Kougar

Alright, since ya wanted more details. I bought a G0 Q6600 to replace my E6300.

"G0" Q6600 VID 1.200v
B2 E6300 (14 months old) VID 1.325v

E6300 OC's like a dream, fully small-fft & large-fft stable for 24 runs including 24/7 F@H use at 3.8Ghz. 7 x 540FSB max on both a 965P DS3 and P35 DQ6 boards, anything higher is not stable at any voltage.

Q6600, this one has been an overclocking nightmare. Spent over a month trying to figure it out. It DOES overclock, but gives very specific errors once at or over 400FSB.

Let me preface by stating the Q6600 is 24 hour 4x Small-fft or 4x Large-fft stable at up to 3.6Ghz. I did not try 3.8Ghz due to the F@H problems. The CPU boots Vista at 4.1Ghz long enough for a quick SuperPI run. I had almost a dozen screenies of various test runs that I spent the entire month of August collecting, most which got lost when I corrupted Vista to the point the recovery disc couldn't find the instllation. These are a few I had on another system, some are from XP and some were from the nuked Vista install. Once I noticed the F@H problems I re-ran many of these runs for 16-25 hours but didn't bother to make screengrabs.

3.0Ghz 12 hours 1.2v Link
3.2Ghz 22 hours 1.34 to 1.36v Link
3.4Ghz 20 hours 1.3v Link

This chip would run at 9x333=3Ghz @ stock 1.2v for 24 hours stable in any test or program I could think of to try. I used 2x Orthos programs and also the 4-thread capable Prime95 V2 program. I finally ran fully stock settings at 6 x 400FSB = 2.4Ghz and still had the errors in folidng@home applications, so my only conclusion I could make was it has a highly specialized defect on one of the FSB portions of the chip.

I have tried the SMP client and 4 of the console Folding@Home clients and all exhibit the same premature NaN and EUE of work units when < 1% complete, even after a 24 hour Prime/Orthos test run when using 400FSB or higher regardless of CPU multiplier. F@H works fine at 9 x 333FSB.

I know the motherboard is capable of a stable 540FSB with my E6300 and using F@H, so the board is not a limiting factor nor is the RAM. Fiddling with the FSB voltage and relaxing system timings and the MCHBar strap had no effect on the Q6600.

As another piece of the puzzle... total system power draw from the back of the PSU is an about 285 watts STOCK with GPU idle. 3.6GHz @ 1.325v is around 360watts. 3.6Ghz @ 1.4v is around 440watts. 3.8Ghz @ 1.4v is 460watts. Actual core temps don't even reach 70c or come close, but my P3 Kill-a-Watt meter does not lie. I ran ATI tool to load the single 8800GTS just to spike the power draw a bit higher and test the PSU itself... Link When the Q6600 pushes the system to a peak of 465 watts with the GPU idle the system will reboot itself.

Here is what it comes down to. Either Orthos/Prime has to be correct, or F@H simple has a bug in it. However: if the Folding@Home software was at fault then why does it work error free at stock, and at a 333FSB. I spent over a month troubleshooting the Quad, before I figured it had to be a specialized defect in the chip itself. That other article mentioned on Madshrimps regarding why some chips have "FSB walls" seems to back this up. I leave the system at 3Ghz for 24/7 use now, is at least half of what I thought I would get.

Sidney

Some of the Q6600 G0, but not all with this problem?
Did you get it from newegg?

__________________
lazyman

Opteron 165 (2) @2.85 1.42 vcore AMD Stock HSF + Chill Vent II

Kougar

Oh, I extremely doubt most have this issue else I would have easily been able to find at least one other person with the same problem. I think it was just (bad) luck of the draw...

It was from ClubIT, with a pack date of about 7/26 going off the top of my head.

I can live with this Q6600 G0 since it does 333FSB... saving up for Nehalem and a good Nehalem capable board now.

Sidney

"Beam Me Up, Scottie!"

__________________
lazyman

Opteron 165 (2) @2.85 1.42 vcore AMD Stock HSF + Chill Vent II

thorgal

Honestly, I still think it's your motherboard.

Why do you think you can compare a FSB OC of a dual core versus a quad core. To scale a quad core beyond 400FSB on a 965 chipset without a GTL-ref mod, you need to be A) Lucky B) Very lucky C) Extremely lucky. I hope you catch my drift

P35 is another story, and needs PLL tuning and/or GTL-ref tuning (if the board supports it) to scale very high. P35 does scale quads higher than 400 though, but still far from 500FSB without any GTL or PLL tuning. +500FSB on any board and a quad core : show me

Edit : for the record : your chips is probably one of the most impressive I've ever seen. And I don't mean jsut the VID of 1.2V Don't know what cooling you're on, but a windows run at 4.1Ghz, even totally unstable is absolutely incredible.

__________________

Kougar

I have not tried the Q6600 in my 965P-DS3 because it is a Rev 1 board. But I also agree, shouldn't expect a great overclock using 965P and a Quad. Revision 1's only have 3 Vregs, so if I plug a Quad into it I would blow my DS3 mainboard easily. Big reason why every Intel board Gigabyte made since then has a minimum of 4 Vregs, usually 6 now.

Regarding the P35, I have simply not seen others having issues with the P35-DQ6. The funky thing is, the P31-DS3L I am testing offers a GTL-Ref ratio setting (That I don't know how to use, yet)... but my P35-DQ6 does not. Yes, both running latest BIOS. No Gigabyte board I have seen to date offers PLL settings, unfortunately.

I honestly never even gave thought about 500FSB for a Quad, I completely agree that takes a great deal of fine tuning that I personally wouldn't want to mess with to such a degree. All I had originally wanted was 9 x 400FSB = 3.6Ghz... basically 1/3rd of the original speed, and perfect for 1:1 ratio operation with the RAM, which also lets me use a tight MCHBar strap for best performance. I compared the E6300 because it is the only other CPU I have, and I've tested it thoroughly on this DQ6.

I just wish I knew what F@H was doing that Prime95/Orthos is completely unable to detect, because my faith in Prime/Orthos has been marred from this. Not even using Orthos with its built in Gromacs core test running 4 threads detected any problems.

Anandtech's G0 Q6600 has a VID of 1.5v. Their top stable overclock was not very good either, according to Anand it was similar to their B3 Q6600!! My only issue with 4.1Ghz is that the CPU needed higher voltage... and higher voltages caused it to peak over 465watts and instantly reboot itself once it crossed that threshold.

But honestly if the CPU tests stable but still errors in specific apps at anything near or above 400FSB then it just isn't very meaningful in my humble opinion to test higher, 4.1Ghz or not. If I can't use the overclock then it's just not any good. Maybe Gigabyte will "unlock" a hidden GTL setting on the P35 since P31 boards come with it, but I am not holding my breath.