Q9450 OC / Temps / Settings

[Cob]

I quite agree with you - I don't think there is a widely accepted or standardized guideline of the vNB. Being on air - I think I'm being a bit more conservative. I think Cob and some others on Water are up much higher then I'm at.

My NB was 1.58v for 4ghz, and is now down to 1.52v for my 3.9ghz day-to-day clock.

I just have an 80mm fan sitting on top of my graphics card bowing air onto the NB. Not even sure if I need it.

[LUCI5R]

My NB was 1.58v for 4ghz, and is now down to 1.52v for my 3.9ghz day-to-day clock.

I just have an 80mm fan sitting on top of my graphics card bowing air onto the NB. Not even sure if I need it.

Awesome. Cob, did you take a peak at the settings I'll be trying later today? Any pre-advice?

[Cob]

CPU Voltage 1.400v - I think this is a tad too high for 3.7ghz with loadline enabled. Now I know all chips are different, but I only needed 1.38v under load for 3.8ghz in my IP35 Pro.

CPU GTL Voltage Reference (0/2) 0.67 - Start at 0.63x
CPU GTL Voltage Reference (1/3) 0.67 - As above

CPU PLL Voltage 1.54v - I've a feeling you may need a tad higher, but try 1.54v to start with

FSB Termination Voltage 1.34v - Probably not far off where it should be

DRAM Voltage 1.98v - Mines set at 2.02v

NB Voltage 1.28v - Plays a huge part. Might need to go into the mid 1.30's

NB GTL Reference AUTO - Almost as important as the CPU GTL's IMO. Once you get near stable with the correct CPU GTL's this becomes the clincher. But leave on auto til you sort everything else out.

SBridge Voltage 1.20v - Fine

PCIE SATA Voltage 1.60v - Never had to move mine off 1.5v, even for 4ghz

[LUCI5R]

Cob, thanks for your response!
I've noted all your recommendations and I'll be giving them a shot as I try to achieve stability.

CPU Voltage 1.400v - I think this is a tad too high for 3.7ghz with loadline enabled. Now I know all chips are different, but I only needed 1.38v under load for 3.8ghz in my IP35 Pro.

I myself would like to keep it <1.40v but I think out of frustration of not attaining stability @ 460FSB I'm hitting 1.40v. With the other settings being changed, I'll start this back at 1.36v and see where it takes me.

CPU GTL Voltage Reference (0/2) 0.67 - Start at 0.63x
CPU GTL Voltage Reference (1/3) 0.67 - As above

Actually that's what I intended, didn't imply it correctly. Will do!

CPU PLL Voltage 1.54v - I've a feeling you may need a tad higher, but try 1.54v to start with

About how high can I go on this? I'll start at 1.54v but I'll up it if needed.

DRAM Voltage 1.98v - Mines set at 2.02v

This is one question that has lingered with me - when an overclock fails - how do I know if it's because RAM is not getting enough voltage? How Do i know if it's the CPU or the RAM?

NB Voltage 1.28v - Plays a huge part. Might need to go into the mid 1.30's

Interesting - I thought 1.28v was the cutoff on this. Again, how high can I do on this?

PCIE SATA Voltage 1.60v - Never had to move mine off 1.5v, even for 4ghz

I'll give it a shot and see if it helps!

Thanks again!

[rdv]

COB I have a question for you. With all the great advise your giving LUCI5R, I am assuming you are quoting BIOS settings but not real voltages? I ask this as when I set vPLL BIOS to 1.54v the real is 1.632v. Am I correct that you are quoting BIOS settings? If so then LUCI5R may also be over the <=1.6v guideline that we have been discussing.

LUCI5R can and do you record your BIOS Thermal Monitor voltages versus your BIOS settings?

Thanks for any input you can give.

[Cob]

LUCI5R -

PLL - Shouldn't need to go higher than 1.58v.

DRAM - It'e estimated that the board overvolts the memory by 0.08v, so 2.02v should give exactly the recommended 2.1v for the memory. And you shouldn't have any problems with the memory til your FSB is over 515mhz

Nb - I see no reason why 1.28v would be the cut-off. We're running 45nm quads with high FSB's. The northbridge needs a good kick in the arse to do so.

[LUCI5R]

COB I have a question for you. With all the great advise your giving LUCI5R, I am assuming you are quoting BIOS settings but not real voltages? I ask this as when I set vPLL BIOS to 1.54v the real is 1.632v. Am I correct that you are quoting BIOS settings? If so then LUCI5R may also be over the <=1.6v guideline that we have been discussing.

LUCI5R can and do you record your BIOS Thermal Monitor voltages versus your BIOS settings?

Thanks for any input you can give.

rdv, though Cob will have to confirm, but as far as I know he's quoting BIOS settings, and not "actual" volts. And yes, you're right, we're 'all' possibly over the <=1.6v guideline.

I actually do not record BIOS Thermal Monitor volts - I'm just going by the BIOS settings.

[Cob]

COB I have a question for you. With all the great advise your giving LUCI5R, I am assuming you are quoting BIOS settings but not real voltages? I ask this as when I set vPLL BIOS to 1.54v the real is 1.632v. Am I correct that you are quoting BIOS settings? If so then LUCI5R may also be over the <=1.6v guideline that we have been discussing.

Yeah I'm quoting BIOS settings.

Setting PLL above the lowest 1.5v setting is a necessity for clocking the Q9450. 1.6v real may be advised limit by some people, but that may be as low as the 1.5v BIOS setting, and you'll not get over maybe 3.5ghz on that.

[LUCI5R]

LUCI5R -

PLL - Shouldn't need to go higher than 1.58v.

DRAM - It'e estimated that the board overvolts the memory by 0.08v, so 2.02v should give exactly the recommended 2.1v for the memory. And you shouldn't have any problems with the memory til your FSB is over 515mhz

Nb - I see no reason why 1.28v would be the cut-off. We're running 45nm quads with high FSB's. The northbridge needs a good kick in the arse to do so.

Gotcha!!! I'll start PLL @ 1.54v and take it as high as 1.58v if needed, not more then that. I'll set DRAM to 2.02v as well.

vNB - that's comforting! I'll start off at 1.24 - 1.28 as needed, and bring it up into mid 1.30's if needed.
I do have a 40mm fan strapped to the North Bridge, and with all the Case Fans & PCI Slot Fan I have, the case & motherboard is always pretty cool. I had a little Thermal Compound disaster (Detailed Here) which I'll fix today and that should help my CPU temperatures.

Thanks!!!

[WiTcHkInG]

Which chips getting the best result?
X3350 or the Q9450.

[Nordmann]

Hard to say really, I've seen some wicked clocks on both chips but if you ask logic, the xeon chip "should" be a notch better than it's desktop brother as all processors go through a screening process and those chips that excel the tests are picked out for server use and are then put through a rigid test to ensure the stability required for server use. Everything that fails the tests end up as desktop chips if they are not scrapped.


The bios setting CPU GTL 0/2 & 1/3, what does that do exactly?

[Nordmann]

a little update.

I killed my mobo the other day. had a "little" accident removing the original chipset heatpipes.. damn those little pins on the socket are so fragile.

so, I ordered me a new motherboard which I received and installed yesterday and today I went to try it out.
My main goal was 3.8Ghz stable and it it turned out it was remarkable easy to hit 3.8Ghz.

I checked to see that everything was set to to AUTO, set the fsb to 475 and rebooted. after it posted I went into the bios again and took a good look at the bios real volts and went to lower all volts a notch or two and then rebooted to windows. So far I'm prime stable for one hour and counting


CPU PLL is set to 1.54v in bios, bios real is 1.600v
memory timings are 5-5-5-18 T2.
Loadline is turned on.

The temp's are a bit strange. Sensor 1 is the one I trust as it shows reasonable idle and load temps, sensor 2 shows extremely low temps on idle. lowest I've seen is 18c(?) at idle. Sensor 3 and 4 are stuck, won't go one degree below 40. at load those two sensors doesn't seem to agree on what temps they should show and they keep dropping and jumping all over the place. one second they can show 53c, next second it's 61, then 57c, 60c, 55c and so on. highly unstable and I dunno why.

but do you know what the funny thing is?
My main goal was 3.8Ghz stable. If I could hit that I'd be a happy camper but considering how easy it was to hit that I'm starting to wonder how high I can go.

[Earzz]

Nice overclock Nordman, looks promising for me since i also bought the Rampage yesterday.
Got tired waiting on the Maximus II
Hope i will get 4 Ghz with my X3360, 471 FSB is all i need
This weekend we will know, first have to build the damn thing.

[jadeiceman]

Hi, I'm new to overclocking and I've been having issues trying to get my 3.2Ghz OC stable. I have an EVGA 780i mobo, TRUE120 HS with a fan on it, Antec 900, 2x1gb RAM @ 4-4-4-15, and I don't know how to interpret the results I've been getting. Here's a link to the spreadsheet I made. http://spreadsheets.google.com/ccc?k...LxAFs1HA&hl=en
If someone could give me some advice, it would be really helpful.

[jas420221]

I cant see the spreadsheet (im at work, not your fault) so forgive me if I ask or say something that was covered in there....with that said:

Cpu voltage?
SPP voltage?
Stock memory speeds? Current?
CPU FSB voltage?

Maybe create a signature for easier viewing of your rig specs...

[jadeiceman]

Let's see... The best prime95 blend run I had was at
DRAM:1.85 (Auto)
NB:1.45
SB:1.6
VTT:1.4
VCC:1.26875 BIOS (1.232 in Windows)

It lasted for 9 hours and failed on Core 4

[jas420221]

Those volts (NB and VTT) are damn high for 3.2Ghz. I need that for 3.8Ghz...!

First, NOTHING should be on auto really, set the memory voltage manually. Is 1.85v what it needs for stock speeds?

I assume you are at 400x8 so the memory is at stock speeds yes?

However, 9 hours is "Prime stable" to most. I would now add some volts to the GTLref for that core and see if it craps out at 9 hours again. Those volts will not put you in a danger zone but you shouldnt need what you have listed for the speeds you have now.

[jadeiceman]

Yes, memory is at stock speed and the rated timings

I'll try adding some volts to the GTLref and post the results

Thanks for the advice!

[jadeiceman]

If it helps at all there was a run with everything on auto except for the vcc, which was 1.26875v (1.232 real), that lasted 6 hours, but then the computer froze.

[rdv]

The OC goal: 24/7 stable OC with the PC lasting 3+ years to stretch the initial investment

1) Statistics:
a) Q9450 lot #: L803B543 with a stock vid: 1.216v
b) 463FSB; @3.7Gz 39.10&#37; OC with:
Vcore BIOS 1.35v real 1.336v; vPLL set to BIOS 1.56 Real 1.648; vNB BIOS 1.51 Real 1.536.
c) Temperatures during the 24hr prime95 test run:
Lowest (Celsius): Cores 33-37-33-33; Mobo 33; NB 42; SB 38
Highest (Celsius): Cores 71-70-67-67; Mobo 34; NB 50; SB 40
Room Temperature range: 24-25C
d) Stability Test notes:
> All quoted temperatures are from a closed case.
> All core temps reflect Realtemp calibrated values. See section "Observations" for calibration details
> All 24hr+ stability test sessions were performed with Prime95's options: "blended" and 4 cores.
e) Utility versions:
Realtemp v2.60
CPU-Z v1.45
ASUS Probe II V1.04.29
Prime95 v24.14
Phoronix Test Suite (v1.00) (a free linux system benchmarking test suite)
f) OS:
>> Stability testing 32 bit Vista Home Premium
>> System Performance benchmarks 64 bit Ubuntu v8.04

2) Proof:
CPU-Z @3.7GHz and memory at 1112MHz:
24hr Prime95 and Realtemp:
24hr Prime95 and zero errors:

3) OC versus stock Benchmark comparisons:
The html file must be downloaded : http://www.mediafire.com/?9u1xtneybuj

4) Observations/notes:
a) The NB/SB Rampage heat sink was removal and the stock paste was replaced with AC5 paste before the mobo was ever used
b) Pencil Vdroop is a MUST!!! Reduced Vcore fluctuations, essential to achieve the highest OC at reasonable Vcores
c) A 96mm fan directly blowing on the NB reduced the NB temperatures by ~5C
d) All core temperatures reflect the adjustments made after the prescribed Realtemp calibration instructions.
>> Before the Realtemp calibration the temps read (Idle/Load):
#0-37c/47c; #1-33c/37c; #2-39c/43c; #3-40/45 with a room temperature of 25c
>> Realtemp adjustment values for cores 0 to 3 were respectively: -1.5 0.2 -2.0 -3.0
>> After the Realtemp adjustments were applied the temps read (Idle):
#0-33c; #1-34c; #2-33c; #3-32 with a room temperature of 25c
e) The OC strategy was to raise vNB fairly high then concentrate on Vcore until a 24hr stability test was passed. Then test lower vNB's until stable 24hr sweet spot is found.
f) Watch PLL on AUTO as mine was over the <=1.6 target (1.648). In the end I was not able to reduce the real PLL voltage.
g) During stability testing there seemed to be a relationship between the need for more Vcore voltage and core3 while if core2 failed more NB voltage was required. This held true until the last bit of fine tuning.
h) At times core3 fell behind the prime95 testings as much as 5 tests 10 hrs in. Core1 3 tests for 3.6Ghz testing but not 3.7Ghz and all cores eventually caught up before the end of the 24 hrs.
i) When the NB and core load temps lowered despite raising voltages it seemed to indicate the sweet spot had been found. This was observed during both 3.6 and 3.7 testing.
j) Seeking the lowest stable vNB and vPLL took the most time as some tests lasted between 20.5, 19, 12-13 hours and 7hrs before failing.
k) It has taken quite a while to finalize the 24hr stable OC settings:
> I may have not been as patience if I did not have a working alternate rig.
> Assembly started the morning of June 14th first fire up late afternoon. First stability tests started Jun 15th. OC Stability finalized including both Memtest86 and Prime95 >=24hr no errors on June 26th.
l) I used VNC to remote monitor stability tests over home lan and my easy chair. Make sure to turn off any screen saver or power saving features on OS and BIOS as that stops VNC updates can cause a trip to the OC rig to shake the mouse.

5) Other information:
a) BIOS settings:

Test Sessions BIOS Settings	#22: 3.7G 463FSB    vPLL 1.56 MAX vNB STABLE
Extreme Tweaker menu:	
CPU Level Up [Auto]	Auto
Ai Overclock Tuner [Auto]	Manual
OC From CPU Level Up [Auto]	Auto
CPU Ratio Setting [Auto]	Auto
FSB Strap to North Bridge [Auto]	333
FSB Frequency [XXX]	463
PCIE Frequency [XXX]	100
DRAM Frequency [Auto]	Auto
DRAM Command Rate [Auto]	Auto
DRAM CLK Skew on Channel A/B [Auto]	Auto
DRAM Timing Control [Auto]	Auto
	
1st Information : 4-5-5-15-3-42-6-3 (These values are auto-detected)	CPU-Z says    5:6 1112MHz
CAS# Latency [5 DRAM Clocks]	5
RAS# to CAS# Delay [5 DRAM Clocks]	5
RAS# PRE Time [5 DRAM Clocks]	5
RAS# ACT Time [15 DRAM Clocks]	18
RAS# to RAS# Delay [Auto]	Auto
REF Cycle Time [Auto]	Auto
WRITE Recovery Time [Auto]	Auto
READ to PRE Time [Auto]	Auto
	
2nd Information : 8-3-5-4-6-4-6(These values are auto-detected)	
READ to WRITE Delay (S/D) [Auto]	Auto
WRITE to READ Delay (S) [Auto]	Auto
WRITE to READ Delay (D) [Auto]	Auto
READ to READ Delay (S) [Auto]	Auto
READ to READ Delay (D) [Auto]	Auto
WRITE to WRITE Delay (S) [Auto]	Auto
WRITE to WRITE Delay (D) [Auto]	Auto
	
3rd Information : 13-5-1-5-5 (These values are auto-detected)	
WRITE to PRE Delay [Auto]	Auto
READ to PRE Delay [Auto]	Auto
PRE to PRE Delay [Auto]	Auto
ALL PRE to ACT Delay [Auto]	Auto
ALL PRE to REF Delay [Auto]	Auto
	
DRAM Static Read Control [Auto]	Auto
Ai Clock Twister [Auto]	Auto
Ai Transaction Booster [Auto]	Auto
Pull-In of CHA PH1	
Pull-In of CHA PH2	
Pull-In of CHB PH1	
Pull-In of CHB PH2	
	
CPU Voltage [Auto] (Vcc)	1.35000
CPU PLL Voltage [Auto]	1.56
North Bridge Voltage [Auto] (NBCore)	1.51
DRAM Voltage [Auto]	2.1
FSB Termination Voltage [Auto] (VTT)	1.34
South Bridge Voltage [Auto] (SBCore)	Auto
SB 1.5V Voltage [Auto]	Auto
Loadline Calibration [Auto]	Auto
CPU GTL Voltage Reference [Auto]	Auto
NB GTL Voltage Refernece [Auto]	Auto
DRAM Controller Voltage REF [Auto]	Auto
DRAM Channel A Voltage REF [Auto]	Auto
DDR2 Channel B Voltage REF [Auto]	Auto
CPU Spread Spectrum [Auto]	Auto
PCIE Spread Spectrum [Auto]	Auto
	
	
Advance Menu/CPU Configuration:	
CPU Ratio Setting [Auto]	Auto
C1E Support [Disabled]	Disabled
CPU TM Function [Enabled]	Enabled
Vanderpool Technology [Enabled]	Disabled
Max CPU ID Value Limit [Disabled]	Disabled
Intel Speedstep Tech. [Enabled]	Disabled
	
	
Advanced Menu/North Bridge Chip:	
Memory remap (for 64bit OS's) [Enabled]	Enabled

b) My OC Testing spread sheet with the following individual pages:
>> Guidelines: Voltage and temperature guidelines and Q9450 Intel specific information
>> Component Details: My rig component details and a collection of Q9450 lot#'s, users and their achieved OC stats
>> Formulas and OC Info: OC Formula and stats (e.g. FSB -> GHz, FSB -> Memory Mhz ... etc)
>> Asus Rampage Formula BIOS OC Settings (Other's settings): Settings from forum members which OCed this mobo
>> Test Plan: Major check points in OCing rig until it is deemed ready to be installing as the 24/7 main rig
>> Test Sessions BIOS Settings: BIOS settings both passed and failed referenced in test sessions
>> Test Results: Statistics (temps,voltages), what/when failed and notes for each OC test session
NOTE: Not many tests were recorded which did not post, boot Vista or failed prime95 in under 10 minutes
>> There were total of 45 recorded test sessions and 22 different BIOS settings tested
Download OpenOffice v2.x (.ods) format: http://www.mediafire.com/?mdwn5m1fam2
Download MS Excel 97/2000/XP (.xls) format: http://www.mediafire.com/?jjmh5zax4oe

6) My favourite OC resources:
a) Howto: Overclock C2q (quads) And C2d (duals) http://forums.overclockersclub.com/i...howtopic=71656
b) Pencil Vdroop Mod - (I used a 3B pencil) http://forums.techpowerup.com/showthread.php?t=60341
c) NB/SB HS removal (Search for text "Talonman's advice on the removal of the stock ROG cooling") http://www.xtremesystems.org/forums/...d.php?t=162085
d) Q9450 OC / Temps / Settings http://www.xtremesystems.org/forums/...9450+overclock

7) Comments on the results:
a) Positive:
> Exceeded expectations as 3.7Ghz is above the initial target of 3.6GHz
> All except vPLL voltages are below upper limits set in guidelines
> North Bridge voltages (vNB) and temperatures under load were reasonable after tuning
> Despite the seeming ridiculous number of fans the rig is very quiet. I attributed that to using 120mm fans which can push more air at lower RPMs. My HDPC rig has a small case and 4 small fans versus the OC rig with 5x120mm, 1x96mm and 3xMemory fans. The OC rig is as quiet as the HDPC.
b) Negative:
> For a 24/7 3+ yr rig the vPLL voltage exceeded recommended guidelines of <=1.6v by +0.048v. Although the Rampage Formula has LED lights to warn you when voltages are OK/High/Crazy. My vPLL led is still green: CPU PLL Voltage Green:1.50000~1.60000 Yellow:1.62000~1.80000 Red:1.82000~ so I may actually still be safe.
> Under prime95 stress load the highest core temps exceeded targets of <=55C with a 71C to 68C range reached. Considering that I am not a gamer or will not be doing folding the high CPU core temps are a minor issue. Also the 71C was infrequently reached with sustained high temperatures being an average of around 62-58C.

8) I am asking for any comments, warnings and/or suggestions for improvements. I have the following questions:
a) Do you feel that this rig meets the original goals or am I taking too much of a risk?
b) Can anyone suggest how the vPLL can be lowered but without violating other "safe" settings?
c) What would you try to improve/change if this was your rig and you had the same stated goals?

Any comments or questions are welcome.

I hope this information can help someone else trying to OC. In a small way that may pay back all the help I have received from reading these forums.

[jadeiceman]

Holy crap... This is awesome.

[LUCI5R]

Jesus Mary, Mother of God!
Feels like a very professional, complete & authoritative Analysis & Report. Bravo! That is an awesome job & and an equally awesome report.

There are many pointers, tips & understanding to be taken from your achievements & process.

I'll be going through your documents in more detail during the day - but one thing I do want to add, in response to your Question 8 (c), I would certainly, personally if it were my own rig, be concerned about running the rig 24/7 for an extended period on that vNB & vPLL. Quite honestly, I do not know for a fact if it would do any harm at all even in the long run - I have no evidence to support my concern - but I personally would be hesitant.

Regardless, great job!

PS: Haven't been able to get to my rig in the last two evenings (home matters), will be beginning OC today evening and will report back.

[jadeiceman]

If it helps at all there was a run with everything on auto except for the vcc, which was 1.26875v (1.232 real), that lasted 6 hours, but then the computer froze.

I tried adding some volts (65mv) to the Core that crashed, and lowered the VTT to 1.3, but I got errors on Cores 2,3,4 after 2 hours.

And I also kept VTT at 1.4 and Core 4 failed in 1 hour 5 minutes.

I really suck at this.

[rdv]

Thanks LUCI5R, your input is appreciated. I wish I could have formatted the report better (numbers and bullets) but the forum formatting is limited. The report was already very long and I did not want to discourage members from grabbing any nuggets of information that they may have found useful. Please check out the benchmark report as I found the OC Vs Stock comparisons very interesting. It really confirms the positive value of OCing.

The areas I wanted to improve performance was in encoding video and audio (I have large DVD, CD and Vinyl collections) and the benchmarks indicate major improvements with that type of processing.

Good luck on your OC efforts.

[LUCI5R]

The areas I wanted to improve performance was in encoding video and audio (I have large DVD, CD and Vinyl collections) and the benchmarks indicate major improvements with that type of processing.

This part of your post was particularly striking. The primary reason for my upgrade (from P5W DH Deluxe/E6420/2GB PC2-5400) to P5Q Deluxe (P45), Q9450 & 4GB PC2-8000 is Video Encoding, Editing & Authoring. I use my PC primarily for all sorts of audio/video encoding & editing. Mainly High Definition Full HD 1080p Encoding using various formats (x264, VC-1, WMV-HD, WMA 10 5.1 Audio, etc.)

One of the main reasons I chose Q9450, aside from being a Quad Core, was SSE4.1 ... specifically aimed at Video Encoding (For applications that support it). I use Mar '08 version of TMPGEnc Xpress which has full support for SSE4.1 and the upcoming full Expression Encoder & the new version of Vegas Pro, both will support SSE4.1

While @ stock - the difference between my new & my overclocked old setup wasn't breathtaking - I'm expecting "Major" difference after I have successfully stabilized my overclocked new setup. Your 40% overall increase and over 40% increase in encoding processes is extremely promising. I'm impressed, and excited at the same time to try out my processes.

For a video encode process using the following:

Feature length, 2 ~ 2.5 hrs Video
h.264 1080p (1920 X 1080) 6 Channel DTS (5.1 Surround) 12 Mbps MKV Container --> WMV-HD (VC-1) WMA 10 5.1 Surround 12 Mbps 2-Pass VBR

My OC'd E6420 (350X8: 2800Mhz) took 24 ~ 26 hours
My Stock Q9450 (333X8: 2666Mhz) took 17 hours

Can't wait to run the process in a stable OC'd Q9450 (preferably 3666Mhz ~ 3800Mhz)

I'll report back on that.