3GB, 4 sticks of Crucial Ballistix DDR2-1000

[BonzaiDuck]

Couple months ago, I had a mal-installed driver for my AverMedia tuner-capture card, which was hogging much of my 2 GB (2x1GB) Crucials. Before we traced the problem to the mal-installed driver so that reinstalling corrected the problem, I decided to boost my memory from 2 GB to 3, by adding a 2x512 kit of the same.

This is a Q6600 (B3) system with the Striker Extreme 680i board and next to latest BIOS. I'd originally "certified" it as a stable over-clock to 3Ghz, and found a higher over-clock to 3.2Ghz.

I left the 3GB installed after fixing the driver problem. I set the command-rate to 2T. I thought that retesting stability with the double-ORTHOS small-FFTs test was good enough, and also ran the large FFT test on the RAM. Didn't think to run the Blend test.

I've been tweaking my voltages again, and thought to give the system another "shakedown" test at the 3.2Ghz setting. Now I find that I get one core failing the Blend Test after what seems to be almost a certain 2 hours, 52 minutes. And no matter how I (cautiously) up my voltages, the Blend Test fails in this less-than-3-hour timeline.

I'm wondering if this might be a symptom of some weakness or failure in the 2x512 modules. I had trouble getting them to socket properly when they were first installed -- the system would fail to post until I switched sockets and pushed them with greater-than-usual effort into the sockets.

The system has seemed stable over the last two months with this 3GB setup, but I reset the over-clock to the more conservative 3 Ghz (1,334 FSB) setting, and it STILL fails the blend test after 2 hours, 52 minutes, give or take a minute or two.

Any ideas?

 

[BonzaiDuck]

Well, so far it appears nobody has any ideas about this.

When I first installed the 2x512 kit in the second pair of slots, I posted a question here about whether I'd have to loosen timings or perhaps adjust voltage after filling all four slots, and someone suggested "Probably not." My testing at that time (without the Blend Test in double-ORTHOS), seemed to indicate they were correct.

Upping the VDIMM voltage doesn't seem to help, and the ORTHOS test still seems to die almost exactly after 2 hours 52 minutes. I'm now loosening the latencies to see if that corrects the problem.

If not, I think I can live with 2 GB in this system -- running XP MCE 2005 and SP2. Or -- maybe I can get Crucial to replace under warranty.

 

[BonzaiDuck]

Well, loosening the timings didn't help.

And it's peculiar that the Blend-Test fails almost exactly between 2 hours, 52 minutes, and 2 hours, 56 minutes -- every . . . . stinkeen .. . . . time. Whether I bump of the voltages or loosen the timings -- 2 hours, 50-some-odd minutes. And invariably, the test fails at the Lucas-Lemar "20K" calculations.

I can't see how the problem would be the CPU, since it ran 6+ hours the other day on the small-FFTs test -- without fail. And I can't see precisely how it would be a problem with the motherboard.

Nope. Next thing to do, is remove that 2x512 kit of Crucials.

 

[BonzaiDuck]

OK -- 1 GB (2x512) less. And now I remember why I didn't like running the Blend Test of ORTHOS to begin with. With only so much memory, it is "paging" constantly, with one instance hogging all the memory and the other instance stalling and paging.

So it may be that the TEST is flawed, and the hardware is fine.

I've also noticed that everyone has suddenly moved to a newer, four-instance version of PRIME95 for stress-testing. There must be some reason for that, and I also remember that some over-clocking gurus here recommended the "small-FFTs" test (which I used) in ORTHOS, while another opinion cites "Blend-Test," but in reference to PRIME95.

If anyone has any skinny or insight about a windows-based memory-stress test that can be relied upon, other than MEMTEST86 (which boots by itself and may not complete all tests on four modules) -- please let me know.


And any other insights will be appreciated, too . . . .

 

[BonzaiDuck]

One more comment: Someone remarked that they prefer ORTHOS because the new, four-thread version of Prime95 "doesn't load up the cores" as much.

That's baloney.

And I'm banking on this version to run "Blend" without problems. then, I'm gonna resocket my 2x512 kit, and see if the same problem occurs.

 

[BonzaiDuck]

So it looks as though this worry about ORTHOS "Blend Test" has cost me almost a couple days -- for nothing.

This raises, for me, the question as to why there seems to be a scarcity of reliable stress-test software available.

Let me speculate.

No software firm will undertake a project to build "stressful" stress-test software, for fear that dweebs and geeks who may not apply it thoughtfully to their hardware might sue for damages -- if they do damages.

Instead, motherboard makers -- occasionally memory module producers -- offer bundled software that does some "mild" stress-testing.

so we go on, getting shareware in a network of enthusiast sites, trusting that the software has no serious bugs or other malware embedded in it.

And there you are. there you are.

 

[boomhower]

I say you are fine. I have never been a fan of long test sessions. I usually give my stuff a one hour run. That is stable enough to me. If you are not doing some long critical calculations you do not need to stress test over a prolonged period, in my opinion. I know a lot of people swear by eight hour runs and some twenty four. Yet on the other end you have the people who are satisfied if it will run their benchmarks and games without crashing. Stability is all in the eye of the beholder. If it stable to you, then let it be.

 

[BonzaiDuck]

Well, there's been time lost, but to good end. This focus of attention has got me attempting to tighten latencies even more. The earlier settings were "low-hanging fruit," but this final go-around has taught me some things, and I've plucked some pretty big apples, oranges and kumquats.