What's the latest skinny on the best DDR2-800/1000+ 4GB kits?

[BonzaiDuck]

I can see everyone's all goosey about DDR3 now, but I have to replace a 4GB kit of G.SKILLs. I'd return them under warranty, but since they went "South" after a few days of use, I think I want to try a different "flavor of manufacture." [And -- Geez! -- I only turned up the RAM voltage to 1.975V on these, when the G.SKILL spec says "2.0V to 2.1V."]

I'd be partial to a set of DDR2-1000's that are "downwardly elastic" in latency -- that is, such that a latency spec of (say) 5,5,5,15 @ DDR=1000 can be run at DDR=800 with settings like 4,4,4,12.

I never got a chance to tweak the G.SKILLs. I worked on the preliminary over-clock settings with memory at default, and was planning to save the "latency tweaks" for later. But I read enough customer reviews that I'm concluding that the G.SKILLs are lackluster for "latency elasticity."

PRices are low right now. Any suggestions? Crucial has come out with 4GB kits that I didn't think existed 6 months ago. AData looks promising. Mushkin looks "not bad." You can even get a set of OCZ Reapers for around $90 -- with the funny-lookin' heatpipes sticking up from the top of the modules.

 

[DSF]

The G.Skill DDR2-1000 kit is actually one of the most commonly recommended around here. Everybody gets a bad part once in a while, that's just part of the hobby. I had to return my P35-DS3L because the NB heatsink was broken, and my Spinpoint because it was DOA.

I would take advantage of the warranty service and get a free kit rather than spending the money all over again.

There's a 2x2GB Mushkin kit that runs at DDR2-800 5-4-4-12 @ 1.8V if you do decide to buy new RAM. Mushkin is a company with very good customer service from everything I've heard. I'm curious as to why you rate them as "not bad."

 

[BonzaiDuck]

You're probably down center-field with your response, DSF.

Generally, on these "limited lifetime" warranties, I've had service as a whole that is "also -- not bad." The best of it was OCZ's willingness to replace a set of Gold EL's (DDR) with a pair of XTC's. The worst was Crucial's replacement of a DDR2-1000 kit with a pair that simply wouldn't run at a 1T command-rate, even though every Crucial kit I've used and purchased brand-new would do that. Of course, when I raised it as an issue, Crucial came down on my like a ton'a'bricks: "If you run them at 1T, it will void your warranty!!" [That rhymes!] Both OCZ and Crucial were swift and accommodating in the replacement.

But as you said -- "that's just part of the hobby." The reason I might just buy a new pair -- even while RMA'ing the old -- is that I can get the new ones in much less time than the warranty-replacement turnaround. And "part of the hobby" can be an "impatience to get things done" so I can clean up the cyber-junk in this g*****n room!!

And -- you're absolutely right about the Mushkins. I can't say that I've gone through -- or just used -- more memory sticks than a gorilla goes through bananas on a food-binge. But even Crucials have had problems: I've spoken to many here who've had experience with the modules crapping out after just pushing the voltage to within 0.05V of the recommended limit. Variations in the way motherboards match "set" against "reported," miscalibration of voltage sensors -- these things might contribute to an over-volting of RAM with the user comfortable that he's being safe and careful.

But running those G.SKILLs at 200 Mhz below the rated DDR2 spec at default latencies, while setting the voltage at well below the limit or at the "Auto" default? Leaves a bad taste in my mouth. The customer reviews may have been swamped with 5-out-of-5 ratings, but I'd seen enough of the less-stellar customer reviews reporting the same sort of "infant-mortality," DOA and other results similar to mine.

But . . . . we take these risks, because, as you say . . . . "That's just part of the hobby."

Anyway, I'm impatient -- gotta do something. Once one's seen the difference between a VISTA-64 configuration of 4GB and one with 2GB, the latter of the two seems like molasses on a cold winter day in the Northwest Territories. It's a SLUG with 2GB!

EDIT: . . . and I don't run my memory at 1T anymore -- after my little "Crucial tutorial." Word to the wise, and a lesson-learned.

 

[Ratman6161]

What do you mean when you say they "went south"? From what you say, you were undervolting them (i.e. "I only turned up the RAM voltage to 1.975V on these, when the G.SKILL spec says "2.0V to 2.1V."). I would assume you at least tried running them at 2.0 - 2.1???

 

[n7]

Sounds like you're using an nForce board?

Not ideal for clocking 2x2 GB...

Other than the Crucials which seem to be using D9JKH, everything good in 2x2 GB is PSC (Powerchips).

They all clock similarly, though obviously some are much better binned.

CAS 4 is not a strong point of PSC...you'll get DDR2-900 at absolute best, but more likely DDR2-850 or less.

1T may or may not work...i wouldn't count on it.

Most of the 2x2 GB kits are best run at a ratio higher than 1:1 in the CAS 5 DDR2-1000 to 1100 (or higher) range.

Keep in mind to do decent speed, a high tRFC is needed for 2 GB DIMMs.

 

[nevbie]

Originally posted by: Ratman6161
What do you mean when you say they "went south"? From what you say, you were undervolting them (i.e. "I only turned up the RAM voltage to 1.975V on these, when the G.SKILL spec says "2.0V to 2.1V."). I would assume you at least tried running them at 2.0 - 2.1???

These G.Skills run at 800mhz/1.8V at default, which, if I understand correctly, most "overclocking" memory won't do. So he raised voltage from this default.

I hope my GSkills won't turn bad after I start tweaking, because they seem quite nice. I prefer that they default to JEDEC standard even if they are validated for higher clocks.

 

[BonzaiDuck]

Idontcare will confirm that the Striker Extreme board has downwardly biased VDIMM settings and reports lower than actual voltage in the BIOS monitor -- which in turn serves the Windows-based monitoring software. Or -- the same sensors report the values, and I haven't discovered any difference so far between, say, Everest Ultimate and the BIOS monitor -- for voltages.

The Crucials gave people trouble at the high end of their legal range, or 2.2V. Not completely sure about other 680i boards, although I seem to recall similar reports -- AigoMorla was running a eVGA 680i and burned a few Crucials. IdontCare burned a few sets . . . too . . . and I burned a 2x1GB DDR2-1000 set of Ballistix.

I didn't undervolt the G.SKILLs, or at least I was bumping up the voltage on them from an initial fixed setting of 1.925V. But the Striker would report 1.96 at that setting. if the report itself was biased, they were already running close to 2.0V, but I haven't the means to confirm it. IdontCare seems to think the cumulative bias added up to 0.10V.

I was focusing on the Small-FFT test, and VDIMM increases were confirmed effective more than once at making the system stable. But then, the Blend-Test precipitated the BSOD. If it was just a matter of undervolting, I'd simply get an error in one of the cores stopping that particular thread, and I would've bumped up the DIMM voltage in BIOS setup and continued testing.

Crucial seemed to be the last to start producing 2GB modules, and if you check the paltry few customer reviews they've had at NewEgg -- they are not stellar. With their 2x1GB kits, I was pleased, because they were just right for the combination of mobo and CPU at over-clocks of 20 to 35%. Right now, on a similar system (same mobo with a Q6600), I'm running a set of Tracer 800's at 875 and tighter latency settings than spec'd for 800 -- 4,4,4,10 as opposed to 4,4,4,12. Takes a licking, keeps on ticking.

I sent the RMA request to G.SKILL, with the message that "I want a factory-new replacement kit." My Crucial RMA experience led me to suspect that Crucial was receiving oodles and gobs of RMA modules -- testing them -- binning the good ones and chucking the bad ones, then repackaging the good modules for RMA replacement. There is no other explanation as to why the replacement set I got would not run at 1T. My point is this: even if wisdom says not to run at CMD=1T, if every single kit of Crucials I'd had brand-new was capable of it, then so should the RMA replacement kit.

I've agonized over putting a spare set of Ballistix 2x512MB in the box with the temporary Tracer 800's that are in there now -- I might try it, but the G.SKILLs are coming back, and I decided to try something else. I ordered a set of Corsair CAS=4 (4,4,4,12) 800's, which Corsair matched to this board under the "Enthusiast" category. I'm not sure, though, if I can push them much past 800 as I had with the 2GB kit of Tracers in this system I'm accessing to post this reply.

Back to the G.SKILLs. If the "Auto" setting produced clean MEMTEST86+ results showing a reported voltage of 1.875V, and if I twisted up the VDIMM to get it stable in small FFTs, -- AND . . . . if, as I say, the Striker board reports lower-than-real voltages, I don't think that I missed anything. Like I said -- if it wasn't a hardware failure, a voltage too-low would just cause an error in one of the core screens of Blend Test -- not a BSOD heart-stopper.