Looking Back 2 Years: An over-clocked Q6600 on 680i

[BonzaiDuck]

In addition to machines built for fam-damn-ily members, I've built two for myself since summer, 2007.

[I should probably edit my "sig" to include "main-machine specs, but for now:]

Q6600, B3, running at 3.2 Ghz first six months, then 3.15 for four months, then 3.0. I'll explain briefly why I stepped it down -- in a minute.
ASUS Striker Extreme motherboard -- 680i chipset, v.1301 BIOS. This version and version 1303 worked well with either B3 or G0 steppings (based on discussions over the last two years with fellow members here.)
Original memory: 2GB (2x1GB) Crucial Tracer DDR2-800 @ 4,4,4,12 and various speeds (for 3.0Ghz, it was 835 Mhz DDR.)
3Ware 9650SE 4-port RAID controller, and 4x Seagate 7200.10 320GB in RAID5
[BFG] nVidia 8800 GTS 640MB graphics
AverMedia M780 Hybrid tuner-capture
Creative Labs X-Fi PCI audio
[The specs for the DVD-burner, front-panel USB hub, fans etc. are incidental]

THE RECENT PROBLEM:

I moved slowly to upgrade OS on my systems. I still have an old machine with Windows 2000 Pro. Everything else -- three home machines including my 680i/Q6600 system -- deployed 32-bit XP Pro or XP-MCE-2005 (same thing). As a former computer professional, I knew we were slowly moving into a 64-bit era, but I was determined to move cautiously. I can't afford watershed expenditures on new software upgrades if I can help it.

SO . . . . to that end . . . . after building a 780i gamer with VISTA-64, I was sure that a 64-bit OS would yield performance improvements over 32-bit that would almost seem like "over-clocking." I'm sure I'm right about this. So I reset the 680i system to stock settings after backing up files and inventorying software -- then, installed VISTa-64 SP1.

And . . Everything was fine, except that the Q6600 system only had 2GB of Tracer memory.

Now -- here's the kicker. I had been saving a pair of G.SKILL DDR2-1000 2x2GB -- previously deployed in the 780i system. I replaced them with G.SKILL Black-Pi modules -- which could be run at tighter latencies and the same speed and voltage. And I had a mishap this year, when setting up a test-system with a GA-EP45-UD3R motherboard and a cheap [BFG] power supply -- the latter which quickly went south and (I'm fairly sure) fried the motherboard and/or processor. I had socketed the G.SKILL DDR2-1000's in that board.

Forgetting that, I went forward to replace the Crucials with the G.SKILLs. Knowing the latency settings and voltage for the G.SKILLs, I skipped resetting the over-clock settings to stock, suffered BSOD problems, and finally cleared the CMOS and reset the BIOS to stock values. So far -- fine.

Then I attempted to re-implement the (3.0 Ghz) overclock settings. I could only run the G.SKILLs at their default latencies. After a few PRIME95 runs and restarts, it appeared stable with 11 hours small-FFTs 0-errors and 10 hours Blend 0-errors.

Even so, I'd had to bump up the VCORE two additional notches to avoid small-FFT failure -- two notches above what the Q6600 required for running at 3.0 Ghz for more than a year.

Then, I chose to run OCCT 3.0.1, and the test failed in 5 minutes. Tweaking other voltages, such as NB-core and CPU_VTT only bought me two extra minutes.

So now, I've replaced the G.SKILL modules with a set of Corsair DHX DDR2-800 4GB 2x2GB.

Everything is running "stock," and my next step is to tighten the latencies to the Corsairs to their 400Mhz/SPD-EPP settings of 4,4,4,12.

But I'm wondering. I had -- for the most part -- run the CPU VCORE well below the upper limit of the "safe-range" spec. Frankly, I don't think it was ever shown to be above 1.31V in CPU-Z even when the BIOS setting was 1.41+ or 1.42.

Maybe two years has taken a toll on a modest over-clock setting -- maybe not.

I'll have to continue tweaking and report back here in this post.

Any observations -- per longevity of systems that are mildly over-clocked, or the possibility that the G.SKILLs were damaged -- are welcome in the meantime.

ONE MORE POINT: This system does multi-duty as a multi-media, HTPC, and "regular business" system. We have always had a rule of thumb that one should avoid using an over-clocked system for "business" purposes, but then, a 3.0 Ghz overclock for a Q6600 -- on a motherboard spec'd to run at 1,333 FSB -- is not much of a stretch. And for the moment, any problems are "an inconvenience" -- all data is backed up.

In the current economy, and under summertime belt-tightening due to an interruption in income, I'm not eager to replace the processor with a last-gen G0 Stepping of the same Kentsfield core. I don't want to replace the motherboard with a comparable ASUS 780i so I can run a Yorkfield.

The constraints are: to replace the motherboard, I might prefer the eVGA 780i for performance reasons, but the Striker II Formula fits the same physical specs for processor-placement and the cooler that sits on top of it. This is relevant to the issue of modifying my ducted-cooling setup for the TRUE.

Any comments on how I might have run out my string, so to speak, for having this system over-clocked consistently for two years?

I'll report back on what happens now as I attempt to raise the clock-speed with the newly-socketed Corsair DHX's.

 

[BonzaiDuck]

MY UPDATE TO A RATHER UNNECESSARY THREAD:

I sometimes post while fixing a "problem" or troubleshooting: it attenuates the loneliness inherent in the trouble.

AigoMorla said on a very recent post to another thread -- at least twice or in two ways: It's probably the memory. He wasn't addressing my current issue, but his remarks reassured me of my own diagnosis.

I think it was those damn G.SKILLs -- and not for being G.SKILLs, but for getting fried by a bad power-supply.

The old over-clock settings seem to be rock-solid under a preliminary 30-minute OCCT/LinPack test. As I said, with the (apparently damaged) G.SKILLs, the test would invariably fail within the narrow range of 4+ to 7 minutes.

[Apparently, for short OCCT tests, you have to set the test for considerably longer than the timespan or interval within which you might expect error: a ten-minute test leaves a four-minute idle monitoring period].

Well, like Hyman Roth says in "The Godfather part II" -- "Dat . . . . is da bidnis . . . . we're in!" A lot of trouble -- to switch OS's, run down the detour of unexpected bad memory, recertify over-clock settings, etc.

But -- Hey! Computing with a 64-bit OS, whether VISTA or Windows-7-64-bit -- it's stunning. It's stunning with stock settings, and it's stunning with over-clock settings.

I think it's coming time to bid good-bye to the 32-bit OS, folks.

 

[BonzaiDuck]

Well -- it turns out -- my problem may not be faulty RAM.

There are two possibilities: the processor has slowly degraded somewhat, or there is some voltage-setting equilibrium that changes when a 2GB RAM kit is replaced with 4GB RAM kit.

I'm not sure which it is. Obviously, the 4GB kit draws more power. How that affects how a motherboard VCORE setting yields a different CPU_Z voltage reading, I don't know. I can only say that I've had to kick up the VCORE four notches above what it was with the 2GB RAM kit to reach the same level of stability.

Comments? Observations? I don't recall having made this sort of change in components before, so it isn't something I've observed first-hand. And frankly, I haven't kept all the same hardware in an OC'd system for two years consistently.