HOWTO: Overclock C2Q (Quads) and C2D (Duals) - A Guide v1.7

[Acanthus]

Memory speed and timings have very very very small performance differences in games.

Encoding and Winzip/Winrar are the only things i can think of where it makes any difference at all that isnt within a 1% margain of error.

 

[RobertR1]

Originally posted by: Acanthus
Memory speed and timings have very very very small performance differences in games.

Encoding and Winzip/Winrar are the only things i can think of where it makes any difference at all that isnt within a 1% margain of error.

Thanks for the reply.

I was just reading this AT article: http://www.anandtech.com/memory/showdoc.aspx?i=2916&p=6

Some interesting stuff in there. I'm gonna stick to getting my q6600 oc'd and stable before even going after the memory.

 

[myocardia]

Originally posted by: Acanthus
Memory speed and timings have very very very small performance differences in games.

True, but when it does make a difference, keeping the RAM 1:1 with the lowest timings possible always outperforms higher speeds (non 1:1), with higher timings.

edit: You'd never be able to tell the difference, though, unless all you do is benchmark, with memory-intensive benchmarks.

 

[RobertR1]

Originally posted by: myocardia

Originally posted by: Acanthus
Memory speed and timings have very very very small performance differences in games.

True, but when it does make a difference, keeping the RAM 1:1 with the lowest timings possible always outperforms higher speeds (non 1:1), with higher timings.

edit: You'd never be able to tell the difference, though, unless all you do is benchmark, with memory-intensive benchmarks.

So in theory what I'm proposing is ideal? CPU at 400x8 and RAM set to 4-4-4-12 (or tighther) at 800?

 

[hokiealumnus]

@ OP - FYI, the Speedfan beta addresses the necessity for the Core temperature offsets. The issue had to do with identifying the stepping of the chips but has been fixed and should read all cores accurately without the offset.

 

[graysky]

@robert - you can do the experiment: select your top 5 applications/games and test them with mem 1:1 @ 4-4-4-12 and 3:2 (or whatever) @ 5-5-5-15 and see if you get a difference. Anything less is just speculation.

 

[cubanx]

Great guide. Excellent work

 

[myocardia]

Originally posted by: graysky
@robert - you can do the experiment: select your top 5 applications/games and test them with mem 1:1 @ 4-4-4-12 and 3:2 (or whatever) @ 5-5-5-15 and see if you get a difference. Anything less is just speculation.

Actually, I already had, however I didn't benchmark any games; I had already read that games benefit slightly from higher RAM speed, than from lower speed @ tighter timings. Of course, from my benchmarks, using only two different video endoding/conversion softwares, there was a very slight decrease in time (increase in performance) from 1:1 266 Mhz/533 DDR) @ 3-3-3-8, than there was from having it @ 1:4 (or would that be 4:1?), with the RAM @ 5-5-5-15 @ 1,000 DDR.

This review/comparison actually shows that some encoding apps benefit slightly, while others either don't, or lose ground: firing squad article. Also, that article has quite a few games benchmarked @ all of the different timings & speeds. Notice that with every app, including the games, it would require a benchmark to tell the difference in performance, like I had said in my first post.

edit: IIRC, the Athlons gain considerably better performance with lower latencies, at the same RAM speed, because of the IMC. At least, that's what I remember a few articles about them saying. Of course, those same articles also said that the Athlons also benefitted considerably from higher speed RAM.

edit #2: My tests were performed with a Q6600 @ 2.4 Ghz, and 2x1 GB of Crucial Ballistix PC8000. And no, I didn't take any screenshots, because 1) I only did them for myself and 2) like the article that I linked, there were no huge, or even large, gains either way.

 

[QueBert]

great guide, I'm new to Intel systems (last week...) on my MSI P35 Neo Platinum I can't find an option to change the CPU multiplier, it's @ 7 right now, so I set my memory to 1:1 4-4-4-12 (Crucial ddr2-800) which is what Crucial specs it at. And pumped my FSB up to 400 to hit 2.8ghz. Is this going to be give me results similar to a higher multiplier with a lower FSB? I'm running the stock heatsink for now but plan to upgrade to something better this weekend. I'm confused why I can't change the multiplier, PCClub says the cpu I got is a G0 step. Coretemp is showing both cores at about 60, I know stock fan sucks, not doing anything intensive. Maybe I'll push the FSB back down to 333 and see what they read at.

any suggestions on how I can access the multiplier, in the cell menu there's nada, I went thru the rest of the Bios, and unless I'm blind the option just isn't there. From what I've read this is a pretty decent mb for OC'ing.


*EDIT* damn I'm a tard tonight Coretemp is showing F, 60f is way low ha. Any ways going to run prime95 while I sleep and hope for the best. If I get no errors by morning maybe I'll push up the FSB to 415 and press my luck.

 

[wgoldfarb]

Graysky,

Excellent guide! :beer: I had read lots of other guides to OC, but yours was the most complete and straightforward I found. After several months of "thinking about it" I finally used your guide to do my first OC.

I now have my E6600 stable at 3.0 Ghz (333x9) after 8 hours of Prime 95 testing. BIOS Vcore setting is 1.2875V. Actual Vcore is 1.248 at idle and 1.232 under 100% TAT load. I have my RAM dividers at 4:5 so I am overclocking my DDR800 ram slightly. CoreTemp temps for my hottest core are 42C at idle and 57C after 10 minutes of a 100% TAT load. CPU Cooler is a TR Ultima 90 with a 120 mm Tricool fan running at medium, so I have some headroom in terms of my temps. Given all this and based on the feedback of people in this forum, I am trying for 3.2 GHz.

My first attempts at 3.2 have not worked so far. I lowered my RAM divider to 1:1 to undercclock my RAM and eliminate it as a bottleneck. I raised my FSB to 356. I had to increase Vcore three notches (!) from 1.2875 to 1.325V in the BIOS to get the system to even boot into Windows, and the OC is still not stable (it will boot into Windows, but will restart after less than 10 minutes of a 100% TAT load, which is what I use to measure load temps). I'm not sure why I needed such a big jump in Vcore, seems unusual. My FSB termination voltage, NB Vcore, SB Vcore and Chipset voltage are all set to auto. Could there be something else that is causing the lack of stability, or do I simply need more voltage?

I am hesistant to raise Vcore further because the VID of my E6600 is 1.325 -- same as my current Vcore BIOS setting. So, before I continue I want to make sure I understand what VID is.

If I understand VID correctly, it is a value set individually for each CPU at the factory to indicate its safe operating Vcore under "normal" conditions. The Intel spec finder says that the E6600 VID ranges from 0.85V to 1.5V, but my VID is 1.325V, so I assume 1.325 is the upper limit of how much Vcore can "safely" be provided to the CPU and stay within its operating parameters. Is this correct? If not, what is the significance of VID, and what is the "normal" safe Vcore that I can provide to the CPU without exceeding its "normal" operating parameters?

Assuming my understanding of VID is correct, I don't want to exceed my VID because I want this to be a long term OC that I can use 24/7. However, with my Vcore setting of 1.325 in the BIOS and my Asus P5B Deluxe, the real idle Vcore is actually well below 1.325 (it is just shy of 1.30V). Given this, I assume it is safe to increase Vcore one more notch. I am guessing that --regardless of whatever the BIOS Setting is-- as long as real Vcore stays well below the 1.325V VID, my CPU is within its "safe" operating specs even if the BIOS setting would suggest otherwise.

So, is all this correct? based on how Vdrop and Vdroop on my ASUS P5B Dlx seem to behave, if I increase my Vcore BIOS setting one notch to 1.3375V I am guesstimating that my "real" idle Vcore will be around 1.312V and my load Vcore will be around 1.296V. Is this a "safe" voltage for my E6600?

I am quite happy with the 3.0GHz I already have, but if I can safely make it to 3.2 I will be even happier Still, if it requires upping my voltage from 1.248 to 1.312 I may scale back to 3.0 GHz for 24/7 use, unless the consensus here is that 1.312 is still a safe operating Vcore for every day use.

Thanks again for a great guide, and thanks in advance for any guidance!

 

[QueBert]

interesting, I ran prime95 all night when I got up coretemp was showing my temp @ 39c, stock cooling with my FSB @ 400. Since my chip's multiplier is locked all I can do is push the FSB up a bit more. I know it varies from chip to chip, but what's a decent expectation for me to hit on stock cooling? My chip only went up about 6c going from 2.3 to 2.8 I expected higher temps. I googled and read people claiming 475 on air but that's probably with some monster heatsink and even then 475 sounds awfully high.

 

[Build it Myself]

Sort of off topic with the discussion at hand but I was curious about something I had read before. Does M0 stepping have a significant impact? The reason I ask is I am going to buy a processor friday, either a 2160 or a 2180. @ clubit they are selling chips with M0 advertising and I can't find a definitive source for information regarding this. Intel's site just says it's updated processor code, but not much else.

Thanks for pointing me in the right direction!

John

 

[wgoldfarb]

update (Follow up to my question 3 posts above this one):

I read somewhere that the P5B seems to have an FSB "hole", from about 330-350FSB to around 400 FSB, where it is very hard to get the board stable. I admit I did not understand the explanation but it is related to how the P5B handles straps. This "problem" disappears as soon as the FSB reaches 401. I have seen posts from people with unstable overclocks at, say, 350FSB, but perfectly stable OC at 401 or above. So, I tried increasing the FSB to 401, leaving the RAM at 1:1, leaving Vcore at 1.2875, and lowering my multiplier to 7 to underclock my CPU -- just to make sure the board itself was stable to 401. These settings were Prime95 stable, so the Board itself can handle 400FSB no problem.

I then increased the multiplier to 8 to get 3.2 GHz. Once again I had to increase Vcore in the BIOS all the way to 1.325, and the overclock was still not stable at that Vcore level (as before, it rebooted after a few minutes of 100% TAT load).

If I understand these results correctly, it looks more and more like I simply need more Vcore to get to 3.2 GHz. So I am back to my original questions about VID:

- Is it safe to exceed a BIOS Setting of 1.325V (for 24/7 use), if the VID on my E6600 is 1.325?

- regardless of the BIOS setting, is my CPU within its "safe" operating parameters as long as real Vcore (i.e. after Vdrop and Vdroop) stays well below the 1.325V VID, even if the BIOS setting would suggest otherwise?

- More specifically, is it "safe" (for 24/7 use) to increase Vcore setting in BIOS to 1.3375 (I am guesstimating that real Vcore at idle will be at or below 1.312, and Vcore at load will be at or below 1.296)?

Any insight will be greatly appreciated!

 

[graysky]

@wgoldfarb - man, that's a huge post!

First off, glad you found the guide useful. Nothing is more rewarding (except maybe cash) than knowing people are actually getting some mileage out of work like this.

You might not get 3.2 so easily. I have found that there is a pretty linear zone for a chip in terms of vcore scaling with FSB.. up to a point (around 333 MHz on my system). To go beyond that and get an extra 10 % for example requires a disproportional amount of vcore to do it. My q6600 is like this. I can run 2.4 GHz - 2.7 GHz on a mere 1.150 V, to step up to 3.0 needs 1.2625. 3.24 requires over 1.3 or close to 1.4 I forget since I never try anymore... don't forget that heat increases with the square of the voltage.

Also, a note about "auto" setting of other voltages... it doesn't always get it right. You might wanna manually set them higher to see.

I wouldn't exceed the recommend VID unless you don't care if you damage or take some life off the chip.. but that's just me. I know others in here and else routinely run their chips @ higher-than-recommended vcores.

About VIDs... I don't understand them. I would go with Intel's recommended vcore range. You might also try the vdroop mod for your P5B-Del... it works like a champ on mine (link in the main overclocking guide).

Yeah, the P5B and all P965 boards have a "hole" around 360 MHz as your describe it, see this url for more.

To answer your questions in your last "update" post:

No (I think)
Yes
Maybe

Again, pencil mod that board which will help you a lot at your idle state. Personally, I won't exceed the max vcore. How are your coretemp temps at that high vcore by the way. Also, don't use TAT to stress your system. Prime95 v 25.x is MUCH better.

 

[daviclond]

Is my CPU speed being throttled?

Here

At full load im getting steady temps of 65-66 on two cores and 63 on the other two. When I first loaded up cpu-z it had my CPU speed at 305x6, despite the fact it's set at 305x10. It eventually updated to the correct value, but on instance of Orthos is still reading it at a 6x multiplier.

I realize Orthos just makes an estimate that might not be accurate, but is it possible something could be automatically restricting the clock speed of two of the cores? Or should I put the differing multiplier down to error and the lower temperatures on two cores down to coincidence?

 

[graysky]

6x to 10x back to 6x is speedstep and totally normal... it depends on your load/idle state. My Q6600 jumps between 6 and 9 for example.

As you pointed out, orthos estimates the multiplier and FSB. You should ignore them; CPUZ is right.

What kinda cooling do you have? Those temps are on the high side for that vore @ 3.0 GHz. You might wanna try lapping your HS and your IHS. I did it and dropped my load temps by 7-10 depending on the core (12-15 %).

Q6600 IHS lapping
Ultra 120 Extreme

 

[jmmtn4aj]

Guys, is the voltage CPU-Z displays accurate? I always assumed it wasn't because it tends to be really far from the voltage set in BIOS, but after reading up on vdroop.. Can vdroop really be so much? I have vcore set to ~1.49 in bios, but CPU-Z shows 1.408 while folding on the SMP client, and 1.392 (fluctuating to 1.408) on Prime95

 

[graysky]

If you're using version 1.41 (they fixed the vcore problems earlier versions of CPUZ had) then it's probably accurate. Vdroop gets worse at higher vcores by design. My advice is to look for a pencil vdroop mod for your board. My P5B-Del has 0 vdroop after doing it.

 

[wgoldfarb]

Originally posted by: graysky
@wgoldfarb - man, that's a huge post!

LOL, I know. Sorry. I wanted to give you as much details as I could, and after I had posted it I realized I had rambled far too much.

Originally posted by: graysky
How are your coretemp temps at that high vcore by the way.

With BIOS Vcore at 1.325 my temps increased to 42C at idle and 59C at 100% TAT load. My room temp was 2C lower than before so maybe I should add 1 or 2 degrees C to make everything comparable. Either way I still have some headroom in terms of temps -- I can always increase the speed of my fans.

Originally posted by: graysky
Also, don't use TAT to stress your system. Prime95 v 25.x is MUCH better.

I do use Prime 95 to test for stability, following your recommendation in the first post of this thread. I only use TAT for 10 minutes to see how high my temps would go in a "worst case" scenario.

Thanks for all the info, again! :beer:. I will follow your advice and NOT increase Vcore beyond 1.325. I don't think I would feel comfortable doing that, even with low temps. As far as the mod, i still think of myself as very much a newbie (this is my first build) -- I don't think I'm ready to mod my board just yet (i'd like to learn more about Vdrop and Vdroop before I do, so I have some reading to do). I will keep experimenting with other settings and see if I can get the OC stable at 3.2 with 1.325 Vcore. If I do I will post here. If not, well, 3.0GHZ is not bad for a complete newbie doing his first OC! (A newbie with lots of great help from this forum!)

 

[daviclond]

Originally posted by: graysky
6x to 10x back to 6x is speedstep and totally normal... it depends on your load/idle state. My Q6600 jumps between 6 and 9 for example.

As you pointed out, orthos estimates the multiplier and FSB. You should ignore them; CPUZ is right.

Alright, thanks for the confirmation. It just raised my suspicion that cpu-z initially reported a 6x multiplier and an orthos torture test still displayed it to be 6x after 45 odd minutes of testing.

What kinda cooling do you have? Those temps are on the high side for that vore @ 3.0 GHz. You might wanna try lapping your HS and your IHS. I did it and dropped my load temps by 7-10 depending on the core (12-15 %).

Q6600 IHS lapping
Ultra 120 Extreme

I've got a Zalman 9700, I got this system from Scan.co.uk and previously had their 3XS watercooling system installed but was getting insanely high temps (60 at idle, 85-90 full load) so had to return it and ask for a part change. Given that it ran high on two different HSFs it probably is a problem that could be fixed with lapping, but to be honest I'm not so hot at messing around with computer hardware.

I'll consider it in the future maybe -and thanks for the link- but this is my first OC so I'm being pretty moderate with my speeds (currently debating between going for 3.2 or 3.33 Ghz). I don't think the extra heat will be too big an issue given that I'm trying to leave plenty of extra leeway for safety anyway- I'll see how things pan out

 

[graysky]

Just updated the first post of the thread to reflect new info, rewordings and updates. (Version is now 1.3).

Enjoy!

 

[WaiWai]

Very well-written.
Thank you so much. It is very helpful.

 

[arthur2007]

Thanks for the awesome guide - I used it to overclock my comp. Here are the spec report by CPU-Z

CPU : Intel Core 2 Quad Q6600
CPU Arch : 4 Cores - 4 Threads
CPU PSN : Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz
CPU EXT : MMX SSE SSE2 SSE3 SSSE3 EM64T
CPU Cache : L1 : 4 x 32/4 x 32 KB - L2 : 2 x 4096 KB
Core : Kentsfield (65 nm) / Revision : G0
CPUID : 6.F.B / Extended : 6.F
Freq : 2136.61 MHz (356.1 * 6)
----------------------------------
MB Brand : Gigabyte Technology Co. Ltd.
MB Model : P35-DS3R
NB : Intel P35/G33/G31 rev A2
SB : Intel 82801IR (ICH9R) rev 02
----------------------------------
RAM Size : 2048 MB
RAM Freq : 356.1 MHz
RAM Type : DDR2-SDRAM Single Channel
RAM Ratio : 1:1
RAM Timings : 4-4-4-12
----------------------------------
Memory G.Skill 2x 1024MB (PC2-6400) +0.2V

However I have a question, the speed report by cpu -z is way to low 2136 (356x6) since the current I set in my Gigabyte-GA-p35-dsr MB is CPU 3.2Ghz = 356x9 at vcore 1.33125

Is this a problem ? other program such as coretemp, system information report correctly that the speed is 3206mhz
The version I have for CPU-z is 1.41
. Any suggestion is welcome

 

[wgoldfarb]

Arthur,

The only thing I can think of is that you may have C1E or Intel SpeedStep (EIST) turned on. Either one of these options will lower the voltage/multiplier on your CPU when it is idle. So, even if you have your multi set at 9, the moment your CPU goes idle C1E or EIST will come into play and lower your multiplier to its lowest value -- 6 for your Q6600.

Try loading your CPU and see if the speed reported by CPU-Z increases. If it does, you likely have C1E or Speedstep enabled in your BIOS.

The other alternative, of course, is a bug in CPU-Z

 

[WoodButcher]

Originally posted by: wgoldfarb
Arthur,

The only thing I can think of is that you may have C1E or Intel SpeedStep (EIST) turned on. Either one of these options will lower the voltage/multiplier on your CPU when it is idle. So, even if you have your multi set at 9, the moment your CPU goes idle C1E or EIST will come into play and lower your multiplier to its lowest value -- 6 for your Q6600.

Try loading your CPU and see if the speed reported by CPU-Z increases. If it does, you likely have C1E or Speedstep enabled in your BIOS.

The other alternative, of course, is a bug in CPU-Z

Nope, I had this too @ 3.6 and higher C1e speedstep all disabled, cpu-z was right. I was using prime 25.4. I read somewhere that it is an internal failsafe. I don't know this to be true as I'm sure I read this in a forum posted by god knows who...