Strange problem I need help with

[Megatomic]

My setup:
Abit NF7-S Rev. 2.0 with BIOS 25D
AMD AthlonXP 2500+ 0313
2 x 512MB of Corsair Value Select PC3200
Windows XP Pro SP1a

Here's my problem: I typically run my system at 3200+ with my timings at 11-3-3-2.5 and the ratio at 6:6 and all is well. If I go to 11.5 x 200 I lose memory memory bandwidth in the Memory Read test of AIDA32 to the tune of 300MB/sec. My Memory Write score goes up by about 100MB/sec. If I overclock by FSB all my scores go up correspondingly...

Why on earth does this happen? What BIOS setting is causing this to happen? What other details do you all need to help figure this out? Thanks for any assistance guys.

 

[o1die]

You may need to increase your memory voltage to 2.6 or 2.7. Sometimes this helps stability; it all depends on the memory chipset or brand. You can also try raising your cpu voltage.

 

[Megatomic]

My CPU is stable up to 2.3GHz at stock voltage (passes Prime95 easily for many hours) and my Vdimm is already at 2.7V. I can try to up the Vcore and see if the bandwidth problem goes away though. Thanks for the advice.

 

[DAPUNISHER]

Bench a few games and/or apps you use to see if it shows up as a real world difference, otherwise I wouldn't pay it any heed. It seems odd that the multi change impacts bandwidth, but there are so many oddities with nForce2 that it isn't too hard to believe

 

[ectx]

Originally posted by: DAPUNISHER
Bench a few games and/or apps you use to see if it shows up as a real world difference, otherwise I wouldn't pay it any heed. It seems odd that the multi change impacts bandwidth, but there are so many oddities with nForce2 that it isn't too hard to believe

Like he said. It is odd but you never know ...

Definitely try some other bechmarks

 

[Megatomic]

I'll try but I had been using AIDA32 for my memory benchmarks as it has been far more reliable than SiSuck. What would be a good memory dependent benchmark to use?

 

[DAPUNISHER]

Originally posted by: Megatomic
I'll try but I had been using AIDA32 for my memory benchmarks as it has been far more reliable than SiSuck. What would be a good memory dependent benchmark to use?

That's kina my point brudda, if the slightly lower bandwidth doesn't show up as a crutch in the games/apps that matter to you then F it It would be great if someone knew the technical reason why this is happening or if it's just a bug with nF2 and AIDA or what?

 

[Megatomic]

I understand what you're saying. I'm going to try some things and see what happens. Unfortunately, SiSuck reported the same phenomenon so it's not just AIDA freaking out...

 

[Megatomic]

Well, I did find one useful thing out while I was tweaking my BIOS to test out this issue: my system is stable at 1.625Vcore while the CPU is set at 3200+. I can overclock my CPU by about 400MHz and undervolt it by .025V at the same time. I'm gonna Prime95 it overnight to see how that goes.

I can say this already: it was Doom 3 stable for 1.5 hours.

 

[Megatomic]

Prime95 Blended Test stable for 8hrs and 1min. I stopped it as I need the computer for work. I guess it is stable enough for me. I'll stop whining about losing bandwidth at 11.5 now that I can undervolt my chip.

 

[ViPerZX]

I don't know if its true or not, but I've heard that you will lose memory speed by using 1/2 multipliers.