DFI NF4 Ultra Infinity Overclocking

[QuietIce]

The CPU must have given up - MemTest doesn't stress the CPU a whole lot.

What was the vCore ...?

 

[HoTsHoT]

The CPU must have given up - MemTest doesn't stress the CPU a whole lot.

What was the vCore ...?

Stayed at it's default setting at 1.35v, then I bumped it up to 1.4v to no avail. What has made me stumped is, it was possible before to do a blend test with this configuration.

 

[QuietIce]

I've come to dislike P95 blend. I either use large or small FFT's now since I've had several CPU's crap out under blend then run SETI forever without a bad work unit. In the past SETI would sometimes crash a system that was P95 blend stable so I wonder if they've changed the P95 program?



Anyway, you've found the board top end so the next step, IMO, is the CPU OC. I know infinitevalence shows RAM next in his OC thread but to me that's backward. Since CPU speed is by far the most important aspect of an OC I always do it first then RAM test such that I don't ruin the CPU over-clock. I also try to keep RAM speeds near their stock rating while OC testing the CPU - and that's even different than my AM2 OC sticky (I really need to edit that one of these days!) ...

 

[HoTsHoT]

I've come to dislike P95 blend. I either use large or small FFT's now since I've had several CPU's crap out under blend then run SETI forever without a bad work unit. In the past SETI would sometimes crash a system that was P95 blend stable so I wonder if they've changed the P95 program?



Anyway, you've found the board top end so the next step, IMO, is the CPU OC. I know infinitevalence shows RAM next in his OC thread but to me that's backward. Since CPU speed is by far the most important aspect of an OC I always do it first then RAM test such that I don't ruin the CPU over-clock. I also try to keep RAM speeds near their stock rating while OC testing the CPU - and that's even different than my AM2 OC sticky (I really need to edit that one of these days!) ...

I've already considered doing processor overclocking first over memory after discovering that the processor speed affects the whole overclock. But with my new cooler arrival stalled, it'll probably have to wait till then. One of the main things why I love overclocking is actually like a test of luck. It's exciting to know if your processor is either a lump of gold, or a dud. Either way, I still live with it

210 Mhz 1:1 at 10x seems quite stable for now, and I won't be having any work done on this PC here for a while, guess I'll play around with it. Maybe I'll take some time to finish up Fallout 3

Cheers to you QuietIce, you've been a great help

 

[QuietIce]

Just a final note, then.

As long as you stick to stock vCore your temps won't go up much, so push the CPU all you can at stock volts and let it run that way while you await the new cooler. Generally 55°C at full load is acceptable, 60°C max for 24/7 computing ...

 

[HoTsHoT]

Just a final note, then.

As long as you stick to stock vCore your temps won't go up much, so push the CPU all you can at stock volts and let it run that way while you await the new cooler. Generally 55°C at full load is acceptable, 60°C max for 24/7 computing ...

I was able to get an Arctic Cooling Freezer 64 Pro for an affordable price, it'll definitely work better than an XP-90 or the stock four heatpipe AMD cooler.

Once it arrives, time to crank up the volts

 

[HoTsHoT]

From what I started:

Stock AMD cooler --> Thermalright XP-90 --> Arctic Cooling Freezer 64 Pro --> Evercool Buffalo --> Sunbeamtech Core Contact Freezer

I've seriously blown my budget big time. No matter, let the fireworks begin!

 

[HoTsHoT]

A little update. Managed to push it up to 2.4GHz with 1.525v. THis puppy isn't going as far as I expected. Currently running 267MHz clock on 9x multiplier, 5:6. The Trc and Trfc timings played a big role in perfecting stability, as I recently found out.

So far, I've hit a dead end. I can't increase my multiplier without getting that Nvidia BIOS IDE RAID ROM appearing during boot. I've read around and it says that it's a common problem among NF4 boards, and also appears during over-the-edge overclocking. I like playing dangerous, so is there any way to fix it? Things I have tried:

-Disabling RAID on BIOS
-Disabling one of my DVDROMs (apparently there's a bug where the RAID message pops up when you have two of them)
-Disabling one of my hard drives
-Moving my harddrives to different SATA ports
-Changed the nvraid.rom to a newer version

Things that would trigger it include:

-Raising the memory voltage too high
-Changing multipliers (I'm stuck at 9x, going higher or lower causes this problem)
-Changing memory timings (especially the Trc and Trfc)

Also, made a small observation. I have to pump correct vDDR at the correct speed range. For example, if my RAMs are running stock, they have to be fed 2.6v. Given more than that would cause BSODs. Never thought that this kind of complications would actually happen.

 

[QuietIce]

Sorry to hear about the problems. Must be a DFI thing, I didn't have those kind of issues with my NF4 builds ...

 

[HoTsHoT]

Sorry to hear about the problems. Must be a DFI thing, I didn't have those kind of issues with my NF4 builds ...

Don't have to be sorry man, I would have been pointlessly wasting a lot of time if not for your advice

Finally things have turned better. I tried the 10x multiplier again, and all of a sudden it booted without a hitch, when I previously had those NVIDIA RAID messages. A quick test revealed that I was able to hit 2.6GHz, with the 5:6 ratio to adapt to my RAM's limitations. Any higher and the sticks would crap out.

Now 2.6GHz, that's a nice number to have - till I started testing it. This processor I'm having here must have been the worst out of its batch, as it has serious voltage hunger problems and doesn't work as how much it is fed. I could only get 2.4GHZ stable at 1.55V previously, which already was a bad omen. Pumped in 1.775v and it still died within a few minutes on the Small FFT test. To make it worse, I read that the voltage I gave it is only suitable for PCs which use phase-change

Everything's not all bad though, as I don't experience the NVIDIA RAID messages anymore, even if I raised or reduced voltages. It seems like a multiplier problem here

My PSU's starting to strain its back too, as I'm beginning to see drastic changes on my 12v rail everytime I change the Vcore. And here I thought that a 480W should be enough for everyone

Also, a question came into my mind. Is it safe to run the processor at very high voltage levels, even if the HSF is able to keep it cool? If it is, does it mean that you can subject the processor to as many volts you want, as long you have the cooling to pair with it? I know electromigration would eventually kill it, but I'm talking short term here

 

[QuietIce]

Processor stepping makes a big difference with the Toledo's. I've got a couple of late run Opty 165's running almost 3.0 GHz on air but my Opty 180 hit a wall at 2.8 GHz on water. It's all a matter of the stepping since AMD was binning them pretty tight for awhile and playing with the silicon a lot.

A fluctuating rail at those voltages could just mean you're close to burnout on the motherboard MOSFETs. The whole system starts to demand more and more as the voltage goes up - mostly due to greater resistances at higher temps. That same rule applies to the voltage converters and regulators as well as the CPU. How hot are the motherboard chips - have you checked them (fingertip "probe" ) at those voltages?

My personal limit for s939 K8's is 1.55v. I run a couple of my rigs at that vCore for about a month/year during the SETI races. Usually, though, they're turned down to 1.45± vCore just to keep room temps down. I've seen people run 1.6v semi-long term with reasonable temps - like OC'ing it's a matter of personal taste (and risk) ...

 

[HoTsHoT]

Processor stepping makes a big difference with the Toledo's. I've got a couple of late run Opty 165's running almost 3.0 GHz on air but my Opty 180 hit a wall at 2.8 GHz on water. It's all a matter of the stepping since AMD was binning them pretty tight for awhile and playing with the silicon a lot.

A fluctuating rail at those voltages could just mean you're close to burnout on the motherboard MOSFETs. The whole system starts to demand more and more as the voltage goes up - mostly due to greater resistances at higher temps. That same rule applies to the voltage converters and regulators as well as the CPU. How hot are the motherboard chips - have you checked them (fingertip "probe" ) at those voltages?

My personal limit for s939 K8's is 1.55v. I run a couple of my rigs at that vCore for about a month/year during the SETI races. Usually, though, they're turned down to 1.45± vCore just to keep room temps down. I've seen people run 1.6v semi-long term with reasonable temps - like OC'ing it's a matter of personal taste (and risk) ...

Well, my processor was a random one so I guess I should be happy with what I've got, rather than complaining about something that I can't or don't have.

Motherboard MOSFETs? The 12v rail on load fluctuates as I raise the vCore, shouldn't this problem be associated with the PSU not providing enough juice? As for the motherboard chips, I had a guide on where the hotspots were on the DFI boards, and added a few heatsinks on my own so it shouldn't be much of a problem.

Right now I've pretty settled down. 2.5GHz, 10x multiplier on 1.65V running on 5:6 ratio with RAMs running on 208MHz, 2.5-3-3-7, 2.8V. To get stable on 2.55GHz I would need to feed 1.7v, which is ridiculous for such little improvement. Passes all P95 tests with ease too, with the processor bordering 70C on a hot day.

A humble screenshot:

 

[QuietIce]

2.5 GHz is not bad for a 3800!

If it's just a random chip then you could have easily gotten one of the mid-run CPU's and, if so, that's a very good OC ...

 

[HoTsHoT]

2.5 GHz is not bad for a 3800!

If it's just a random chip then you could have easily gotten one of the mid-run CPU's and, if so, that's a very good OC ...

I actually set a goal to reach at least 2.6GHz, but a 100MHz short from my target might not be so bad.

Done a little benching with Sisoft Sandra and my processor is about on par with a C2D E6300. Goes to show how a new architecture can really optimise things

 

[HoTsHoT]

Crap, I used Modbin to unlock the 183MHz divider in my BIOS, flashed it and now my PC won't start. It would just power up, then nothing happens. My BIOS beeps still work fine (like when you don't insert any memory in), and I've tried clearing the CMOS.

It seems to be a 'voltage memory' problem, where if you feed too much of voltage to any particular component, it would show funny symptoms like unable to run at stock voltage anymore. I ran into a few problems like this before, but currently now I can't boot the PC at all - any ideas?

Damn, serves me right for being greedy.

 

[HoTsHoT]

Finally, back into my comfort zone. My experience told me that re-enabling hidden settings in the BIOS with Modbin6 couldn't pose a problem, but I was wrong. Fortunately, a spare P4 rig and Winflash got me running back up.

 

[QuietIce]