Raised multiplier woes...

[3.0charlie]

I'm stumped. Rig in sig, but not at 4.0GHz (400 FSB) anymore... It ran without a hick-up at 4.0, until I changed the PSU for a Corsair HX1000W (from a Corsair HX620W). It ran for 2 days, then I shut it down for maintenance - adding a second HD4870. Re-started it, and the Monster came to Life... the BIOS POSTs, but when it reaches the DMI pool data, it reboots in a loop.

Here goes:

10x 400 is a no-go, whatever volts I pump in it - VCore, VFsb, VMch.
9.5x 400 is a no-go, whatever volts I pump in it - VCore, VFsb, VMch.
9x 400 is a no-go, whatever volts I pump in it - VCore, VFsb, VMch.
8.5x 400 is stable OCCT 1 hour, BOINC folding.

The lowest stable FSB for a 9x multi is a mere 333MHz, whatever volts I pump in it - VCore, VFsb, VMch. Ram is out of the equation with settings of 5-5-5-15 2N 1.9V @ 820MHz.

Now, check this out:

8.5x 400 is stable OCCT 1 hour, BOINC folding.
8.5x 440 is stable OCCT 1 hour, BOINC folding.
8.5x 450 is stable OCCT 1 hour, BOINC folding.
8.5x 460 is being tested - OCCT 9 minutes to go, BOINC will be done overnight.

A FSB of 460MHz that is looking to be very promising, with a Quad AND 2x 2Gb sticks?

What is preventing the board to POST higher than 9x 333MHz, but it will run happily at 8.5x 460MHz?

[magictorch]

Annoying problem!

Just out of curiosity have you tried switching the old psu back and running one 4870 just to see if the cpu can run as before.

Also just running 1 4870 with the new psu to rule out any weird PCI-E instability problem in crossfire. Maybe a bios fix could fix that.

Also maybe try another crossfire capable psu like a silencer quad to rule out anything else.

Just thought id throw in some random stabs.

[3.0charlie]

Yes, very annoying... especially if 460FSB is stable! We'll see tomorrow.

I'll test that theory tomorrow night, removing 1 4870 and re-installing the HX620W (I just edited my initial thread - I forgot about that one...). And I'll try the OCZ 600W too, just for the fun of it. No BIOS fix possible, I've tried both that are available at Gigabyte's site (F6 and F7b).

Unfortunately, I don't have access to any other PSU.

[MpG]

Damn, that's weird.

This is entirely a stab in the dark (and I'm not familiar with exact Gigabyte BIOS options), but is there any possibility that some MCH latencies or GTL settings might be responsible? You're running two GPUs, so there's definitely got to be more traffic crossing the northbridge now. Quad-cores are more susceptible to issues involving the synchronization between the memory and the processor, and with a few settings, too much and too little can be equally problematic.

That aside, I'd be interested in seeing if the second card is what's giving you grief.

[3.0charlie]

Well, 460 FSB is stable with a Quad and 2x 2Gb sticks... that X48 board is amazing. I had a hard time reaching 400 with my old Q6600 and 2x 1Gb sticks on my old P35C-DS3R... I'll perform some more tests tonight; your explanation makes sense MpG though. It'll be rather interesting.

If the second card is the culprit, which settings could I change to allow a multi of 9x? The BIOS does not have too many options regarding latencies and GTL settings...

[MpG]

Again, I'm just guessing, but my own board/chip combination is a complete bastard over 440FSB. I've found that the processor GTL setting needs to be 0.61X in order to keep things happy -down to 0.59X or up 0.63X are both bad, and the former would sometimes give me a no-post at higher frequencies. I'm still banging my head over the NB GTL settings, but I suspect they're equally important. Do you have any sort of "clock skew" or "performance boost" option that's set too aggressive?

I'd be trying the video cards first. Much simpler and more obvious. Maybe a bad card? Pop one in at a time and see if things are okay for both. You haven't overclocked the PCI-E bus?

[3.0charlie]

But your explanation makes sense. Problem is that this X48 board does not have too many adjustments regarding GTLs. I remember a clock skew setting, but it is voltage-related. PCI-e bus is overclocked to 110MHz. I'll try a default setup with one card, at 100MHz.

[MpG]

Since both the board and the cards are PCI-E 2.0 compliant, I'd be pretty amazed if any further boost was even needed there. And a pair of full x16 slots running at that speed... who knows?

[3.0charlie]

Well, +100 smackaroos for MpG. You hit the nail squarely. I was bored at work, and decided to learn more about the skew control of the DQ6 (AnandTech: The Gigabyte GA-X48T-DQ6 - Redefining the High End?). Low and behold, I'm typing this at 10x 420MHz (4.2GHz!), with Crossfire enabled!

The differences? A simple MCH clock skew control adjustment of 50ps (pico second) and a clock-down of the PCI-e bus overclock back to 100MHz, from 110MHz. That's it.

Thank you very much to both MpG and magictorch. I knew something was wrong, but couldn't put my finger on it...

[BALISTX]

John,

Good to see you got it figured out and nice to see such a beautiful overclock with Crossfire enabled.