Overclocking (in)stability (Q6600)

[Shmee]

it was 2 sticks, 4GB, of Gskill pi. My friend has the system now, he says its stable at 3.2 GHz with another gskill kit in there as well. Not even the same kit, and different voltage. Seems to work fine he says.

 

[BonzaiDuck]

780i is fine for OCing Q6600s, at least the EVGA FTW version I had was. It took some work, but I had it at 3.6, just the temps were too high with the cooling I had. My regular OC (and now my friend's) is about 3.2 GHz on the board with a Q6600 G0. Keep in mind with quads, MCP overvolting may/probably will be required.

Per Lopri's response: I had the Striker Extreme 680i. It was certainly "quirky." And you couldn't count on it working with all four memory slots filled, regardless the BIOS revision.

The following year, I picked up the cheaper (non FTW) EVGA 780i board, and it had no problems whatsoever. I OC'd an E8600 to 4.2Ghz, mixed two G.SKILL kits (950 pi's and DDR-1000's), and it's still tip-top.

But there's one more thing I now remember about the Q6600 and certain temperature monitoring software -- I should've thought of it in responding to IDontCare's posts.

RealTemp and CoreTemp got the TjMax on the B3 stepping wrong from the get-go. This meant that temperatures were reported some 10C higher than they should've been. It was only later -- a year or more I think -- that this was corrected. I was stunned that my load temperatures were really as low as reported. It may have been that the responsibility lay entirely on Intel.

This probably discouraged people from pushing the B3-stepping further. I could entertain going to my brother's house and giving him a little extra "zoom" in that system, but . .. "don't fix it if it ain't broke."

 

[lefenzy]

Here's an update. I think I am stably at 3 GHz and 1.25 V with 889 MHz memory, and I will keep it like this for 24/7. Intel speedstep is enabled, and temperatures are fine. That 2% tweak made the difference. The board is an MSI P6N-SLI, using the GeminII cooler.

I could explore whether I could get a few hundred more megahertz, but I probably won't anytime soon.

 

[Schmide]

Here's an update. I think I am stably at 3 GHz and 1.25 V with 889 MHz memory, and I will keep it like this for 24/7. Intel speedstep is enabled, and temperatures are fine. That 2% tweak made the difference. The board is an MSI P6N-SLI, using the GeminII cooler.

I could explore whether I could get a few hundred more megahertz, but I probably won't anytime soon.

Edit: I looked around. It's probably running at a 4:3 ratio. So disregard everything after this.

Something is not exactly right here. Not saying it isn't stable, but what you're reporting doesn't add up.

As far as I know, Kentsfield/Yorkfield do not have half multipliers.

If you're running at 889mhz memory, that would be a 5:4 memory ratio, equaling a 355mhz bus and thus a x9 = 3.2ghz OC. To get 3ghz you would need an 8.5 multiplier which these chips lack.

Maybe download the latest OCCT/CPU-Z to check your results?

The possible FSB speeds for 889mhz

1:1 = 444mhz x 9 = 4ghz
5:4 = 355mhz x 9 = 3.2ghz
4:3 = 333mhz x 9 = 3ghz
3:2 = 296mhz x 9 = 2.6ghz

You really should be running a 1:1 memory ratio till you find the best settings. Higher ratios really yield very little performance benefit vs better overclocks.

 

[lefenzy]

The memory is rated for 800 MHz, so even at stock 266 MHz FSB, the memory would be operating with a divider. I don't think there is any harm in running the memory at 889. My goal was to get to 3 GHz, and after that I decided to see if I could raise memory clocks as well.

I am a little confused by the different clocks for memory. Is DDR2 memory rated at 800 MHz clocked at 200 MHz or 400 MHz?

 

[Schmide]

The memory bus runs at 2x the FSB.

So to get the memory speed

FSB x 2 x ratio

so example

266 x 2 x 5:4
533 * 1.25 = 667

or in the case I think above

333 x 2 x 4:3
667 * 1.33 = 889

All motherboard bios have different ways of getting the divider you want.

I believe for the MSI to get a 1:1 memory ratio. You set the memory speed to 533.

Edit: There is no harm in running it faster. What I'm saying is up your memory speed after you've figured out your overclock. 889 is on the edge of ddr2 800's comfortable range without changing timings and adding voltage.

Edit2: You basically get very little or any improvement running faster than 1:1 memory ratio. Think 0-3%. Easily made up for with a better overclock.

 

[Idontcare]

Edit2: You basically get very little or any improvement running faster than 1:1 memory ratio. Think 0-3%. Easily made up for with a better overclock.

When I was running Kentsfield I was curious how much the ram speed mattered. So I did a few tests:

Mind you I did this at 4GHz, so if kentsfield was going to be data-starved (ram bandwidth limited/impacted) then it was going to show up all the more here versus the impact in a lower clocked system.

And the result I observed for performance in real-world apps in increasing the memory divider above 1:1 was a mere <1% (Case A versus Case B).

Of even less value to performance was worrying about the actual FSB speed, as Case B versus Case C was intended to highlight in which the difference there was 333MHz vs 400MHz FSB (ram was clocked to DDR2-800 in both cases)...performance improvement in real-world apps was basically zero.

And going all out to find how just how much FSB and memory speed mattered to a 4GHz kentsfield CPU, compare Case A versus Case B where the ram goes from DDR2-667 to DDR2-800 and the FSB goes from 333MHz to 400MHz...the improvement is notable in the synthetic benches (that is what they are good for, marketing dollars and all that) but the improvement in real-world apps was barely more than 1%. :\

Pretty much only one thing impacts performance when it comes to Kenstfield processors - CPU clockspeed.