1366 x58 Xeon Enthusiast overclocks club

[ZorchThatCPU]

Yep, I guess it is starting to drag a little. We're not benching, so spending another 20 hours tweaking to squeeze out another 1% performance isn't all that appealing to me either. When the weather turns, it'll be like getting a whole new pair of CPUs for me, so I'm still not tempted to spend any more. Until then, I'll probably putter around with re-creating improved versions of the various BIOS profiles I lost, since those were rough drafts that didn't have the memory tweaked very much. And of course this time I'll make sure to save them to a flash drive in the event this board decides to re-flash the BIOS from the backup chip again. But I reckon that's about it for now on my end too.

I'll probably check back in sometime around November. Perhaps by then, I'll have enough posts here that I can upload screenshots, if that's how that works. I'm not really sure.

Anyway, have fun and enjoy your awesome rig. Cheers.

 

[GonzoP]

Yep, I guess it is starting to drag a little. We're not benching, so spending another 20 hours tweaking to squeeze out another 1% performance isn't all that appealing to me either. When the weather turns, it'll be like getting a whole new pair of CPUs for me, so I'm still not tempted to spend any more. Until then, I'll probably putter around with re-creating improved versions of the various BIOS profiles I lost, since those were rough drafts that didn't have the memory tweaked very much. And of course this time I'll make sure to save them to a flash drive in the event this board decides to re-flash the BIOS from the backup chip again. But I reckon that's about it for now on my end too.

I'll probably check back in sometime around November. Perhaps by then, I'll have enough posts here that I can upload screenshots, if that's how that works. I'm not really sure.

Anyway, have fun and enjoy your awesome rig. Cheers.

Yeah if you look at the guys in the first five pages of this club many of them post much much less now. The new car smell fades after awhile. Plus with minor exceptions we're all ending up with nearly the exact same settings and issues. Sure one guy can get 100Mhz more OC or use .05v less vcore but really its the same ballpark.
I too am looking forward to what an ambient 60ºF less might yield me. But as you said that's months away

 

[Zoson]

See everybody talks about QPI like more is better up to the 1.35v. I tend to use as little as possible myself. That is 1.3v for 4.4 and 4.5, I do have to go to 1.32 for 4.6 and 4.7. Ram voltage I'm at 1.65v which I need for my ddr3 1333

Yes, you should absolutely only run as much as is needed to be stable. So if you can use less than 1.35, you should.
On that note, 99% of the time, the point of raising QPI/UCLK/Vtt is to stabilize the IMC. And the IMC is normally where you run into performance limitations. Here are things that I found necessitated running higher QPI/UCLK/Vtt:
1. More memory = more voltage. 3x2GB needs less than 3x4GB, and likewise, populating all slots requires even more than either of those.
2. Increasing memory frequency directly increases UCLK frequency at a given divider. So, increasing memory frequency usually necessitates more QPI/UCLK/Vtt.
3. Tightening timings, and command rate - these increase the internal workload of the IMC, even though clockspeed doesn't increase, so can often necessitate more QPI/UCLK/Vtt.

Now, the strategy I ended up going with was: set to 1.35v, and then find the fastest UCLK speed/Ram Speed/Timings combination I could get at that voltage.

These two shots demonstrate the performance difference nicely.
4604 Core, 3656 UCLK, 1083 RAM - ~88 to 89 GFlops


4610 Core, 4009 UCLK, 1002 RAM - ~90 GFlops


So, you can see that RAM speed doesn't really do anything unless there's enough UCLK to support it(While the limit was lowered to 1.5x, the performance rule of 2x still holds true - that is, you want UCLK speed to be twice that of your memory speed), and correctly balancing your UCLK with your memory will yield measurably better performance.

How are your temps at 4.6? Have you done any LinX testing with the latest linpack binaries?
You should also check out CPU-Tweaker. By far the best timings breakdown of anything I've seen for our cpus.

 

[GonzoP]

Yes, you should absolutely only run as much as is needed to be stable. So if you can use less than 1.35, you should.
On that note, 99% of the time, the point of raising QPI/UCLK/Vtt is to stabilize the IMC. And the IMC is normally where you run into performance limitations. Here are things that I found necessitated running higher QPI/UCLK/Vtt:
1. More memory = more voltage. 3x2GB needs less than 3x4GB, and likewise, populating all slots requires even more than either of those.
2. Increasing memory frequency directly increases UCLK frequency at a given divider. So, increasing memory frequency usually necessitates more QPI/UCLK/Vtt.
3. Tightening timings, and command rate - these increase the internal workload of the IMC, even though clockspeed doesn't increase, so can often necessitate more QPI/UCLK/Vtt.

Now, the strategy I ended up going with was: set to 1.35v, and then find the fastest UCLK speed/Ram Speed/Timings combination I could get at that voltage.

These two shots demonstrate the performance difference nicely.
4604 Core, 3656 UCLK, 1083 RAM - ~88 to 89 GFlops


4610 Core, 4009 UCLK, 1002 RAM - ~90 GFlops


So, you can see that RAM speed doesn't really do anything unless there's enough UCLK to support it(While the limit was lowered to 1.5x, the performance rule of 2x still holds true - that is, you want UCLK speed to be twice that of your memory speed), and correctly balancing your UCLK with your memory will yield measurably better performance.

How are your temps at 4.6? Have you done any LinX testing with the latest linpack binaries?
You should also check out CPU-Tweaker. By far the best timings breakdown of anything I've seen for our cpus.

Good Info Zoson.

LinX I don't have that benchmark downloaded. As has been my theme today I'm reaching my saturation point on all the benchmarks and stress tests. I mean I can only run FryBench, IBT, Prime, Cinebench, 3DMark, Pi, AIDA stress test etc so many times. I get the idea of being stable. My best test is to just use my PC. Gaming is about as stressing as it gets for me as I don't render images with any frequency or regularity. So if I can run BF4, Far Cry 3 or Crysis 3 for an hour or two I'm content. I'm not a professional benchmarker after all. Now that I have SpeedStep on I'm probably typing this at 2.4Ghz
Again not dismissing the mild need for some stress testing for the sake of one's OC but I'm not obsessive about it.

EDIT Temps you asked about. At 4.6Ghz with 1.425v vcore I was hitting low 80C on stress tests

 

[Zoson]

Good Info Zoson.

LinX I don't have that benchmark downloaded. As has been my theme today I'm reaching my saturation point on all the benchmarks and stress tests. I mean I can only run FryBench, IBT, Prime, Cinebench, 3DMark, Pi, AIDA stress test etc so many times. I get the idea of being stable. My best test is to just use my PC. Gaming is about as stressing as it gets for me as I don't render images with any frequency or regularity. So if I can run BF4, Far Cry 3 or Crysis 3 for an hour or two I'm content. I'm not a professional benchmarker after all. Now that I have SpeedStep on I'm probably typing this at 2.4Ghz
Again not dismissing the mild need for some stress testing for the sake of one's OC but I'm not obsessive about it.

EDIT Temps you asked about. At 4.6Ghz with 1.425v vcore I was hitting low 80C on stress tests

IBT and LinX are almost the same thing. They're both frontend GUIs written for Intels Linpack Math Kernel Library. In both cases, though, you need to manually update the linpack binaries to make sure that you're stressing correctly, and AFAIK, IBT never updates them for distribution.

Here's the direct link:
https://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download

You simply download the windows zip, and then replace the corresponding files in the IBT folder(same with LinX). Also, If you have 12GB or more memory and you have not disabled HT, test with a problem size of 9126 or you'll get improper loading due to Hyperthreading overload.

Basically, IBT/LinX correctly updated will produce 'worst case scenario' temps on your cpu.

 

[GonzoP]

IBT and LinX are almost the same thing. They're both frontend GUIs written for Intels Linpack Math Kernel Library. In both cases, though, you need to manually update the linpack binaries to make sure that you're stressing correctly, and AFAIK, IBT never updates them for distribution.

Here's the direct link:
https://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download

You simply download the windows zip, and then replace the corresponding files in the IBT folder(same with LinX). Also, If you have 12GB or more memory and you have not disabled HT, test with a problem size of 9126 or you'll get improper loading due to Hyperthreading overload.

Basically, IBT/LinX correctly updated will produce 'worst case scenario' temps on your cpu.

OK I guess I'll add this to the file. Thanks.
You are the king of the good OC links brother

 

[FrgMstr]

One ban. Just pointing out I was not joking. I am tired of having to mind this thread like a baby sitter.

 

[CrimsonKnight13]

How'd a thread about Xeon OCing have people who'd piss off our benevolent leader off? (sorry I missed the fun?)

 

[primetime]

How'd a thread about Xeon OCing have people who'd piss off our benevolent leader off? (sorry I missed the fun?)

Was kinda wondering myself...lol i must have missed a post or 2 idk....I didn't even know there was a problem.....yikes!

 

[Deimos]

@Deimos: Question about your ES chip. Does CPU-Z still show it as s B1 revision, or is it an earlier revision than retail releases? I always find ES chips interesting, although sometimes disappointing for OCing, as it sounds like with yours.

Yeah B1.

Also keep in mind that this chip has a 27 multi so I only need 166 BCLK to reach almost 4.5Ghz. But as I said, its a shitty overclocker

 

[primetime]

Originally Posted by Deimos View Post
It would be good if you guys could post the full spread of your voltage settings to achieve your overclocks.

[/URL][/IMG]

[/URL][/IMG]

[/URL][/IMG]

[/URL][/IMG]

Just posting these for a fyi.......voltages i uses are by no means 100% safe, but this pc passes any and all stress test no problem at no more than 80c temps....
It's really weird cause i have tried to lower all the recommended voltages that are considered unsafe but my pc just loves these settings.....even with my previous xeon almost like its specific to my mb to run correctly.....who knows?

 

[Cecil]

Easy does it everyone the sheriff has spoken.

Back on topic. I've been spending today truly seeing where my BCLK and RAM ceilings are. With my i7 it was between 210-215 and thus 1680-1720 ram at default 9-9-9-24 timings. I posted the other day my 4.8 shot with a 219 BCLK. So to take my ram out of the equation for the BCLK test I dropped my multi to 12 and set the ram at just 6x multi. Started at 219 and went by 2's. 221 check. 223 check. Tried 225 and it was very wonky, never tried 224 but whatever. With this Xeon I'm fully 10BCLK higher than my i7 920 D0. Just a better memory controller? I guess so.
Then I tested ram. Just ran 133x12 and set the ram multi to 10. Went to 180 BLCK for 1800Mhz and she seemed fine at 9-9-9-24 [remember I only have ddr3 1333 so I was pleased with 1800. 1850 would not happen. I could try relaxing the timing and probably could get up to 2000+ with some kinda 10-11-10-30 settings. But I've fiddled enough for one day. I'm done for now. Sitting here at 200x22 I am tempted to maybe try the looser timings and keep the 10x multi over the 8x. From what I've gathered 2000Mhz with 11-11-11-31-ish timings will generally beat at a 1600Mhz 9-9-9-24 set up

What is your PCIe setting? Once you get to 220-225BCLK, you need to set PCIe to 103, and slowly start increasing that as you raise BCLK more.

 

[GonzoP]

What is your PCIe setting? Once you get to 220-225BCLK, you need to set PCIe to 103, and slowly start increasing that as you raise BCLK more.

Ok that might explain things. I'm still at 100. I've heard of guys running 105 but never really knew this method behind the madness

 

[TyrantTru]

Its incredible how cheap you can get these chips..and the performance..wow.

 

[GonzoP]

Its incredible how cheap you can get these chips..and the performance..wow.

Yeah its the best deal in the cpu world right now by a landslide. $70 and my cpu will stand toe to toe with a 4770k at 1/5th the price. Its why some guys with say SB i5's are selling off their systems and "downgrading"

 

[Zoson]

*Primetime, Why are you clock skewing and changing your cpu diff amp? Both of those things can break your OC and only really become necessary when going to 5GHz+ speeds. They're intended to compensate for a degraded clock signal because of the clock rate being extremely high. You might want to try resetting those, as well as putting LLC on full control.

Here's an image showing what skewing your clock does:

You're literally moving your clocks out of sync. This could possibly explain why your cpu needs excessive voltage to be stable... You have to apply a higher bias to get the clock to active high in time since you're delaying so much. Also, cpus that have been electrically degraded will often just require more and more juice to be stable.

And some info on LLC:
LLC and You

Anyway, here are my bios shots.
Clocks:


Digi+:


Volts:


RAM:

 

[GonzoP]

Gonzo, Why are you clock skewing and changing your cpu diff amp? Both of those things can break your OC and only really become necessary when going to 5GHz+ speeds. You might want to try resetting those, as well as putting LLC on full control.

Um...I think you have me and primetime mixed up if you are referencing the BIOS screenshots further up this page

Having said that, once again I appreciate you sharing own insights.
Your BIOS is even more intricate than the one I have on my p6x58d but that stands to reason since the Rampage 3 is the king of the x58.

Yeah myself I have a lot of stuff on auto like the IOH/ICH. I've got LLC enabled which is all my BIOS allows [disabled/enabled]

I am using quite a bit less QPI than you and I see prime is way up at 1.4, yikes

 

[GonzoP]

Zoson as an education piece for me and likely someone else, can you remind me/us why clock skewing and cpu diff amp are employed? If memory serves I was using these features on my i7 and once I got over 4Ghz with her it did seem to help there.

 

[Zoson]

Zoson as an education piece for me and likely someone else, can you remind me/us why clock skewing and cpu diff amp are employed? If memory serves I was using these features on my i7 and once I got over 4Ghz with her it did seem to help there.

Yep, realized I should have gone more in depth, I updated that last post with some more information. Basically when you get to really high clockrates you can end up with a degraded clock signal that is 'sluggish' for lack of a better term, so you have to add a delay on the other clocks with a skew in order to get them to 'line up' correctly.

 

[GonzoP]

Yep, realized I should have gone more in depth, I updated that last post with some more information. Basically when you get to really high clockrates you can end up with a degraded clock signal that is 'sluggish' for lack of a better term, so you have to add a delay on the other clocks with a skew in order to get them to 'line up' correctly.

Gotcha yeah I have most things on auto in my BIOS quite frankly and most things turned on. Basically everything is on or on auto I think except for the Spread Spectrum stuff was always told that's not your friend when OCing

 

[novadaemon]

I just got my X5650 in.

4ghz IBT stable @ 1.2v! (20x200)
12gb (6x2GB) Kingston ECC DDR3-1333 CL8 @ DDR3-1200 CL7 (It drops a stick if I try to OC the ECC RAM to 1600mhz)
P6T Deluxe V2 Bios 1202
Zalman CNPS 10X Single Fan Air Cooler

Everything in my bios is set to Auto except CPU voltage is set to 1.2 and my RAM timings are tightened to 7-7-7. HT is ON, but Turbo is OFF. I'd love to get 4.4 stable but I don't want to increase power consumption too much to do it. I don't think it's worth it to go from 4ghz @ 1.2 to 4.4 at 1.4 if that's going to be what it takes. Anyone have any tips to squeeze out more performance?

My RAM is only 1200mhz, but my uncore is 3600mhz. Is that ok? I don't see any instability so far. Would that hinder me from getting to 4.4?

Also, the ECC ram seems fine at 1600mhz except for the problem where it drops a stick. Any tips to fix that? I'm running 6 DIMMS.

 

[Zoson]

Sticks dropping out is solved by raising QPI/UCLK/Vtt voltage. Don't exceed 1.35v real for extended periods of time.
QPI Speed is NOT the same as UCLK speed. They're usually different.
UCLK speed should be 2x your memory speed *or greater* for best performance. So it's not a problem to run a UCLK speed higher than 2x your memory speed, there just won't be a whole lot of performance gain.

Most Uncores won't go much faster than about 3.6GHz to 3.8GHz at 1.35v though, limiting the speed at which memory is effective to between 1800 and 1900MHz depending on your individual CPU.

 

[GonzoP]

I just got my X5650 in.

4ghz IBT stable @ 1.2v! (20x200)
12gb (6x2GB) Kingston ECC DDR3-1333 CL8 @ DDR3-1200 CL7 (It drops a stick if I try to OC the ECC RAM to 1600mhz)
P6T Deluxe V2 Bios 1202
Zalman CNPS 10X Single Fan Air Cooler

Everything in my bios is set to Auto except CPU voltage is set to 1.2 and my RAM timings are tightened to 7-7-7. HT is ON, but Turbo is OFF. I'd love to get 4.4 stable but I don't want to increase power consumption too much to do it. I don't think it's worth it to go from 4ghz @ 1.2 to 4.4 at 1.4 if that's going to be what it takes. Anyone have any tips to squeeze out more performance?

My RAM is only 1200mhz, but my uncore is 3600mhz. Is that ok? I don't see any instability so far. Would that hinder me from getting to 4.4?

Also, the ECC ram seems fine at 1600mhz except for the problem where it drops a stick. Any tips to fix that? I'm running 6 DIMMS.

CL7 is probably what's hurting your ram at 1600 that's awfully tight for 1600 try 8 and then 9 if you need. I think across the boards CL9@1600>CL7@1200.
You can likely pull off 4.4 with a bit less than 1.4v. I'm at 1.36v myself. If you have a personal vcore limit just take it to that and settle there. Surely you're good for at least 4.2 on maybe 1.25-1.275ish

 

[Zoson]

BTW, forgot to address the ICH/IOH voltage. Mine are set to 1.16v because I'm running SLI.

There are two scenarios where you would want to increase ICH/IOH:
1. You're running more than one high power thing in your PCI slots, for example, two gpus.
2. You're taking your QPI link speed up above about 7800MHz. QPI links the CPU with the IOH, so increasing BCLK will increase QPI, and cause more stress on the IOH.

 

[novadaemon]

Sticks dropping out is solved by raising QPI/UCLK/Vtt voltage. Don't exceed 1.35v real for extended periods of time.
QPI Speed is NOT the same as UCLK speed. They're usually different.
UCLK speed should be 2x your memory speed *or greater* for best performance. So it's not a problem to run a UCLK speed higher than 2x your memory speed, there just won't be a whole lot of performance gain.

Most Uncores won't go much faster than about 3.6GHz to 3.8GHz at 1.35v though, limiting the speed at which memory is effective to between 1800 and 1900MHz depending on your individual CPU.

Thanks for the reply. My QPI and UCLK are BOTH at 3600mhz. They are technically set to Auto but the motherboard automatically readjusts QPI to 7200MT/s and Uncore to 3600mhz. I just didn't bother to post a screenshot of my memory tab showing the NB frequency at 3600mhz.

CL7 is probably what's hurting your ram at 1600 that's awfully tight for 1600 try 8 and then 9 if you need. I think across the boards CL9@1600>CL7@1200.
You can likely pull off 4.4 with a bit less than 1.4v. I'm at 1.36v myself. If you have a personal vcore limit just take it to that and settle there. Surely you're good for at least 4.2 on maybe 1.25-1.275ish

Thanks for the reply too. When I'm running at 1600 my RAM timings are on auto which is 8-8-8. I could try loosening them to 9-9-9, but I somehow doubt that that's going to make the stick magically appear. I will try it when I get some time though.

I'll see what I can do about 4.4. If I can't get there with less than 1.35v I might settle for 4.2.

 

[Zoson]

Thanks for the reply. My QPI and UCLK are BOTH at 3600mhz. They are technically set to Auto but the motherboard automatically readjusts QPI to 7200MT/s and Uncore to 3600mhz. I just didn't bother to post a screenshot of my memory tab showing the NB frequency at 3600mhz.



Thanks for the reply too. When I'm running at 1600 my RAM timings are on auto which is 8-8-8. I could try loosening them to 9-9-9, but I somehow doubt that that's going to make the stick magically appear. I will try it when I get some time though.

I'll see what I can do about 4.4. If I can't get there with less than 1.35v I might settle for 4.2.

Explanation as to why a stick would fail to appear:
All x58 based CPUs suffer from a design symptom called 'IMC Clock drift' where because you have three separate memory channels, when your board posts, it has to attempt to initialize all three.
Clock Drift

The whole thing is *every* board has different length traces connecting the memory sockets to the cpu, and intel had to set a hard limit on 'i'll wait for this channel to respond for x amount of time'. So you end up with situations where you need to drive a higher voltage through the IMC to make it 'wake up' more quickly on the channel that's dropping out.

This is a common x58 issue, and it was way worse with 45nm CPUs even. You can solve it by raising your QPI/UCLK/Vtt voltage - just don't exceed 1.35v.

You can actually see this in both mine and primetime's bios shots, it's called "Round Trip Latency" and you can see on my shot where the R3E reports there's exactly 1ns difference between my 1st and 2nd memory channels, and then even another 1ns difference between my 2nd and 3rd memory channels.

Then there's even yet another hidden setting called tRL that was only exposed in the 0003 Rampage 3 Extreme bios - however later on they added 'memory recheck' which forces your CPU to scan the memory channels twice, which in almost 100% of all cases will be a long enough duration for a missing channel to wake up and they removed tRL.

 

[BigTroll]

I just got mine in, I got the Asrock X58 extreme, was running an i7 920 at 3.6, got a x5650 for $69 on ebay running it at 4.0ghz 182 bclk and 20x multi, I did have to lower uncore to the lowest to be stable, however it does 8 hours of linpack occt just fine and i did 24 hours of large fft on occt and it passed fine. running around 1.26 volts.

love this upgrade!

 

[Zoson]

I just got mine in, I got the Asrock X58 extreme, was running an i7 920 at 3.6, got a x5650 for $69 on ebay running it at 4.0ghz 182 bclk and 20x multi, I did have to lower uncore to the lowest to be stable, however it does 8 hours of linpack occt just fine and i did 24 hours of large fft on occt and it passed fine. running around 1.26 volts.

love this upgrade!

You mean 22x multi? 182*20 = 3640

 

[BigTroll]

oopps still thinking of the i7 yeah running the 22x multi on it with no problems

 

[primetime]

*Primetime, Why are you clock skewing and changing your cpu diff amp? Both of those things can break your OC and only really become necessary when going to 5GHz+ speeds. They're intended to compensate for a degraded clock signal because of the clock rate being extremely high. You might want to try resetting those, as well as putting LLC on full control.

Come to think of it, it was just something i probably picked up from a very old article and i can try them on auto and see how that goes, but yea at my current clock speed it probably wouldn't make a difference....i guess it did make more sense when i was shooting for 5ghz in the beginning except without high end water cooling it wasn't going to happen anyway (i was personally seeing to high of temps at 4.6) So yea i can put them back on auto for time being

 

[GonzoP]

Come to think of it, it was just something i probably picked up from a very old article and i can try them on auto and see how that goes, but yea at my current clock speed it probably wouldn't make a difference....i guess it did make more sense when i was shooting for 5ghz in the beginning except without high end water cooling it wasn't going to happen anyway (i was personally seeing to high of temps at 4.6) So yea i can put them back on auto for time being

If you were like me you probably had it on back from your i7 days. Maybe just a habit.


Certainly is nice to see so many new guys riding the wave

 

[primetime]

If you were like me you probably had it on back from your i7 days. Maybe just a habit.

you are 100% correct once i thought about it

Edit: I made the changes that was recommended and figured i would see how sky diver bench compares and it was the highest yet, so so far looks to possibly increase performance, and i can't complain with that ...so far it seems a good tip

[/URL][/IMG]

i think these were the settings you said needed changed right?

[/URL][/IMG]

 

[Zoson]

Yes, those are the ones. And now that you're no longer skewing your cpu clock out of sync, you might want to try reducing PLL again. PLL is the voltage that drives the 'phase locked loop' which is the oscillator that generates the cpu clock.

Increasing PLL makes the waveform more 'perfect' but as mentioned, and noted in intels specsheet, you really don't want to exceed 1.9v.

 

[Ultra-m-a-n]

I got my x5650 up and running on my ex58-ud3r, but I have noooooo idea how to over clock this thing

This is my first run at x58!

Any tips where to start overclocking this?

 

[GonzoP]

I got my x5650 up and running on my ex58-ud3r, but I have noooooo idea how to over clock this thing

This is my first run at x58!

Any tips where to start overclocking this?

Ultra I haven't seen once which failed to hit 4GHz. Not saying you start there though. If memory serves I believe I started at 133x22 just to make sure it wasn't dead. I messed around in windows for maybe 5 minutes and then restarted to 166x22. I initially started the volts at like 1.125 and then moved to 1.175v for 166x22 then I picked whatever would give me an even 4Ghz which is what 182 moved up to 1.225v. After that its a bit more delicate balancing the ram, the BCLK and the voltages but nothing major. 95% says if you actually on first boot in your BIOS set 182x22 on say 1.225v you'd just sail right on in. Then you can play the game of raising clockspeed while dialing in your voltages. For 4Ghz the range I've seen is anywhere from less than 1.2v to more than 1.25v. Some guys go with more and reduce as they can. I usually do the opposite, use as little as possible and if I'm unstable then I kick it up a notch or two. No best way really. Most of these chips end up in the same ballpark.

 

[GonzoP]

Yes, those are the ones. And now that you're no longer skewing your cpu clock out of sync, you might want to try reducing PLL again. PLL is the voltage that drives the 'phase locked loop' which is the oscillator that generates the cpu clock.

Increasing PLL makes the waveform more 'perfect' but as mentioned, and noted in intels specsheet, you really don't want to exceed 1.9v.

I've read somewhere [can't remember to find the link] that more PLL isn't always a stability cure all and that sometimes reducing PLL will improve an OC. I know on my mobo 1.8v is the lowest value possible but I wanna say on other boards guys have been going below that.

 

[GonzoP]

you are 100% correct once i thought about it

Edit: I made the changes that was recommended and figured i would see how sky diver bench compares and it was the highest yet, so so far looks to possibly increase performance, and i can't complain with that ...so far it seems a good tip

[/URL][/IMG]

i think these were the settings you said needed changed right?
[URL=http://s1362.photobucket.com/user/Sean_Adkins/media/IMG_20140811_014346353_HDR_zps02bf8e32.jpg.html][/QUOTE]
Pretty Good Prime, good enough for the #2 spot amongst [H] Xeon +7970's :p
[IMG]https://lh5.googleusercontent.com/-kcQ8YxU0LLA/U-jsMbksaPI/AAAAAAAAGeM/dajVyeaX-hA/w1173-h143-no/SkyDiver.png
Just kidding buddy.

I'm going to guess based on your Physics score you did that test at 4.2? I know I have a slightly higher OC on the 7970 than you as well. What baffles me is your combined score beats mine
Granted this is a test I ran over a week ago when I was still running the synthetic tests so I almost don't care. I just find it off that if my cpu is 4.4 to your 4.2 I score better in physics and my gpu OC is higher than yours I score better in graphics, how I am losing the combined?
Again its a moot point since I'm not one who lives and dies by my 3DMark score; just find it highly curious.

 

[Zoson]

I've read somewhere [can't remember to find the link] that more PLL isn't always a stability cure all and that sometimes reducing PLL will improve an OC. I know on my mobo 1.8v is the lowest value possible but I wanna say on other boards guys have been going below that.

Right, on the boards that allow you to reduce PLL below 1.8, a LOT of people find that their CPUs run better between 1.2 and 1.4v PLL.

My main concern, for *primetime, though, is he's WAY over the 'absolute max' specified by Intel: they say 1.89v, I've been telling you guys 1.9v, and he's at 2.1v!!!

I'm also suggesting that his broken skew and amplitude settings may have been the entire reason why he needed to raise PLL in the first place at all.

One picosecond is 0.000 000 000 001 of a second - 11 zeroes.
So we can express a CPU clock delay of 800ps, as .000 000 000 800 seconds.
Running at a clockrate of 4400MHz that's 4 400 000 000 calculations per second.
Said another way, every clock would take .000 000 000 2 seconds(200ps).

So, primetime was basically skewing his clock FOUR clock cycles worth of delay, then attempting to make up for it by pumping the PLL full of voltage so it could reach active high as fast as possible. I think it's easy to see how this can cause enormous synchronization/stability problems at the end of the day. I never found skew or diff amp useful except above 5GHz, up until then it only caused problems.

 

[GonzoP]

Right, on the boards that allow you to reduce PLL below 1.8, a LOT of people find that their CPUs run better between 1.2 and 1.4v PLL.

My main concern, for you, though, is you're WAY over the 'absolute max' specified by Intel: they say 1.89v, I've been telling you 1.9v, and you're at 2.1v!!!

I'm also suggesting that your broken skew and amplitude settings may have been the entire reason why you needed to raise your PLL in the first place at all.

One picosecond is 0.000 000 000 001 of a second - 11 zeroes.
You were delaying your CPU clock by 800ps, or .000 000 000 800.
If you're running at a clockrate of 4400MHz that's 4 400 000 000 calculations per second.
Said another way, every clock of your cpu would take .000 000 000 2 seconds(200ps).

You were basically skewing your clock FOUR clock cycles worth of delay, then attempting to make up for it by pumping your PLL full of voltage so it could reach active high as fast as possible. I think you can see how this can cause enormous synchronization/stability problems at the end of the day. I never found skew or diff amp useful except above 5GHz, up until then it only caused problems.

Zoson you're mixing me and primetime up again. He's the one with 2.1v PLL I'm the one at 1.8v.
Prime's the one who had the camera taken BIOS shots.
My shots were the AIDA screens like this one

And as you can see my PLL is in fact 1.8v and my QPI is just 1.3v

No worries though you have earned more than one get out of jail free card with all your helpful educating.

 

[Zoson]

Ugh, right. Yeah, I'm worried about his CPU. He's running voltages real close to what Nate was running when he fried his chip.