ASUS P5W DH - Problems + Fixes Thread

[Falkentyne]

Tomcug (or anyone): has there been any news?
It's feeling awfully lonely here...

Well I might as well liven things up by a few things i've observed.

First, Gskill 2x2 GB "Pi" black PC-7200 dimms have no problem in this board, and you can run 1:1 manual timings past 370 FSB. The quad core limitations are holding the memory back. In fact, with the 4:5 (or the other divider), I was able to reach full RAM spec of 450 Mhz with this memory, timings 4-4-4-12, and 2.1 volts (2.0 did not POST), but this required either 1.85 vmch or 1.75 vmch+1.30 vFSB. I was at 338 FSB to reach 450 mhz on the RAM so i forgot what divider thats called; it was second to the one under the 1:1. i know one is 4:5 but what was the other ? So it's good to see that this board can run the RAM at that frequency of 900 mhz. And it's 2 GB dimms too, which is good for people who are memory starved. So if you have a high multiplier chip or unlocked quad core, you're still good to go for some high mhz on a 1.04g board. Too bad it wont get there 1:1 on the quad (i wonder if TRD and TRFC might help things slightly?)

My QX9650 is stable with vdroop mod, at 4 ghz (333 x 12) @ 1.3625vcore (1.344 idle, 1.350 load), or at 4.09 ghz (364 x 11), and vmch 1.55, vfsb=1.20.

Early on in this thread, someone took a multimeter to the Mch and determined that raising the FSB termination voltage affects the vmch voltage directly! It wasn't a fixed value, but it was something like every +.1 vfsb caused the measured vmch value to increase by about 0.8. I cant find the post right now but it was something like 1.60vmch +1.30 vfsb came out to 1.68 vmch measured, and 1.60mch+1.40 vfsb came out to 1.74 vmch.

I've actually confirmed this in a FSB test in how the board is acting with my qx9650. Note I'm using the 1.04g board, serial# 71xxxxx and g0aaY (showing it was made january 2007)

First, absolute max FSB i can use with my QX9650 is about 380 FSB. 380 FSB require 1.85v vmch to pass loop memtest86+ test 5. BUT, i got the same stability by using 1.75v on the vmch WITH increasing the vFSB to 1.30. This seems to support that the vFSB is somehow giving the mch more voltage for some reason.

370 fsb was fine with 1.65v and 364 FSB had zero problems with the lowest 1.55v value.

Note that all these settings were with the RAM running 1:1 with timings 4-4-4-12. Gskill hz D9's (2x1gb) ran as expected. My corsair XMS2's required a volt bump to 2.1 or 2.15v at 380 FSB. The 2x2GB gskills didn't.

390 FSB was not usable. 1.85 vmch with 1.20 (stock) vFSB crashed memtest86+ almost instantly. I mean crashed. 1.30 vFSB (which as above, im sure raised the vmch above 1.90) caused constant errors in test #5 at almost each address.
1.40 vFSB did a LOT better, with only 3-4 errors per pass on test #5. I fear of how much voltage, without a multimeter, the northbridge is getting here. With 1.50 vfsb, either 390FSB might work, might degrade the QX9650 or cause the entire northbridge to blow up or go into overvolt protection mode, since it would be getting over 2v, if it's following the pattern from the other person's multimeter measurements.

It goes without saying that 400 FSB didn't post. I think I remember it posting with my X6800 but i forgot. If I recall, the "POST" multiplier feature, where the board tries to post at the default multiplier, before reading and setting the requested multiplier, was causing failure to post here, because the board was trying to POST at 400x11 and then switch back to the x9 multiplier. I obviously have no such problem with the high multiplier of 12 on the QX9650, instead it's just the board (GTL signals? reference point?) cant handle the high FSB with a quad, but 380 FSB is still quite nice for such an old board.

I think the comment about the vmch increasing its voltage with a percentage of the vFSB being added may be interesting for some people.

So anyway, it's not the CPU mhz of quads that the board cant handle, it's the FSB limitations by lack of the other two gtl reference dividers that limit quad core overclocking at high FSB. Unlocked CPU's help circumvent this, and 1.04 version (with rectangular shaped northbridge MCH, newer 975X chipset like in the Intel badaxe 2) can overclock FSB on quads more.

It is still unclear whether the 1.04g version has the two missing GTL reference points that freecableguy mentioned that he couldn't find in the board he tested, although he didn't say if that was a 1.02 or 1.04 board.

But its still good news to know that using 2x2GB dimms isnt hurting the FSB overclock any more than the 2x1GB gskills i was using--the Quad is limiting the FSB. I just wonder how 2x4GB would work

I hope this thread doesn't die.....

*edit*
switched tuniq tower 120 for a TRUE Black 120+NBT Panaflo medium output fan, for my QX9650, and not only did my temps DECREASE at least 10C, but 1.3625v (1.52v real) is once again quad prime stable at 333x12 (4 ghz) thanks to the huge temp reduction. Tuniq Tower cant handle overclocked quads...