Asus Rampage Formula (X48) Thread

[RPIJG]

Earlier BIOS versions on the RF had a problem with LLC but the more recent BIOS versions it has been corrected, and is much better than without the LLC I found.

[Demented]

Well, with just turning LLC off, I blue-screened as soon as I rebooted. Then I upped my vCore to 1.365v, with it reading at 1.328v in bios, and just booted up fine. With LLC on before, I had my vCore at 1.35v, and it read at 1.336. So I figured if I disabled LLC, I need to make sure that it at least reads at close to that. However, in order for it to read at 1.336v, I'll have to set it at higher than what my chip's supposed max is, so I'm wary of doing that.

So with LLC disabled, I think I might be back to the drawing board as to finding safe overclocking settings. I think I'll have it run Prime95 overnight tonight but I have a question. I ran Prime for a little while earlier today, before I disabled LLC and had vCore set to 1.35v and reading 1.336, and after about 30 minutes it seemed my second core (01) had a failure, but the first core (00) was ready to keep going. If I were to let it run, and go to sleep, if that happens again, will it cause more problems? I know that perhaps the whole reason Core 01 failed in the first place might have been due to the LLC issue, but just wanted to get an opinion.

[Ghostleader]

Originally Posted by Arbie

I've been running a Rampage Formula since last August, using a QX9650 oc'd to 3.8GHZ on air. It does Orthos/Prime stable for 24+ hrs at 102 deg F ambient. The quad gets hot but it's stable. Never a BSOD.

In setting up this rig I learned a LOT from Kris Boughton's excellent articles on the Anandtech site regarding this mobo. He made a very good point regarding LLC (http://www.anandtech.com/cpuchipsets...x?i=3184&p=6):

"ASUS' implementation of this feature, labeled as Load Line Calibration and included with their latest line of motherboards, is particularly worthy of our attention for a number of reasons. The first is that setting lower voltages with this option enabled actually results in a condition in which the CPU voltage under load is higher than the idle voltage. Imagine our confusion as we desperately struggle to understand why our system is Prime95 stable for days yet continues to crash under absolutely no load.

Suffice it to say, we found it better to leave Load Line Calibration disabled."


Kris has also published some related works on Ars Technica. Tracking back through those and the Intel datasheets he references, it is clear to me that Load Line Calibration is a silly feature put in by Asus to satisfy some poorly-informed enthusiasts who think the power to the CPU can and therefore should be more tightly controlled. They want to eliminate the "VDroop" characteristic purposely specified by Intel to provide more positive and negative headroom for inductive kicks out of the bucking converter when CPU load changes. Turning this feature ON will do no good and will often cause seemingly random blue screens / lockups, especially when going from heavy load to light load (which puzzles everyone). There's no information on what the AUTO does so it's safest to choose DISABLE.

I'll bet that if you disable LLC, your random BSOD problem will go away. There could be another cause, of course, but this is the prime suspect.

Arbie

Hi and welcome to OCN.

Good and informative info.

Originally Posted by Demented

Well, with just turning LLC off, I blue-screened as soon as I rebooted. Then I upped my vCore to 1.365v, with it reading at 1.328v in bios, and just booted up fine. With LLC on before, I had my vCore at 1.35v, and it read at 1.336. So I figured if I disabled LLC, I need to make sure that it at least reads at close to that. However, in order for it to read at 1.336v, I'll have to set it at higher than what my chip's supposed max is, so I'm wary of doing that.

So with LLC disabled, I think I might be back to the drawing board as to finding safe overclocking settings. I think I'll have it run Prime95 overnight tonight but I have a question. I ran Prime for a little while earlier today, before I disabled LLC and had vCore set to 1.35v and reading 1.336, and after about 30 minutes it seemed my second core (01) had a failure, but the first core (00) was ready to keep going. If I were to let it run, and go to sleep, if that happens again, will it cause more problems? I know that perhaps the whole reason Core 01 failed in the first place might have been due to the LLC issue, but just wanted to get an opinion.

Was it P95 small FFT or blend?

Any progress?

[RPIJG]

Originally Posted by Demented

Well, with just turning LLC off, I blue-screened as soon as I rebooted. Then I upped my vCore to 1.365v, with it reading at 1.328v in bios, and just booted up fine. With LLC on before, I had my vCore at 1.35v, and it read at 1.336. So I figured if I disabled LLC, I need to make sure that it at least reads at close to that. However, in order for it to read at 1.336v, I'll have to set it at higher than what my chip's supposed max is, so I'm wary of doing that.

So with LLC disabled, I think I might be back to the drawing board as to finding safe overclocking settings. I think I'll have it run Prime95 overnight tonight but I have a question. I ran Prime for a little while earlier today, before I disabled LLC and had vCore set to 1.35v and reading 1.336, and after about 30 minutes it seemed my second core (01) had a failure, but the first core (00) was ready to keep going. If I were to let it run, and go to sleep, if that happens again, will it cause more problems? I know that perhaps the whole reason Core 01 failed in the first place might have been due to the LLC issue, but just wanted to get an opinion.

Your problem, is the exact reason that LLC exists. When under load, your core voltages are going to drop, if they drop low enough you will get instability. LLC maintains a relatively more stable core voltage even under load.

Think of it like this. In a car at idle, your injectors are pulsing at some base level, whatever is required to maintain the engine running correctly (this is like your idle Vcore). When you mash on the throttle, your cars ECU signals the fuel pump to send more fuel and the injectors to pulse faster giving more fuel so that the car runs correctly (LLC enabled). If when you mash on the throttle the ECU does not signal this, then your car would buck and bang and eventually detonate causing damage ie instability (LLC disabled). If this was the case, you could raise the idle fuel pressure/injector pulse to compensate, but for any time that you aren't running at wide open throttle, you would be wasting fuel and be inefficient. Obviously this is a rough comparison, as there are many differences, but the idea is there.

With LLC enabled, you can run a lower idle Vcore because the corresponding drop in Vcore when load is applied will be managed by the Line Load Calibration.

With LLC disabled, you will need to run a higher idle Vcore to compensate for the drop in Vcore when load is applied.

I personally have always used LLC, even with my Q9650 and the RF, I ran that stable at 4+GHz without any issues, and far better temps than if I disabled LLC.

[Demented]

@Ghostleader: It was the Prime95 Small FFT test.

@RPIJG: I definitely understand the LLC a bit better now. I watched my voltage drop from 1.34v to 1.31v under load before I upped it to 1.3625 in BIOS. With LLC on, it stays a nice steady 1.34v now.

Anyways, I started running another Prime95 Small FFT test last night. I put my vCore to 1.3625v, and had LLC DISABLED. It seemed to be doing fine, vCore stayed at a steady 1.34v (according to HW Monitor), and the cores got to about 58c. Then, a little after an hour into it, 0x00000124 BSOD.

So I left my vCore at 1.3625, enabled LLC and watched a little DVR'd shows before turning it off for the night.

I guess I just don't have the patience to play with every possibility at the moment. I'm guessing that when I give it a go again, I'll leave LLC on, and make sure that my vCore is set to a voltage that will at least read at 1.344v or higher, since if it is more voltage I need, that should do it. At this point though, I'm ready to set aside some time on the weekend, and bring her all back to the beginning. If I do that, should I put my video card back to stock as well? I guess that's wise since I should get one OC stable at a time.

Thanks for all the help so far, and I still am looking for any suggestions that might just nail this issue!

[Arbie]

In case it helps, here are a few other build notes from my Rampage Formula build last year.

CPU is QX9650; 2x2GB of OCZ Reaper 1066 MHz DDR2 RAM. The manual BIOS settings in Tweaker are as follows (I think all others were at defaults):

CPU Ratio - 9.5 (for 3.8GHz)
FSB - 400
PCIE - 100
DRAM - 1066
DRAM Command Rate - 2N
AI CLOCK TWISTER - Strong (see below)
Comm Perf Level tRD - 7 (see below)
CPU volts - 1.35
NB volts - 1.31
DRAM volts - 2.0
FSB volts - 1.3
FSB Term volts - 1.1
SB volts - 1.5
Spread spectrums - OFF
Load Line Cal - OFF

Notes:
- Having Clock Twister = Strong was very important on this baord.
- OCCT gave me nothing but trouble so I finally stopped using it and stayed with Orthos/Prime. - To load the HD4850 graphics board I used rthdribl and later FurBench.
- I got the 1066 MHz RAM to allow tRD=7; slower RAM will probably only work with 8 or more.
- I had no time to try ultra-tweaks like pulling in clocks etc.


As for Load Line Calibration:

The mechanical analogy is irrelevant, as is the fact that some people can enable LLC and have everything work OK. On this motherboard, and on any with a commercially viable power subsystem, you can't get away from some fundamentals. One is that the buck converter output IS going to react to loading changes, and the symptom will be CPU supply voltage swings. That's OK; the CPU is rated to handle this. If you attempt to "stabilize" the supply output and so suppress these swings you will actually force the system into brief over- and undervoltage conditions that the chip designers never expected. Such transients can only be seen on a scope (e.g. Boughton's traces) so if you have only a multimeter you're flying blind.

Bottom line: the price of stability and proper operation is CPU voltage swings, and whatever increased quiescent power dissipation results. But everyone's rig will be different. With a lucky combination of inductors, capacitors, and chips, you might get along "fine" with LLC enabled. But at the height of summer a year from now, with any component aging or deterioration, and some time after you went from 4GB to 6GB, etc., your board may start crashing. Nothing obvious - even rolling back the OC - fixes it, so you figure it's dead (damn Asus!) and rebuild.

I think I've interpreted the source documents correctly, but I'm willing to be corrected by a motherboard / Intel design engineer here. Also, I respect the raw experience of folks who have done a lot of overclocking, modding etc. But you can be fairly successful at that without always knowing what's going on at board level. All the better when some arcane bit IS properly explained, as Boughton did. It helps eliminate a variable.

[RPIJG]

My only comment on this is that those "over/under voltage" conditions will occur regardless of whether or not LLC is enabled or not. In most cases everyone here is pushing more voltage than Intel "recommends" and typically LLC will actually allow you to run less for the most part.

[Ghostleader]

There is a new BIOS available for the ASUS Rampage Formula.

Download BIOS 0902 here.

The problem with the DDR +1200MHz seems to fixed in this BIOS.

(Credit to A-Grey over at Xtremesystems for this)

[TFL Replica]

Originally Posted by Ghostleader

There is a new BIOS available for the ASUS Rampage Formula.

Download BIOS 0902 here.

The problem with the DDR +1200MHz seems to fixed in this BIOS.

(Credit to A-Grey over at Xtremesystems for this)

Fabulous, +rep for spreading the news.

[space@pe]

Thanx for the updated bios m8