to see if I can do better with only 12GB, and if it will be worth it.
 
Before I knock myself out trying to work out why I cannot get the 1066 divider to work like you can - is your ram like mine:

6GB (3x2048MB) T/C Kit PN - OCZ3G1600LV6GK

I think that newegg link is wrong...

I am running 6 sticks in all the red slots...
 
That is the same memory I am running - 6 sticks in the red slots. The link that pulls in is an advertising thing. I didn't add it, and it is wrong.
 
I have used Musky's as template. My main issue was getting memory working.

EDIT 6th October - see second SR2 with x5660 here

EDIT 22 December 2010 see final mature settings in post 142 http://hardforum.com/showpost.php?p=1036223866&postcount=142

EDIT 8thSept: MY LONG NATIONAL NIGHTMARE of RAM IS OVER. New Kingston RAM, problem solved!

After rotating 4 kits of OCZ OCZ3G1600LV6GK PC3-12800 Gold 3x2GB - I have come to the conclusion that it is simply incompatible at anything other than the 800DDR divider (2:6) despite bing on the compatible list. It passes 12 hours of single threaded Memtest86+, but using multiple instances of MemtestPro, is shown to be unstable in less than an hour at 2:8 - including at stock speed. Rock solid at DDR 800, but RMA... and the new kingston stuff works first time PHEW! This was NOT a problem because of having all slots full - for the last few weeks I have been running 12GB, just in the red slots.
Lessons
RAM needs a variety of tests to be sure - a single threaded test is not a representative test of how RAM is stressed in an SR2.
The SR2 BIOS is still shitty, and do not depend on saving settings - photograph your BIOS screens if you want to be sure during testing.
There is still something funny going on, with not all RAM detected on startup every so often - a second reboot and all is well, passes Memtest/Linpack.



Basic system - dual hexacore XEON X5650 2.66GHz, = 12 cores, 24 threads.
EVGA - SR-2 - A47 bios
Kingston KHX1600C7D3K3/6GX - 12 x 2GB = 24GB - new!
Win 7 OEM 64
Intel X-25M 80G
Thermaltake Toughpower 1200W Modular - LOUD - replaced with...
Enermax Revolution85+ 1250W - MUCH quieter
Lian Li PC-P80B - rear fan reversed to intake - draws in cool air to PSU - takes PSU from loud to silent.
GTX460 1GB
Noctua NH-D14 CPU Cooler x 2 with stock fans. Awesome cooling. Very hard to get RAM in and out- no clearance for heatspreaders
2x 2 tb hdd

Currently 4204MHz (20+2 turbo x 191),

Bios settings for 4.0GHz overclock, very cool

Frequency/Voltage Control - main screen
CPU Frequency - 182
PCIE Frequency - 100
CPU Multiplier - 20
QPI Frequency - 4.800GT
Memory Frequency - DDR-1066
MCH Strap - 1067 (auto first - it affects SPD timings - see below)
CPU Uncore = 16x (twice your RAM divider = 2:8)

CPU Configuration screen
Max CPUID Limit - Disabled
Coherency Support - Disabled
A20M - Disabled
Speedstep - Enabled
Turbo - Enabled
C-State - Disabled
ACPI T state - disabled
Everything else is Enabled

Memory screen
see http://hardforum.com/showpost.php?p=1036223866&postcount=142

Voltage
VDroop (both) - Without VDroop
Vcore (all) - 1.31250v
Vtt (all) - 1.350v
Memory CPU0 - 1.65
Memory CPU1 - 1.65
IOH - 1.350

Signal Tweaks screen
Voltage Tweaks - IOH QPI 0 = -72
Voltage Tweaks - IOH QPI 1 = -16

These apparently have an impact, they seem to help more at 190+ bclck. Based of Shaminos BIOS template

Idle = 270watts - (no c state)*
Folding = 528Watts*
Folding + GPU folding = 620watts*

* notes = power includes 2 HDD running = 10w for both, also, these figures are with 12 sticks of RAM in. = about 20watts more compared to 6.

With only 6 sticks of RAM, one SSD, one GTX-460 idle and hard drives off it is 495 to 500watts folding @4Ghz


4.2GHz stable using
Vcore (all) - 1.39375v
Vtt (all) - 1.375v
IOH - 1.4
Voltage Tweaks - IOH QPI 0 = -85

Adds 30w to idle, adds 100+ watts to load - temps folding are 70c at 18c ambient

I did quick 2 frames test of Project 2685: 13mins 07sec = 130,000 PPD

Currently working on another overclock - dropping 1 BCLCK to 190 seems to work on much less voltage - more on this once I certify it.

Once I am up and really folding again I will post more reliable averages.

Look at what the MCH strap does to RAM auto settings. It was news to me Test with 12GB - 6 sticks in red slots. 1T. Setting RAM timings to auto, and changing the MCH strap while on the DDR1066 divider gets the results:
MCH=1867 would not boot
MCH=1600 auto = 11-11-11-20 timings
MCH=1333 auto = 9-9-9-20
MCH=1067 auto = 7-7-7-20

Seeing how the timings try to compensate gives you a good idea of where to start - I have set MCH to DDR ratio , and manually set 7-8-7-20-1T auto the rest.
 
Last edited:
With 12 gb you could get down to 1t as well.
Have you tried 1.65v?
Posted via [H] Mobile Device
 
jebo_4jc said:
With 12 gb you could get down to 1t as well.
Have you tried 1.65v?
Click to expand...

Sorry - tired brain - yes I am on 1T all today, first at 12GB and now back to 24GB. Yay. I think I did weird things to my fingers getting that bloody RAM back in without removing the coolers.
 
Glad you figured out the RAM issues, may i be worthy. Maybe I'll have time tonight to see if I can duplicate your OC results.
 
@MIBW

Turbo will not work with C-State disabled. Since you have a 4ghz OC i would suggest turning Speedstep and Turbo off. They are only going to hurt your stability.
 
I'm using the same CPU configuration settings as MIBW, and they're the same as shown in the SR2 OC BIOS Template. Turbo does work with C-State disabled (only Speedstep needs to be enabled), and Turbo needs to be enabled to hit 4GHz at 182 BCLK. If you disable Speedstep/Turbo, then you have to crank up BCLK to hit the same CPU speed, which means cranking up voltages.
 
I also have the same ocz memory, thought in a X58 board i have had much better luck running tras at 24 instead of 20 even at 1600 speeds. specially at 1680. not sure if this will apply to your SR boards but maybe it will help.
 
SazanEyes said:
Glad you figured out the RAM issues, may i be worthy. Maybe I'll have time tonight to see if I can duplicate your OC results.
Click to expand...

Let us know how you go. My 4.0Ghz I would call easy. Above that it gets challenging very quickly.

nitrobass24 said:
@MIBW

Turbo will not work with C-State disabled. Since you have a 4ghz OC i would suggest turning Speedstep and Turbo off. They are only going to hurt your stability.
Click to expand...

Sorry, but that is utterly wrong. I'm with Sazaneyes on this.

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.

If I turned off turbo, I would have to raise my baseclock overclock from 182 bclck up to 200 to get 4Ghz. I have not been able to get stable at 200 (haven't really tried much) But looking the the difference between my 182 = 4.0GHz and my 192 = 4.2Ghz, a whole lot of voltages have to go from amber to red to make that happen. This board is easy up to about 180, but very tricky above that. On 4.0GHz, I can get by with 1.35v vtt and IOH, rather than 1.375 and 1.4. Once you nudge 1.4 you are getting a bit hot for long term useage, particularly on a machine that is my primary workstation and aircooled.

Far from being bad for stability, Turbo is a godsend - it means you don't have to flog your whole system to get better CPU speeds.

I think of turbo as the difference between a regular basketball player and Michael Jordan. ;) Turbo is when the little guy dunks, with his hand just stretching high enough to still reach the basket. His body does not fly as high as "Air" Jordan, but gets the job done. I don't want to volt my whole system to MJ levels in the stratosphere to get the same points. :cool:

I imagine there is a tiny amount of instability brought in my all the switching, but Linpack does a great job of testing this if you set threads to Auto. And this instability is more than made up for by a more stable system running at a lower baseclock. But the point is, if it introduced instability my tests would have caught it.

I think the issue of whether to use turbo or not should depend on where your memory and system happens to be at - if you are halfway between dividers etc.

I think Turbo gets a bad rap because of earlier implementations, and because people just turn it on without turning down their baseclock, see it crash and spread the word.

lehmann said:
I also have the same ocz memory, thought in a X58 board i have had much better luck running tras at 24 instead of 20 even at 1600 speeds. specially at 1680. not sure if this will apply to your SR boards but maybe it will help.
Click to expand...

Thanks for the tip, but so far I am doing ok at 20. Cross fingers.
 
Im with you on turbo MIBW
Posted via [H] Mobile Device
 
Last edited:
Anyone happen to be running any of the Westmere E5620's? 32nm, 2.4GHz, 12MB L3 but they have a kinda low multiplier @ 18 (19 with turbo). However, for $389 it might be a good "budget" CPU. Thoughts?
 
Tobit said:
Anyone happen to be running any of the Westmere E5620's? 32nm, 2.4GHz, 12MB L3 but they have a kinda low multiplier @ 18 (19 with turbo). However, for $389 it might be a good "budget" CPU. Thoughts?
Click to expand...

I have one of these that I used long enough to flash the bios to support my E5530s. I think that if I were starting over with a "budget" build on one of these, I'd go with E5620s. That 19x multiplier would be a big plus, and $380 (on eBay) is about as reasonable as you are going to get for any new Xeon chip. I am trying to talk NobleX13 into running this chip in his X58 Gigabyte board to see what it will do as far as an overclock. I'd think you could approach 3.8GHz pretty easily with a pair of these.
 
Excellent thread. I'll definitely come back to this one when I buy my SR-2.
 
musky said:
I'd think you could approach 3.8GHz pretty easily with a pair of these.
Click to expand...
Was lurking around on EVGA's forum and there's a guy running a pair of these at 211 x 19 for 4GHz.
 
Tobit said:
Was lurking around on EVGA's forum and there's a guy running a pair of these at 211 x 19 for 4GHz.
Click to expand...

Getting the board to run 211 seems to be the challenge. I don't doubt at all that the chips will run 4GHz.
 
Tobit said:
Was lurking around on EVGA's forum and there's a guy running a pair of these at 211 x 19 for 4GHz.
Click to expand...

musky said:
Getting the board to run 211 seems to be the challenge. I don't doubt at all that the chips will run 4GHz.
Click to expand...

i'm sure it's possible, but i'm not sure you would want to feed all the volts to the mobo that it would require
 
I'll put that E5620 into my gigabyte board and OC the hell out of it later, Musky!
 
I just spent a couple hours trying to hit 4GHz and couldn't do it. Looks like my board hits a wall a little bit sooner than may I be worthy. I probably need to update the BIOS and try again. Here are my current settings.

edit: Updated to A47 BIOS. No change in overclocking. Maybe LinX is not the best test tool. I'll try again later.

Basic system
EVGA SR-2 - A47 BIOS
2 x Xeon X5650 6-core
2 x Cogage TRUE Spirit
Corsair Dominator CMD12GX3M6A1600C8 PC3-12800 6x2GB
EVGA GeForce 8800GS
Enermax Revolution85+ 1050W
WD Velociraptor 300GB
LG Blu-ray burner
no case
Windows Server 2008 R2

Currently 3850MHz (20+2 turbo x 175),

Bios settings

Frequency/Voltage Control - main screen
CPU Frequency - 175
PCIE Frequency - 100
CPU Multiplier - 20
QPI Frequency - 4.800GT
Memory Frequency - DDR-1066
MCH Strap - Auto
CPU Uncore = Auto

CPU Configuration screen
Max CPUID Limit - Disabled
Coherency Support - Disabled
A20M - Disabled
Speedstep - Enabled
Turbo - Enabled
C-State - Disabled
ACPI T State - Disabled
Everything else is Enabled

Memory screen
8-8-8-24 - auto the rest

Voltage
VDroop (both) - Without VDroop
Vcore (all) - 1.300 V
Vtt (all) - 1.350 V
Memory CPU0 - 1.64 V
Memory CPU1 - 1.63 V
IOH - 1.350 V

Signal Tweaks screen
IOH QPI 0 = Auto
IOH QPI 1 = Auto

Idle = 220 Watts
Folding = 445 Watts
Temps around 60 C when folding
 
Last edited:
The signal tweaks didn't help? That is odd. I was thinking that may be needed for the hex cores. I think you are going to need more IOH voltage, and possibly more Vtt. Did you try these two on Auto?
 
Yeah, BIOS update sounds good. Drop the Uncore first I think, that is likely the culprit.

I would also drop your ram to 800ddr temporarily to eliminate that variable. I am still having massive issues getting 1066DDR stable even with 12GB, and am back to 800DDR. Again.

When I am off deadline I will be chasing this up, because I can boot to a usb stick running memtest86+ and pass 12 hours, but use MemtestPro which I run within Windows, running 10 or so threads testing 1GB each, and fail in 10 minutes.

So clearly memtest86+ is single threaded, and does not detect memory problems that occur when you use your system in more real world conditions. (many threads, 90% RAM used)

Tried upping vtt and IOH. More next week when I can get time.
 
I tried Vcore at around 1.325V, Vtt at 1.375V and IOH at 1.4V, and I tried manually setting QPI 0 to -80 and QPI1 to -16. I test using LinX at 10GB, which seems like a good worst-case scenario. Typically what would happen is the first pass would finish and tell me it failed, although occasionally the process would crash or I'd get a BSOD.

After I update the BIOS, I think the next step will be determining maximum BCLK before I have to crank up the voltage. It seems to be between 175 and 182, but I need to nail it down.
 
Ok, I used your settings for the most part, but used turbo boost, left blck at 200.

This has me running at 19x200 for 3.8GHz and 1600 memory.

The memory alone should help, not to mention the extra 200mhz. I hope I'm stable....
 
Kendrak said:
Ok, I used your settings for the most part, but used turbo boost, left blck at 200.

This has me running at 19x200 for 3.8GHz and 1600 memory.

The memory alone should help, not to mention the extra 200mhz. I hope I'm stable....
Click to expand...

I hope so too. I'll be running my CPUs 50 MHz slower than your, but my memory will be 64 MHz faster than yours. We should be almost identical ppd.
 
musky said:
I hope so too. I'll be running my CPUs 50 MHz slower than your, but my memory will be 64 MHz faster than yours. We should be almost identical ppd.
Click to expand...

Ohh forgot, left the memory voltage at auto as well.

Just finished the first full frame since I restarted. I went from 28:44 on a 2684 to 26:09.

Crappy WU, that sure is a speed bump.
 
Mine is doing 26:07 on a P2684, so they are almost identical. You'll see right at 75K ppd on a good work unit, with frame times in the low 19's.
 
musky said:
Mine is doing 26:07 on a P2684, so they are almost identical. You'll see right at 75K ppd on a good work unit, with frame times in the low 19's.
Click to expand...

I'm going to let this go a WU or 2 and see if I can push it just a bit more since I have an extra notch on my multi.
 
That's rite.... the B0 has trubo too... sorry.

Hard to keep all these different flavors strait.
 
B0's don't have turbo, so i am stuck with an 18x multipler. Yours are not ES's, right?
 
OK.... I'm just upped the blck to 210, lets see if I crash.....

210x19=3989 (4GHz)
 
Well it was good for a bit over an hour......

I just had a hard crash. I'm going to try setting the CPU Vcore to 1.35625 and give it a spin.

I have some headroom on temps (70C), so if I can keep it stable, I can run more juice me thinks.

Edit: New temps seem to be holding at 77C..... I hope this works. 4GHz will be quite nice :cool:
 
Last edited:
You must log in or register to reply here.
Back
Top