clarknova

Hi,

I'm new to OCing and new to the forums, but trying to learn the ropes In this post I'm attempting to adhere to the terminology and principles outlined in durch's "DDR2 Memory Speeds Explained".

Before playing with the RAM or any voltages I'm trying to find my max stable CPU reference clock. Stock is 200, I'm up to 269 and the system appears to run stable. When I run prime95's torture test, I get errors, but only under certain circumstances. This is what I need help unerstanding.

I only get errors under 2 conditions:

1. The CPU must be running at full speed. I can manually adjust my cpu in software to run it's multiplier at 10.5, 10, 9, or 5x. I only get errors at 10.5 (100%)

2. The torture test must be number 2 or 3. In other words, there must be RAM involved, if I have read the test descriptions correctly.

Here's where I get confused. My RAM speed in the BIOS is set to DDR2-400 to be safe, although I'm using DDR2-800. cpu-z confirms that my RAM divider is lockstep with the cpu multiplier, i.e., when the cpu reference clock is set to 269, the RAM will show 269MHz in cpu-z, except when the cpu is running at 100%. In this case, the cpu multi is 10.5 and the RAM divider is 11 according to cpu-z, resulting in a RAM speed of ~257.

So if my RAM is running below stock speed (400MHz), then why do I get prime95 errors only on torture tests 2 and 3?

I think I'll have to run memtest86+ to find out if it's really a RAM error. I'll also try boosting my RAM voltage to see if the errors disappear. Anybody seen this before? Any insights?

Hardware:
Athlon X2 4050E (2.1GHz stock: 200*10.5)
ASUS M2A-VM
2*KVR800D2E5K2/2G= 4*1GB ECC running in non-ecc mode (1.8V)
Gigabyte 8600GT pciex16
Intel Pro 1000PT pciex1
Scythe Shuriken low profile HSF

Software:
Ubuntu 8.04 x86_64
Vista Ultimate x64
Server 2008 x64
prime95v25.6 (linux & windows)
cpu-z

BIOS settings:
vcore: auto (reports as 1.02 to 1.25 depending on cpu scaling)
cpu clock: 269
cpu multi: locked (10.5)
+1.2VNBSB: auto (adjustable 1.2-1.5)
RAM: DDR2-400 (adjustable 400-533-667-800)
vdimm: auto (adjustable 1.8-?)
pci delay: 64 cycles (default)
HT: 600MHz (reports as 807 (269/200*600))

db

clarknova

Sorry to cry wolf. It did eventually show errors on torture test 1, it just takes longer to do so. I guess I can't blame the RAM.

db

NrGx

Torture test 2 and 3 also do stress the CPU, it just stresses it differently by using different sizes of FFTs.

__________________
Q9550 @ 4.0GHz

ddlyspdr

if u get any errors (to my understanding) the system is unstable. try getting OCCT. keep the multi at stock, then up the FSB when u get an error, try either doing something with the ram(lower the divider), up the ram voltage, or cpu voltage.

__________________
-<[{DdlySpdr}]>-