I found an interesting workaround for this problem.
This method will let you run the PC at a slightly lower FSB during normal use (and during suspend/resume), yet your FSB will automatically (and predictably) increase under load. So you can effectively overclock past 266 mhz FSB (or wherever your S3 suspend/resume wall is) and then get S3 working again.
Here's what you do:
1. Disable C1E/EIST.
2. Find a target o/c.
3. Find voltage, timings, etc. that are stable at this speed.
4. Lower FSB by a certain amount (i.e. 64 mhz?).
5. Enable C.I.A. 2 auto-overclock, which will bump the FSB up under load.
For example, here are the C.I.A. 2 levels on my board:
-- 266 FSB --( +20 )--> 286 FSB @ load w/ (1) Cruise
-- 266 FSB --( +32 )--> 298 FSB @ load w/ (2) Sports
-- 266 FSB --( +44 )--> 310 FSB @ load w/ (3) Racing
-- 266 FSB --( +52 )--> 318 FSB @ load w/ (4) Turbo
-- 266 FSB --( +64 )--> 330 FSB @ load w/ (5) Full-thrust
So here are my settings:
CPU ......... 8x
FSB ......... 266 MHz
C.I.A.2 .... Full-thrust
RAM ........ 2x multi, 5-5-5-15 (subtimings are auto.)
VPCIE ..... +0.1v
VFSB ....... +0.1v
VMCH ...... +0.1v
VDIMM .... +0.1v
VCORE .... 1.35v
This gives me:
266 x 8 = 2128
330 x 8 = 2660 (<-- when C.I.A. 2 kicks in, under full load)
So I can run my 330 FSB in games, yet have a stable S3 suspend/resume @ 266 FSB.
Of course, this probably does little for people who must run FSB above 333 mhz load to get a decent overclock, since the C.I.A. 2 boost seems to max out at 64 mhz.
Gigabyte really needs to fix this problem... :|
One other thing I found today...
Raising PCI-E voltage by 0.1v was the only thing that helped me hit higher FSBs (333 --> 345 mhz). All the traditional changes -- RAM timings, chipset voltage, etc. -- do not make my PC stable above 333 mhz. This was a reproducible result. In fact, none of my memory subtiming adjustments did anything for this little FSB wall.
I'm not recommending PCI-E voltage increases. I'm just mentioning them because it seems wierd. Does the board have some sort of problem supplying power to the PCI-E bus at high FSBs?
Raising PCI-E voltage had no effect on my S3 suspend/resume problems, unfortunately.