Screensaver interfering with VISTA-64 running Blend-Test

[BonzaiDuck]

Just a heads up -- and for people who may become too casual about stress-testing RAM under PRIME95 Blend Test with (at least) VISTA-64.

My system has two 9600 GT cards in SLI, DDR2-800 4GB 2x2GB RAM. I had OC'd my E8400 to 3.6 (and actually, beyond . . . ) with at least the 3.6 setting using the DDR2-800 4GB RAM kit at its stock DDR=800, [4,4,4,12,2T] and recommended voltage settings.

While small-FFTs gave me a clean-bill of health (13 hours and manually terminated with 0 err and 0 warns) at the 3.6 setting, Blend-Test would result in a BSOD crash after 20 minutes (the default screensaver activation period) -- "Hardware failure -- consult hardware vendor."

RAM tested for around 14 hours error-free with additional runs for specific tests (like Test #5 -- latest version of MEMTEST86+ -- showed two successive 4GB kits (G.SKILL and CORSAIR) exhibiting the same 20-minute-to-BSOD behavior. A 2GB kit of Crucial Tracers would not do this.

While I updated the VISTA-64 driver for the 9600 GT cards, it appears more likely that the screensaver was responsible. Disabling it now shows Blend-Test just ticking along just fine with the 4GB kit after 1 hour and counting.

Perhaps there is some post in the Over-clocking and stress-testing guide stickies about what to disable when stress-testing CPU and RAM. But I thought I would bring this to the attention of members who hang out at this particular forum.

 

[VirtualLarry]

It sounds like your RAM isn't stable, or something isn't stable. I'm not so sure I would simply disable the screensaver and carry on with Prime95 testing. This needs more investigation.

This sounds almost like the "Vista causes SuperPI crashes" that was reported, and I reposted it in the OS forum looking for answers, but people with non-overclocked systems couldn't reproduce the issues. It seems that Vista64 is more likely to encounter errors due to overclocking than XP32 does, probably the 64-bit hardware in the CPU is more sensitive to overclocks, or there is more of it to fail, something like that.

 

[BonzaiDuck]

Yo! VirtualLarry!! Long time, no SEE, bro'!!

It appears more likely that this -- COULD BE -- a matter of tweaking some termination voltages -- and my newbie-like impatience and assumption that "if you've been intimate with the board" for a year, you could take "short-cuts" with the OC'ing, or that a 2GB RAM kit would require about the same settings as a 4GB kit.

Anyway, I just responded to IdontCare in this thread -- see my post of today (8/22/08) and 7:10PM:

I'm SOOooooo STEWWWWW-pid . . .

 

[Majic 7]

I have a similar situation on my system. I haven't overclocked yet because I was getting too anal about temps. I have been running Prime95 and OCCT just to see what my temps were under stress. This has only happened once but I was running OCCT and when the screensaver came up after twenty minutes I had a total black screen that I couldn't get out of, no task bar, just black. Everything was still running so I could still bring up task manager with alt tab and stopped OCCT in processes. Not the same but similar.

 

[BonzaiDuck]

I'd have to go back to Graysky's Over-Clocking Guide stickied to this forum. I have trouble remembering the common wisdom.

But whatever "security/anti-malware" suite I'm running, even the best of them start "doing things" in the background when you least expect it. Same with the screensaver -- hardware and OS/software components.

So I close all that stuff when I begin a stress-testing run. Figure it this way: you're stress-testing the hardware -- not whether or not certain software runs well with the stress-testing programs.

But VISTA is new enough that I wasn't initially successful in turning off the screensaver activation. The same dialogs are built-in, but re-organized in the OS, and you have to drill down to find them.

 

[Majic 7]

It's weird it only happened to me once. Might have been some interaction. I use Nod32, used Kaspersky for a while on my XP machine but honestly it was pretty irritating to me, always asking permission and it never seemed to adapt to what I was doing. Other than that my machine is pretty clean. I ran the stress test a lot because I was getting memory errors at stock. Finally found that I had to disable USB legacy in the bios. Thought the Corsairs were bad and ordered some GSkill the day before I found what I was supposed to do.

 

[BonzaiDuck]

Originally posted by: Majic 7
It's weird it only happened to me once. Might have been some interaction. I use Nod32, used Kaspersky for a while on my XP machine but honestly it was pretty irritating to me, always asking permission and it never seemed to adapt to what I was doing. Other than that my machine is pretty clean. I ran the stress test a lot because I was getting memory errors at stock. Finally found that I had to disable USB legacy in the bios. Thought the Corsairs were bad and ordered some GSkill the day before I found what I was supposed to do.

Interesting. I started with the G.SKILLs, replaced them with the Corsairs -- to find that there wasn't anything wrong with the G.SKILLs. And you started with the Corsairs, and replaced them with a set of G.SKILLs.

 

[Majic 7]

I thought you might be interested in that. Actually haven't tried the GSkills yet, just decided to keep them just in case.

 

[BonzaiDuck]

Originally posted by: Majic 7
I thought you might be interested in that. Actually haven't tried the GSkills yet, just decided to keep them just in case.

I really need to do some web-browsing for serious performance reviews on the G.SKILLs. I've been pointed toward them here at Anandtech and the customer-reviews, and . . . gee . . . look at the frequency count of those customer-reviews -- very popular.

There seems to be mixed perceptions about how tight you can set the timings if clocking them lower than the DDR2-1000 spec for that particular 2x2GB kit, but both the DDR2-800s and the 1000s seem to get thumbs up.

Right now, with these 2x2GB Corsair DDR2-800 DHX-ers, I'm wondering if I can stretch them at the spec voltage beyond 820 or 840. My Crucial Tracer 800's are spec'd at 4,4,4,12, but i can run them to 875 and 4,4,4,10 at 0.075V below their rated, maximum spec.

What really makes me feel stupid, though . . . . the panic I had last week over memory was real, newbie behavior. I'd been so used to over-clocking this motherboard without a single BSOD that I missed a calm determination that it was an under-volted CPU_VTT. Nobody here has CALLED me stupid but me, but I'm . . . . devastated with humiliation.

 

[Majic 7]

GSkills seem to be the ram of choice at XtremeSystems forums. At least on ASUS boards. The only thing wrong with that forum is that the threads can go on for hundreds of pages and it can take a while to find what you are looking for. I actually ordered the GSkills first but my credit card company freaked out over my 700$ middle of the night order. By the time I got it straightened out the sale was over so I got the Corsairs with the cool little fans. Didn't use them on the ram but with the motherboard turned upside down in my 830 they sit right on my Noctua and cool the northbridge.

 

[BonzaiDuck]

We're either "early-birds" or "night-owls" -- it's just turned 6AM here Pacific Time.

$700? You HAD to be ordering stuff besides the G.SKILLs . . . .

With luck, their RMA department won't foul up, and they'll return the box as I sent it. They were really quick in response to my "urgent" e-mail telling that I didn't think they were damaged -- asking them to send them back. As I said, I offered to pay the postage, but they're going to cover it.

 

[Majic 7]

7:00 Mountain for me. I work nights so this is my afterwork time. I was ordering some of the stuff in my sig, my first build.

 

[hclarkjr]

i made the switch from OCZ memory to gskill and do not have any regrets at all. i was like you about making the switch as OCZ had treated me well over the years and still do for my power supply. try them, you will not be disappointed

 

[BonzaiDuck]

Originally posted by: hclarkjr
i made the switch from OCZ memory to gskill and do not have any regrets at all. i was like you about making the switch as OCZ had treated me well over the years and still do for my power supply. try them, you will not be disappointed

I DID try them, but for an earlier RAM generation. Their DDR-500 EL Gold modules were touted in Tom's Hardware feature back in January, 2004, and I'd put them on my "gotta have" list for a Pentium 4 Northwood/Prescott build.

Those were replaced with a set of their Platinum DDR-400's. Why was that? Because you could set the voltage well within spec, hold the 400-Mhz-spec latencies at 2,3,2,5, and pump them up to about 452 Mhz. Those modules were legendary!

Somehow, everyone was pointing to Crucial by the time I started on my dual-core DDR2 adventure. But I can't say "somehow." I was looking for "latency elasticity" similar to what I'd found with the Platinums. That's how I burned out a set of DDR2-1000s -- running them closer to DDR2-710 and 3,3,3,6 . . . . and CMD=1T It may not have been so much the tight latencies or just-below-spec voltage: Their tech-support went "Ballistic" when I told them about the 1T command rate. But that was after I'd received my RMA replacement kit . . . .

 

[VirtualLarry]

Originally posted by: BonzaiDuck
Somehow, everyone was pointing to Crucial by the time I started on my dual-core DDR2 adventure. But I can't say "somehow." I was looking for "latency elasticity" similar to what I'd found with the Platinums. That's how I burned out a set of DDR2-1000s -- running them closer to DDR2-710 and 3,3,3,6 . . . . and CMD=1T It may not have been so much the tight latencies or just-below-spec voltage: Their tech-support went "Ballistic" when I told them about the 1T command rate. But that was after I'd received my RMA replacement kit . . . .

I don't understand how you would burn out a set of modules at tighter timings, especially if you were running below-spec voltage. If anything, it would simply fail to work reliably, but not burn out.

 

[BonzaiDuck]

Well, Larry, I bailed out of a physical-science undergrad program after getting through nuk-aler physics and thermodynamics with a lot of sweat, and only remember a few things about voltage, amperage and wattage. So . . . . "electrical engineer . .. NOT!!"

But -- Ah TELL YA, Bro, their tech-support really went bonkers when I mentioned CMD=1T. The latencies may be another story. Either way, tighter CMD and latencies means more likely higher voltage -- that's the trade-off. But -- you know how the rep(utation) of Crucial has taken a doo-doo. [Or no insult to their tech-support rep to say he crapped when he heard "1T."] I've got some Tracers running at DDR= 875 and 4,4,4,10,2T -- since December -- at 2.125V "set." On the ones which died, I'd only had them set at 2.175V, and only briefly (for a day or so) flirted with "set" value of 2.20V

The 1T command rate was worth a chunk of bandwidth in synthetic comparisons, but not as much as squeezing down tRC to >= tRP + tRAS. I didn't tell them about THAT, but I don't think that is detrimental to the RAMs. If anything, it might be detrimental to your data, but it's just never happened. They're rock-solid that way -- the cautious person could make ">" mean a couple integers higher.

 

[BonzaiDuck]

One more, Larry, as separate post rather than edit.

I DO think I discovered that running a divider like CPU : RAM of 3:4 attenuates slightly the need for a voltage increase on tighter latencies, but it isn't much. Or maybe more correctly, I discovered that it attenuates the need for more volts between 2T and 1T. Maybe I imagined it, but I thought for sure that I observed it. Like trout-fly political issues, there's really a difference betwen "believing," "suspecting," "proving" and "knowing." But if "memory" is off forum, that remark goes even further but for its use as analogy.

 

[Dadofamunky]

(Smile.) Just a thought, I gave up on screensavers on Vista 64. Some that are offered on the Net are just horid junk, others are outright spamware.

Now, I just put my system to sleep.