Thermalright Ultra 120 Extreme - not at all flat?

[malfunction]

It's a socket 775 Xeon - basically exactly the same chip as a desktop quad. G0 core - SLACU as opposed to SLACR for the Q6600s. Only 2.13Ghz stock (8 * 266) vs 2.4Ghz for the Q6600. It was the easiest chip I've ever overclocked - 3.2GHz (8 *400) posted and booted @ stock volts, needed a small bump in voltage to be prime stable (I'm using the beta version so I am loading all 4 cores). Have it sitting in an IP35 pro. Tempted to try for more but I'll work on getting the temps down first.

Edit: Have you tried posting @ 400MHz FSB? It may use different timings / strap settings in the chipset and might actually work... I dropped the multi to 6 when I first tried it but when that worked straight away I just went for it. Though the IP35 has a clear CMOS switch on the back panel which helps remove the PITA factor of clocking too high in the BIOS

[matty-hodgson]

that's like a 30% OC.. not bad, not bad at all.
so its pretty much the same but they've dropped the multiplier by 1..
these new chips seem very good at overclocking at tiny voltages which is brilliant!
1.23v for mine at 3GHz
can't work out why i can't hit the 9 * 334 though.. it hits 9 * 333 and is Prime Stable.. so.. confusion all round! :S
any ideas on what i'm doing wrong?

how much was your Xeon btw?

Edit: i'll try posting at 400 now, 3.6GHz, would that be wise? LOL!
i like using the EIST, just 'cause it doesn't come above 2GHz until i start gaming, which is genius in my eyes

[malfunction]

The Xeon was just under £150. Scan were out of OEM Q6600s at the time so I just went for the Xeon as it was £20 cheaper than the retail Q6600 (I was planning on ordering the Ultra 120 anyway) and all of the forum posts I could find (here and elsewhere) indicated that I ought to get a G0 core.

As per my edit above I'd try posting @ 400MHz on a 6x multiplier - should at least let you know if your mobo and RAM are up to the job. I'd check RAM multiplier / ratio is OK and bump up any chipset voltages slightly too not just the CPU volts

[matty-hodgson]

will i need to bump both SB and NB? or just the NB seeing as the SB only takes care of low power things..

i tried it then, but tried it at 3.6GHz :S not at 400 * 6..

i'll try that now.

you got MSN btw?
just easier to talk through that, add me if you do please, my address should be along the left.

[malfunction]

will i need to bump both SB and NB? or just the NB seeing as the SB only takes care of low power things..

i tried it then, but tried it at 3.6GHz :S not at 400 * 6..

i'll try that now.

you got MSN btw?
just easier to talk through that, add me if you do please, my address should be along the left.

I would imagine 3.6GHz would need a higher vcore - most of the posts I've read talk about ~1.4V + at that level though I'm sure there are exceptions. This is my first serious attempt at overclocking any recent CPU but the rules are pretty much the same - find the weakest link then optimise around the limits you have. So try and find the highest FSB but don't overclock the CPU too much (or your RAM) while doing so. Then try and find the highest CPU clock without pushing the FSB too far. Then try and find the sweet spot (CPU, FSB and RAM all overclocked if possible) but until you've tested each separately (well as separately as possible) don't push them all at the same time - otherwise you've no idea which component is letting you down. Note that lowering the timings on your memory may do more for performance than increasing the memory clock speed / FSB. Also remember that a 3Ghz quad core isn't too shabby - it's not the olympics so don't get hung up if you can't get as high an overclock as others (I've been quite sad about it in the past and it's not really worth it)

Edit: No MSN at the mo - not even installed it yet - I'm tuning and rebooting too so I'd be up and down like a yoyo anyway

[IAmATeaf]

I lapped my U120A, it's the normal not extreme version, only used 400 and 1200 grit wet and dry as that's all I could find. I used a craft knife blade to check the HSF base and the first pic shows the pretty grid I marked onto the base which highlights the high points on the base i.e. the centre and left and right vertical edges. Next few pics are work in progress and the last few are to finish off using the 1200 grit.

Does take time but you have to take it slow and easy, also lapped my Q6600, the CPU was way easier then the HSF to lap and after doing the CPU I redid the HSF as I realised it still wasn't quite flat.

[malfunction]

Heh... Now I'm embarrased. Was just in the BIOS again and checking out the fan speed settings when I realised I'd left them on auto. I.e. the fans were running nowhere near full speed (just 300 RPM for the one on my ultra 120!). Temps have, erm, dramatically dropped... Now 25 - 30c idle (edit: 52-59 load so still quite a bit better than before)

[malfunction]

I presume you drew the patten on to see where the high and low spots were? Or were you having a Rolf Harris moment? I'm surprised the non extreme was also quite uneven - looking at the base of mine it looks like the extra heatpipes are the cause of the warped base.

Edit: I've also considered lapping the IHS on the CPU but thought I'd wait until I've had it a week or two first... Was the IHS quite uneven too? I've read various posts saying that they are now a lot flatter than they used to be. I guess the killer question is how much benefit you've seen after lapping both?

[matty-hodgson]

400 * 6 also doesn't work. i reckon it's my RAM 'cause i can't even get the advertised timings on it, the RAM sticker says 4 - 4 - 4 - 15
but i can only get 5 - 5 - 5 - 12..

it makes no sense..
i blame my motherboard!
pile of ****..

thanks for the help though.. it could also be the FSB though, that's at 1331MHz when OCd.

[malfunction]

400 * 6 also doesn't work. i reckon it's my RAM 'cause i can't even get the advertised timings on it, the RAM sticker says 4 - 4 - 4 - 15
but i can only get 5 - 5 - 5 - 12..

it makes no sense..
i blame my motherboard!
pile of ****..

thanks for the help though.. it could also be the FSB though, that's at 1331MHz when OCd.

You can always try loosening the timings. I'm not familiar with your mobo but on the IP35 there's a ton of voltage options (not just CPU core but signalling / termination voltage too). It's a bit of an 'e-penis festival' but you might want to check out XtremeSystems Forums to see if anyone else is using the same motherboard as you and what settings they've used to overclock. Also check your RAM voltage - might need to be higher than your mobo is setting to work at the advertised timings. Same goes for your RAM ratio - if you're running @ 333MHz FSB wise is the RAM at 800Mhz at the moment (~ 5:6 ratio?) or is it already overclocked because you're running a 266MHz FSB chip (~ 4:5 ratio?). Might be worth checking with CPU-Z

[TooNice]

Meh, just ordered one. I wouldn't be too impressed if it needs to be lapped to perform at 90% of it's potential to be honest.

[IAmATeaf]

I presume you drew the patten on to see where the high and low spots were? Or were you having a Rolf Harris moment? I'm surprised the non extreme was also quite uneven - looking at the base of mine it looks like the extra heatpipes are the cause of the warped base.

Edit: I've also considered lapping the IHS on the CPU but thought I'd wait until I've had it a week or two first... Was the IHS quite uneven too? I've read various posts saying that they are now a lot flatter than they used to be. I guess the killer question is how much benefit you've seen after lapping both?

I drew the lines in Rolf Harris mode and so that I could quickly see the high spots being ground away.

The base of my HSF was convex as was the base of the IHS on my quad, put the 2 together and you end up with an arrangement that rocks! So I ended up lapping the quad, one think that I did do which was rather stupid was I lapped the quad before even testing that it worked, was only part way through I thought what if it doesn't work

Anyway all was fine, now got it running at 3Ghz with temps maxing out at around 50 across all 4 cores.