Need help circumventing 1.6A OC problem :UPDATE:

[brinstar117]

I finally bit the 1.6A bug and upgraded my main rig.

I purchased:

Abit TH7-II RAID
(2) 256MB PC800 Samsung RDRAM
Pentium 4 1.6A retail

The rest of my specs are as follows:

Cooler Master ATC-110 Aluminum case
Enermax 431 Watt Power Supply
Radeon 8500 (retail)
Philips Acoustic Edge
Adaptec 29160LP U160 SCSI controller
Fujitsu 18 GB @ 10,000 RPM (MAN3184MP)
3COM 10/100 EtherLink (3C905C-TX)

I'm pretty sure I have a CPU from a good batch.

FPO/Batch# H202035W
S Code: SL688
Plant: Malaysia

I don't know how to tell what week I have. The pack date indicates 1/12/02 but I'm not sure if that's relevant.

Anyways, people on Overclockers.com with batches approximate to mine have achieved the highest overclocks so far. (1.6@2.5-2.6 with stock cooling)


EDIT: UPDATE

As it turns out, I flashed my BIOS with the same version but from the Abit site (-38). Now the fix option works perfectly. I'm able to get up to 140Mhz FSB @ default voltage. I'm able to get higher by bumping the voltage but it's not worth it IMHO.

Now a new problem has surfaced. Before my BIOS was flashed, I couldn't run the RAM with the 4X multiplier @ 133FSB. The system wouldn't even boot. I thought I had a lesser RAM Clock Generator. Now, I'm able to boot into the OS with a 4X mulitiplier and a 133FSB but it's not stable. I dropped down to a 125Mhz FSB and everything appears stable (I'm running Prime 95 now, and I'll try it overnight).

Before, with the old BIOS I couldn't even get past 112FSB with a 4X multiplier. Now I'm wondering if my OC'ed memory is limited by the RAM itself or the memory clock generator. Is the only way to test this is by getting verified ram that is known to OC to 133FSB X 4?

Here are the markings on my RDRAM:

Samsung
Korea 0203
256MB/8
MR16R1628AF0-CK8
800-45 011 (last three numbers in small print)

These are single sided RIMMS as well

 

[mschell]

I don't know how Abit is doing the PCI divisors but most boards drop the PCI back to 33MHz at 133MHz FBS (1/4 divisor). It generally drops again at around 150MHz (1/5) but some boards allow you to choose otherwise. To find the PCI MHz, divide the divisor into the FBS.
Keep in mind that from the Northwoods 100HHz default FBS to around 118MHz the 1/3 ratio is overclocking the PCI bus. If your card is indeed sensitive to an OC'ed PCI bus then it would fail at those speeds as well. If it doesen't then the RAM or CPU just can't handle the speed - don't know much about the TH7 and Rambus though.

 

[ginfest]

Not sure if you did this from your post but,
Go into BIOS/SoftmenuIII/ and look for "AGPCLK/CPUCLK". There should be 3 settings "2/3, 2/4, fix". At 133FSB the 2/4 setting will give you 133/33/66 FSB/PCI/AGP. For higher FSBs the "fix" should hold the PCI/AGP at 33/66.

I believe "fix was broken in any BIOS earlier than "38". Also, I seem to recall that default on mine was set to 2/3, maybe that's whats causing your problem?

I'm running my 1.6A/Th7-II / 2 Samsung 256MB(DS-16device) at 133 FSB, 2/4, Rambus at "400" which at 133FSB=533/PC1066.

 

[Duvie]

I agree with the above...I thought I heard the abit boards came with a pci lock..at least the i845 board I know anandtech reviewed did...

 

[brinstar117]

I checked my BIOS and it's the most recent version, mine is 38. I have PCI lock with the "fix" option. However it doesn't seem to work.

When the FSB is at 133 I have to use the 2/4 divider to make it work. The "fix" option doesn't seem to be functioning.

Also, for RAM settings there is "Differential Current" is that anything I need to worry about?

 

[JameyF]

heh heh...you said circumvent

If you're not a SNL fan....sorry, I couldn't resist.

 

[KenAF]

You have an old bios. You need the 77 bios from www.hardforum.com. It has a "fix options" PCI and AGP bus speeds. The newest bios is not on Abit's site.

 

[brinstar117]

Thanks for your help guys, here's one more issue to tackle. Hopefully soon I'll reach my goal of a perfect OC'ed system that's stable (well, stable enough for me not to notice any problems)

 

[Clevor]

Another possible problem is you have the 8-device 256 MB Samsung PC800, which some people feel does not overclock well. It is single-sided, but uses denser chips. It seems the 16-device 128 MB double-sided chips might do better, although both should get to 133.

Plus no telling how well the latest ram does, and yours was made in Feb. 02. Maybe they are already binning PC1066 chips, if they can't get the .13 micron (or whatever) tooling up by the May launch, e.g. PC800 chips selling now are ones that don't make the grade.