memory or memory controller degraded?

[Autochthon]

I have some readings in everest I just can't explain. Over a 2 day period of stress testing my average everest benchmarks have dropped from ~6750 to ~6550 for read , 3000 to ~ 2800 (and less) for write and ~38+nanosec to ~39+nanosec for latency. I had made no changes in setting over this period and rebooting hasn't apparently made any differences. I don't see any new software apps or services running but oddly enough my cpu usage seems to have dropped at idle from 10% to maybe 3%. The stress testing was done with 3.2 V on mem, 1.54 to 1.57V on CPU and 1.6V on chipset (all based on bios readings). Do you think my memory or memory controller could be breaking down? Memtest hasn't seemed to change but CPU doesn't seem quite as stable as it had been :(

[soulsaver_8229]

Maybe try upping the volt on the cpu a bit to see if that helps......if not, then all i can tell you is maybe you have some spyware or virus or something boging your system down,.......you are overclocking and maybe your straining something that isnt used to be overclocked /shrug/


soulsaver_8229

[(cf)Eclipse]

"doesn't seem?" ???

run prime95, and do lots of memtesting.

at first glance, it seems like a virus/spyware problem

[Autochthon]

"doesn't seem?" ???

run prime95, and do lots of memtesting.

at first glance, it seems like a virus/spyware problem

I say "doesn't seem" because I can run memtets for 8 hrs with no issues @ 262htt but prime95 errored after 8 hrs using blend @ 260htt and 1.275x123% where as previously it had run for 18hr's at that voltage and htt (possibly even 261htt) with no problems, which may or may not mean anything.

I tried norton and AVG and they showed clear of virus etc. I'll look for spyware or maybe try NOD32 which always worked well for me.

edit: just scanned for spyware showed nothing

[(cf)Eclipse]

how does small fft's do?

[Autochthon]

how does small fft's do?

I'll bump htt back up to 260 or 261 and try it a few hours everest writes get flaky about there and bsod's me at 1.54V. I'm uncertain if that's related to the speed reduction. It may be.

[(cf)Eclipse]

it almost sounds like a power issue

[Autochthon]

I doubt it's power yesterday all my rails looked rock steady and good on my dmm (I don't have access to a scope) and I'm not really pushing this 600watt supply as my fans are running off a seperate supply.

it's been about an hour now on small ftt at 261 htt all looks good 1 run of everest writes has been bsoding me at this speed and voltage. I'll bump up a little more.

[(cf)Eclipse]

hmm, interesting. so the cpu is definitly stable.. it's either the ram or memory controller.

do you have issues if you try one stick at a time?

[Autochthon]

hmm, interesting. so the cpu is definitly stable.. it's either the ram or memory controller.

do you have issues if you try one stick at a time?

yeah running at 263 now and seems happy as a clam.

I'll try the single sticks. You have any thoughts as to which memory slot to use? I've been getting stable memtests up to 265 9/10 with 2 sticks in orange. Earlier I had thought that possibly it was my 4x link was limiting memory speeds but 3x didn't improve anything.

[(cf)Eclipse]

use slots 1+3 on the dfi (orange), you can try yellow if you want though

also, with one stick, i'd just put it in slot 1


also.. the ram dividers are very weird on dfi boards. :(

[Autochthon]

well hell I'm going backwards.

first 1 and 3 are yellow for me 2 and 4 orange but know matter the only place that 1 stick would boot from was orange farthest from cpu ( 2 for me).

Single sticks have much lower read a writes and the same latencies. First stick BSODed everest write at the usual ~260 second stick BSODed at my standard stable speed (259) :( put both sticks back now I get BSOD in everest at 259 which used to be stable. Upped voltage to 3.3V still BSOD at 259 :( :(. tried yellow slots wouldn't complete post.

btw: both sticks bh5

Man what a mess.

edit: doesn't seem like it's memory to me

[(cf)Eclipse]

wow, that's odd.

try a new bios. there's a nice list of them here:
http://www.dfi-street.com/forum/showthread.php?t=6369
i recommend 5-10-1 or -3, it's down a wee bit

also

Yellow slots 1 and 3 are supposed better suited for HI voltage and therefore UTT/BH-5 types of memory.
Orange slots 2 and 4 are supposed better suited for normal lower voltages and therefore TCCD/TCCC/TCC5 types of memory.

[Autochthon]

I'd used 510-3 for about a week before switching to 510-2 . 510-2 seemed to be about 15% better for my memory ( but that's quickly disappearing).

Guess I could try 510-1 or that special 510-2 mod.

[(cf)Eclipse]

ah, hmm... this is very weird. do you have another cpu to test.. or ram?

[Autochthon]

all my other computers are ancient except my laptop and it's a ddr2 pentium ;)

Prime95 did crash at 263, after 15 min or so, but that was after over 2 hr's of gradually bumping it up.

I'm at a bit of a loss where the problem is. I could try bumping chipset voltage or cpu voltage and see what differences it makes.

edit: ok I bumped chipset voltage up to 1.6V and I'm not crashing everest writes every time now. My write read scores still suck though so I still haven't figured out what's happening.

edit2: I've decided the most reasonable answer is that my OS is using more of my memory for some reason. I suppose that could effect the bandwidth/latency scores

[(cf)Eclipse]

hmm, try bumping up LDT voltage too, just out of curiosity on my part :D

and can you make a list of all the timings you're using in the bios, or provide a screen of a64tweaker? :cool:

[Autochthon]

ok took some time but I got image up. 10 years of internet and had never done that before.
System was finicky last night took 3.3V and loosened timing to get memtest stable at 260fsb

would not boot at these timings in yellow slots come hell or high water.
I bumped LDT 0.1V but don't think it mattered seemed fine at stock 1.2 also

http://img138.echo.cx/img138/8517/test4zs.jpg

[Autochthon]

one last try.

I was able to run prime blend for 18 hr's at 261mhz but one pass of everest write benchmark and I BSOD it is really bothering me any idea's on what everest is doing to cause this? uping cpu or mem voltage hasn't seemed to help much. Is it some problem with my memory controller?

[(cf)Eclipse]

have you run memtest? i find it's much more prone to finding memory errors than prime blend :D

and.. bumping up voltage one more notch may work, but 3.3v is kinda my limit for 24/7 on high voltage ram ;)

[Project_2501]

Hey (cf)Eclipse, you do realize that the yellow slots run better with UTT ram, and that the -2s are better with UTT ram? :D You reccomended all the opposites, lol. I learned all this while messing with my Redline. Which btw is now running marvelously, but I'm held back by a stock CPU cooler and the unwanting to raise the voltage w/ the stock cooler. :) I also find that the 5v memory rail is more stable up around 3.1 - 3.2v than the 3.3v rail... Maybe because it's not so close that way, the total, and the available.

I just realized that my 3.3 rail running at 3.23 would be my problem with 3.2v not running right...

[Autochthon]

yeah ran memtest. It's stable in memtest at higher MHz than Prime :(

[Autochthon]

Like I mentioned I found the opposite to be true with my BH5 wouldn't boot on yellow and rock stable in orange. Go figure. I upped my 3.3V line to a bit over 3.4 to try an make certain I was getting enough for 3.2V on the memory.

[(cf)Eclipse]

Hey (cf)Eclipse, you do realize that the yellow slots run better with UTT ram, and that the -2s are better with UTT ram? :D You reccomended all the opposites, lol. I learned all this while messing with my Redline.
oops! my bad, thanks for the correction :D

though, i do know this:
-1 = general
-2 = utt/bh/ch (maybe micron?)
-3 = tccd

[Autochthon]

Well I think I ruled out memory as the problem I just lowered divider to 5/6 ratio and still got the everest error at exactly the same FSB speed. I guess that leaves the memory controller. Some wise soul told me this two weeks ago but I was too stubborn to accept it. ;)

[(cf)Eclipse]

i guess it is then.. how does 2T (CPC disabled) do?

[Project_2501]

oops! my bad, thanks for the correction :D

though, i do know this:
-1 = general
-2 = utt/bh/ch (maybe micron?)
-3 = tccd

Oh, I was pretty sure that the -3 was for general ram and -1 was for TCCD. :confused: Oh well, that could be why my OCZ didn't do so hot, wrong bios optimizations.

[Autochthon]

i guess it is then.. how does 2T (CPC disabled) do?

I tried CPC disabled and loose timings earlier to try and move to 1/1 ratio but it wasn't enough.

I'm trying the new 6/18-2 bios and it may have helped a tiny bit. I was able to lower mem voltage to 3.1V @ 260MHz FSB 220MHz memory and only got 6 errors (test 5) in memtest in 8 hours. So 3.1V may be doable. But for practicle purposes I seem to have reached the limits with this gear on this CPU/memory. More voltage to the cpu seems to add heat but allow only very small increases and the memory just won't do 1/1 without lowering the FSB 20 or 25 MHz and loosing timings.

Something bad happened to the CPU after that long prime95 run a few days back as Superpi and everest scores both dropped noticeably but it still seems like a decent chip 650MHz over stock at 1.55V and 550MHz over stock at 1.4V is pretty decent.

Now I'm waiting on a dualcore as this was just supposed to be a temporary fix anyhow :D

[Project_2501]

I found Hellfire's 510-2s to be better than the 615s that I tried. But those weren't optimized...

[Autochthon]

I found Hellfire's 510-2s to be better than the 615s that I tried. But those weren't optimized...

I'd been using Oskar's standard 510 and it seemed pretty decent. The voltage improvement for 618 may be my imagination. Bu tit seems no worse than 510 at least.