Is a rounding error in Prime95 memory or CPU?

[yacoub]

Need to know if CPU needs more volts or backing off clock a bit, or memory needs more volts or settings tweaked.

Memory is not overclocked really at all (204.5MHz (245Mhz * .83 due to use of 333Mhz divider instead of native 400MHz setting). It ran 220Mhz earlier with 400MHz bus, so I'm confused as to why this would cause an error but that setting wouldn't.

It runs 3dMark2006 without a problem, FEAR Test Settings w/o problem, but Prime95 the middle Torture Test option errors with a rounding error around 3 minutes in, and the third option errors in about a minute. The first option didn't error for at least 10 minutes before I stopped the test. The options increase in memory intensity so I'm wondering if it's a memory issue, but they also increase in heat and voltage use so I'm hoping I just need to turn up the volts to the CPU a little or something. =/

 

[yacoub]

Well 2500MHz (even at the max voltage of 1.55 or whatever it is in the BIOS) will not even get fully through POST so I'm definitely near the max of the cpu core for bus speed at least at the available voltage and with the stock hsf.

I knocked it back to 2400MHz (even though 2450MHz was stable through all testing except Prime95) and am running it at the second highest voltage offered (1.5375 IIRC) to see if it can pass Prime95. I will also try Auto voltage again after this if this errors out. I will also try Memtest86+ again to make sure it's not a memory issue.

I'll give up 50MHz for greater stability. Besides, 2.4GHz is the speed of a 3800+ so that's pretty sweet to me.

This is with a 4x HTF multiplier (so my HyperTransport is a little slower at 960Mhz) and RAM is 333Mhz divider which puts it right back at 200Mhz (CPU/12). Stock 3200+ was CPU/10 (2000/10=200MHz) though that was at 400Mhz (no divider). Hopefully that means the RAM will be extra stable and I've gained 400MHz on the FSB/CPU while the RAM's performance I assume isn't any worse off? Either way, my 3DMark2006 score is up about 50 points and the FEAR performance test is up a few fps as well as a higher percentage "above 40 fps". =)

 

[saltedeggman]

1.55 volts seems a bit too much

what are your temps?

 

[yacoub]

Under max load right now in Prime95, 46C. The fan is at 100% throttle via SpeedFan. It's also been almost ten minutes and no error (yay!).
When idling it's down around 37C.

You say it seems like a bit too much but when I had the voltage on Auto in the BIOS it was reading 1.536v (with fluxes to 1.520) in CPU-Z. It is currently reading 1.568 (with fluxes to 1.584). I guess this shows that the Asus BIOS is actually giving a bit extra voltage than I set it at? Either way, apparently the added voltage and slight drop in MHz on the bus has alleviated the Prime95 error (so far). It's been 15 minutes now and that's already five times as long without an error.


Btw, here's what I got when I did 2450Mhz:
http://users.adelphia.net/~jrockcls/2450.gif
stable in everything except Prime95.
Now at 2400MHz and even Prime is stable so far. Sweet.


final update for the night: 36mins stable Prime95 testing before I stopped it to shut her down for the night. a little extra voltage and a little less FSB cured whatever was just barely flakey at 2450MHz.

 

[compgeek89]

I'd say thats too many volts for a plain old Athlon 64.. opteron it'd be aight but... I dunno man. Seems way too high for a measly 400mhz OC.

 

[yacoub]

What voltage should I be aiming for? Remember, even on Auto it was around 1..536v.)

I've lowered the voltage to 1.525 and am Prime'ing it again. Good so far. (Btw CPU-Z reports the voltage as 1.552 fluxing to 1.568).

update: I am trying it at 1.500v. Voltage shows in CPU-Z as 1.536v fluxing to 1.520v.

Btw, have I lost a lot of RAM performance by going from 400MHz to a 333MHz divider, or being just a divider am I not actually losing any performance since the FSB speed it runs at is still 200MHz (x2 DDR and still dual channel)?

 

[imported_Sincity]

By using the divider, your RAM speed is still same (or close to it 200 vs 204). What mobo are you using that is causing those power fluctuations? Anything above 1.5v is kinda high, though the overclocking FAQ tells you to set it that high. Most people have their 3000+ and 3200+ OC'd using stock voltages. I am at 1.45v running 2.5ghz using a 3200+ E6. Memory @ 2.7v. Prime stable for 12 hours using the CPU stress option.

 

[yacoub]

Is it Prime stable using the second and third test options as well? Those are the ones I'm failing on more frequently. The first one generally doesn't fail, but the second fails and the third fails most quickly, when I had it at 2450MHz. At 2400MHz I'm not failing (yet).

I'll try lower voltage further. Also I might try to up the DRAM voltage (although I don't believe this board (at least not this 1008 BIOS) gives me much range to play with there.)

In CPU-Z mine says "Family: F, Ext. Family:F, Model: F, Ext. Model: 2F, Revision DH-E3"
Is that a good or bad core for OC'ing?

Assuming your "E6" is where mine says "DH-E3", right?



Update- running 1.45v and seems good so far. Guess it's not CPU voltage but RAM that's the issue. The RAM simply doesn't like me when I try to go over 240MHz bus even with the .83 divider (which is strange because I did no divider (400MHz) with a 220MHz FSB without a problem, yet 240Mhz FSB through a 333MHz .83 divider is 204.5MHz FSB for the RAM yet it can't stay stable there. huh?!

Also I'm not sure what safe voltage is for my RAM (Corsair Value Select) so I just leave it on Auto (I tried 2.7v and 2.8v and neither helped it get through a full POST at 250MHz (208.75MHz with the 333MHz .83 divider)).
I also even tried relaxing timings a bit from 2.5-3-3-8 to 3-3-3-8 and 2.5-4-4-8 as well as 1T to 2T. Nothing helped. Must be RAM voltage then? Dunno.

 

[yacoub]

ok that went well, so trying out 1.40v. (reported as 1.408 with fluxes to 1.392 in CPU-Z).

 

[imported_Sincity]

E3 are known to be better overclockers. Did you mention what motherboard you are using?

 

[yacoub]

A8N-SLI Premium, v1008 BIOS. I might have forgotten to mention it in this thread but it's also in my sig. =)
It's very happy at 2400MHz. Idles at 36C, stock hsf, fan at 40% speed via SpeedFan.