Question on some interesting settings

[Ollii]

Hey, as some of might of seen I have been overclocking my system recently but I have finally come to the conclusion that no matter how many volts I through at it 4.2 GHz just doesn't want to be completely stable

So I have now backed it down to 4.0 GHz, much to my PC's relief

I started off with a 4GHz clock (20 x 200) and some fairly high volts (similar to the 4.2 GHz volts) just to see if I could make it really stable, here were the stats:

CPU Voltage - 1.375

PLL - 1.90

QPI/DRAM - 1.50

(Ram was at standard 1600 MHz, 3200 UCLK, 6-8-6-24 Timings at 1.65 volts)

Now with this set-up I thought would be absolutely fine and it would be simply going down with the volts from here on. I was perhaps very surprised then, when it failed the LinX tests on anything above 2048 MiB. The temperatures were a rather nice change though from the 4.2 GHz clock; instead of being 83ish degrees they were at, they were now hovering around the 75 degrees mark.

So after this I thought: Bugger it I'm going to wack the volts up to the highest that I personally would let them run at. Here were the volts:

CPU Voltage - 1.40

PLL - 1.90

QPI/DRAM - 1.60

(Again the Ram was at the standard 1600 MHz, 3200 UCLK, 6-8-6-24 Timings at 1.65 volts) EDIT: I actually did 2 runs at these settings, one with the Ram stats stated earlier and one at 2000 MHz, 4000 UCLK, 8-9-8-24 Timings at 1.65 Volts. The below script is actually from the 2000 MHz run but the lower MHz setting was the same and again struggled to pass anything above the 2024 MiB mark on LinX.

So I started the LinX tests feeling really rather confident that with these volts that I would smash through the 2024 MiB barrier that I seem to have come across. I started off with a 2024 test, just to get things going. It passed it but no real surprises there. I then booted up the 3072 MiB test and from here on it just went downhill; It promptly failed the 3072 test so I put it back down to the 2024 test which it also failed, hmmmmmmm. Lets try it on the 1024 then I thought, It failed that as well. Just for the sake of things and perhaps to try and perk things up a little I put it down again to the 512 test and it failed that as well.

So from this I have concluded that it can't just be the set up that is wrong, I must be doing something wrong with LinX or something. Should I be running any applications at all for instance while it is running.

I am now desperate really to work out what an earth is going wrong, all advice is welcome.

PS) Just to lighten the mood a little, I have managed to get my RAM to the 2000 MHz mark with it being stable etc (So I think

)

 

[Mumbles37]

There's very little chance it's your CPU. It must be your RAM.

I would try to loosen the timings somewhat to check, keeping UCLK at 3200. Try 7-8-7-24 first or maybe even 8-8-8-24.

I've got a setup very similar to yours (950 and Sabertooth) and UCLK will never be stable over 3200 for me simply because I can't raise QPI past 7200. You want to keep UCLK and QPI at an 8:9 ratio, anything over can cause instability. (You need to divide the QPI by 2, then compare with the UCLK manually since Asus doesn't do multipliers in their bios here--3200:3600 is 8:9. That means that if you want to raise the UCLK to 4000, you also have to raise the QPI to a minimum of 9000 to keep the 8:9 ratio, which my particular board can't really handle.) You can check out more information on the 8:9 ratio here.

Anyway, try 7-8-7-24 at 1600, definitely lower your qpi/dram as that is quite high, and also try a lower vcore.

My personal settings for a rock solid 21x200 are:

cpu voltage: 1.31875

pll: auto

qpi/dram: 1.31875

It's possible that your chip/board might need a little bit more voltage, but probably not as much as you're giving them now.