Help with OC'ing X5670

[Collider]

I seem to be stuck with my OC and need some help pushing my X5670 on ASUS P6T Deluxe V2.

I'm having a hard time pushing it past 3.8 - 4Ghz. For 4Ghz IBT fails on Max setting after 3-4 passes. For 3.8Ghz IBT Failed after 16 passes. Prime95 doesn't detect any problems and runs fine 1-2 hrs at time.

My temperatures are pretty good 38C idle and 65C max under 100% load in IBT or Prime95. Using air cooling, Noctua + 2x fans.

Multiplier: 20x
BCLK: 200
DRAM: 1600 (stock spec)
DRAM Timings: 9-9-9-24 (stock spec)
UCLK: 2800 (1.75x)
CPU Voltage: 1.325
CPU PLL: 1.86
QPI/DRAM Voltage: 1.3
IOH Voltage: Auto
IOH PCIE Voltage: Auto
ICH Voltage: Auto
ICH PCIE Voltage: Auto
DRAM Voltage: 1.5 (stock spec)
LLC: Enabled
CPU Spread Spectrum: Disabled

What else should I tweak? Looking to get a 24/7 stable OC.

Is my CPU Voltage too low?

Also should I manually set IOH / ICH voltages? If so, to what and what are the safe ranges?

 

[ClockHound]

Could play with upping the multiplier and lowering BCLK. Also bump up the vcore to 1.35-1.37(ish). Running a 5670 at only 20x is kind of a waste imo.

 

[Collider]

Could play with upping the multiplier and lowering BCLK. Also bump up the vcore to 1.35-1.37(ish). Running a 5670 at only 20x is kind of a waste imo.

Thanks - I was thinking that my VCore is a bit low, but I've read posts of folks getting 4Ghz @ v1.25-1.30 which on my experience is too low. I will try upping the VCore.

You're right, I actually got the x5670 for its 22x & 24x multipliers.

How about IOH & ICH voltages? I dont trust my bios' AUTO settings.. What exactly are they responsible for?

 

[unseenmorbidity]

You don't want to just slam it on 200 BLCK and hope for the best. You got to work your way up there.

Vcore seems quite low for that OC.

 

[ClockHound]

Thanks - I was thinking that my VCore is a bit low, but I've read posts of folks getting 4Ghz @ v1.25-1.30 which on my experience is too low. I will try upping the VCore.

You're right, I actually got the x5670 for its 22x & 24x multipliers.

How about IOH & ICH voltages? I dont trust my bios' AUTO settings.. What exactly are they responsible for?

You can set the IOH & ICH manually to the default values - shouldn't impact OC. (Those voltages are for the i/o hub controller)

Suggest starting with a 24x and BCLK in the mid 170s. (Or 22x Mult & 182 BCLK). Bump vcore to 1.35v and see how far it goes. I expect 4.2 should be stable. Beyond that...well...silicon lottery and cooling get to lend a medling hand. What is your cooler?

Also don't be afraid to under clock memory until you've found a stable setup for the cpu OC. Trying to optimize everything at once is waay harder than getting one thing solid.

You've seen the monster X58/56xx thread here, right? https://forums.anandtech.com/threads/xeon-l5639-overclocking-on-x58.2335636/page-107 - that page has a specific mention of the P6T Deluxe V2 messing with mult when all cores under load. Dang!

Here's some settings from that thread - from Burpo's 5670 in a Sabertooth:

My experience with the 5675 & Sabertooth is a higher mult & lower BCLK results in better stability and lower temps. At 1.32V and 4.4 (25 mult 176 BCLK) all cores in IBT generate 60c max temps under a D15 inside an airflow-optimized and modded HAF XB. Also use offset voltage, so it idles at 1.1V.

My other X58 rig with a 5660 and P6T SE runs around 3.9 with 1.36V, no offset voltage option, so it's only turned on when an extra render machine is required.

These old Xeons are still fun to play around with.

 

[Collider]

@ClockHound thanks for the tips.

What you're saying makes sense, based on that I will change my approach and start with high multiplier, low BCLK and work my way up from there.

As for cooler I'm using an older Noctua air cooler (cant find the model #) - seems to be holding its own so far.

 

[lopri]

CPU PLL is way too high. You should try lowering it, not raising it. Try 1.8, and lower it step by step. (e.g. 1.75, 1.70, 1.65)
All those IOH (north bridge) and ICH (south bridge) votages should be left auto (stock). You are overclocking the CPU, not the NB or SB.

Try multiplier x19 or x21 instead of x20. It has been known that x20 multiplier requires more voltage. (There is an explanation as to why but I forgot.)

Since you are passing Prime95 but not IBT (Linpack), check the memory settings (clocks, timings, voltage, and divisor). If memory is fine, you probably need more vCore or better cooling. And try a different version of IBT, or better yet download Linpack directly from Intel. -> https://software.intel.com/en-us/articles/intel-linpack-benchmark-download-license-agreement

Note that Hyperthreading should be turned off while running Linpack because Linpack's optimized codes stress cores more/better without Hyperthreading.

 

[Collider]

Ok, so it seems I can't get stable x24 Multiplier, upon 100% load it throttles back to x22 according to CPU-z. When CPU is idle CPU-Z reports x24 multiplier.

This is with PLL set to 1.8.

Any advice on how to get stable x24 mutli ?

 

[ClockHound]

See that link above...or here: https://forums.anandtech.com/threads/xeon-l5639-overclocking-on-x58.2335636/page-107
Note there that the P6T Deluxe V2 drops the mult when all cores are under load. Believe it's a power thing in the BIOS.

Is there a manual multiplier lock option in the BIOS? There is in the Sabertooth. Could try flashing the WS bios...but research first....

 

[MongGrel]

See that link above...or here: https://forums.anandtech.com/threads/xeon-l5639-overclocking-on-x58.2335636/page-107
Note there that the P6T Deluxe V2 drops the mult when all cores are under load. Believe it's a power thing in the BIOS.

Is there a manual multiplier lock option in the BIOS? There is in the Sabertooth. Could try flashing the WS bios...but research first....

I do not believe the V2 can do it like a Sabertooth, but I would be leery of flashing the bios on one along those lines unless you seriously research it myself.

I'd be worried about bricking the board personally.

Have a V2 with a X5650 in the bedroom I really never push that hard, the main has a P6T7 with a X5680 in it, I usually leave that around 4.4 just to be stable.

It runs a lot of things at once though.

 

[Collider]

@ClockHound , so I'm testing that lower Bclk w/ higher Multi & already seeing lower temps per Mhz. Getting 55C 100% load in IBT testing 4.2Ghz @ 22x190 w/ vcore=1.344, where before 4.2Ghz @ 21x200 was 65C w/ v1.325. I'm starting to see that lower BCLK yields better results and lower temps with lower vcore. Will test some more and post results.

 

[.vodka]

It was common advice back in the day to stick with odd multipliers because they were more stable than even multipliers on Nehalem, at least on 45nm Bloomfields. I don't know if this rule applies to 32nm Westmere, but it doesn't hurt to try... 23*174 = 4GHz, see if you can get that stable and work from there. That multiplier still gives you lots of headroom, 200*23 = 4.6GHz if your chip can do 200MHz bclk.

Also, wasn't the uncore clock supposed to be at least twice the memory clock on this platform to be stable and avoid issues? If you're running DDR3-1600 that should be 3200MHz for the uncore, in the first post you mention DDR3-1600 and 2800MHz uncore, that could be part of your problem.

Lots of good information here, from ballpark voltages to multiplier combos to try.

 

[ClockHound]

@ClockHound , so I'm testing that lower Bclk w/ higher Multi & already seeing lower temps per Mhz. Getting 55C 100% load in IBT testing 4.2Ghz @ 22x190 w/ vcore=1.344, where before 4.2Ghz @ 21x200 was 65C w/ v1.325. I'm starting to see that lower BCLK yields better results and lower temps with lower vcore. Will test some more and post results.

10 degree drop is much, much better. And yet with increased vcore. Congrats!

Are you using fixed vcore? What are the temps at idle?

 

[Collider]

So I ran IBT for 10 passes at max setting & max ram and it passed at 4.18Ghz @ 22x190 w/ vcore=1.344. Previously 4.2Ghz @ 21x200 w/ vcore 1.325 would fail after 2-3 passes.

Ran IBT again to make results more conclusive and it just failed after 2 passes. So back to the drawing board...

Using fixed vcore, as my board doesn't have adaptive vcore settings. With LLC enabled I do see fluctuations of +/-0.008v when idle vs under load.

My Temps are ~60c under 100% load and about 38c when idle.

Kinda, bummed at the fact that I can't make use of 24x multiplier - that was the reason I chose x5670 over x5650. Flashing BIOS to another board's ROM sounds kinda risky. Wondering if there's any other way to force 24x multi..

 

[ClockHound]

Well...that sucks...Could bump the vcore a bit and see how it goes at 4.18....but shouldn't need that much juice for that moderate clock.

I don't know of any workaround to unleash the beast in the P6T BIOS as it has throttle code that can't be unlocked in the bios unlike the WS version.

Options are:

1) Flash the BIOS to P6WS - Done wrong it ends badly (Altho if it ends badly, it's nature's way of saying... Sabertooth). However, 6T deluxe and WS boards are very similar. Here's a good flash guide: http://linas.16mb.com/asus-p6t-deluxe-crossflash-to-asus-p6t-ws-pro/

2) Watch for a deal on a Sabertooth, P6x58d-e or WS board.

3) Run the P6T at 4Ghz and be happy.

3b) Buy a cheap(ish) W3680 or 3690 - have unlocked mults. 130W beasts, but still. And sell the 5670 to Burpo. Profit!

3c) Or watch for some silly deals as the leets dump their hexcore X99 rigs for a pittance in the scramble to keep their hopes ryzen to 8 cores. And still sell your 5670 to Burpo. Profit!

 

[Collider]

This blows big time, I went down to 3,74Mhz w/ 22x170 @ 1.35 vcore and 1.325 qpi and still failing IBT on max setting.

Tried setting my dram bus voltage to 1.6 & 1.64 (when default is 1.5).

Starting to lose hope here.. It seems that from these voltages I should be able to hit at least 4.0Ghz stably. Starting to wonder if its my ram..

 

[ClockHound]

Could be. Have you run a full RAM test at stock? Memtest86 is your friend. Or worst nightmare.

One of my X58 boards ended up having a flaky ram slot...took weeks to discover it. Don't lose hope, just work the problem. Before it outworks you. ;-)

 

[Collider]

Ok, so I think I found the reason for the instability and now that I think of it its such an amateur mistake.

I clocked my system down to stock settings 21 x 133 and it still failed IBT on max after 12 passes. One of the biggest issues with testing is that it takes forever for IBT to run on max with such large amount of ram. I also tried with 3 another CPUs (I have 3 Xeons laying around so I knew the problem had to be elsewhere.

I've been using 24GB of ram from 3 separate kits. The kits are identical and were purchased together - so I didn't think that would be a problem. I bought them together during a black friday sale years ago for $75 total (25 ea.) which was a steal, even by today's prices (each kit currently ~$60 ea.)

I've been using them for years and can't believe my system was unstable all this time. I've had random reboots from time to time and never paid attention to it until I started running resource demanding loads on my machine (data imports / db work / etc.) and noticed more frequent reboots during loads. I've always attributed the instability to the fact that my CPU was overclocked.

I contacted G.Skill support and they said they they should be willing to exchange 3 separate kits for 1 single 24GB kit as a one time courtesy. Their support seams great. Now just waiting for my RMA.