5960x 4.8 @ 1.3v / 5.0 @ 1.35v (not kidding)

[cekim]

Never hurts to go over your system with a fine-tooth comb... I found my PWM EK pump voltages were way off as I had tuned the wrong thing in the BIOS (I had hooked up the pump control to 4 and the fans for the radiator to 1 and then tuned 1 as if it were the pump).

That wasn't the only issue I found, nor can it fully explain this. Short run-down of issues and possible impact:
- water pump was nearly off/off at any temp short of 75C (and still remarkably effective - but temps are much narrower core-to-core now)

- all fans were running harder than they needed to be (there are 12 fans, 3 140mm, 3 200mm), meaning they were drawing current and inserting noise I am sure through the MB

- random PCIe conflicts with M.2 and 10GbE card (PCIe lanes all busy - SLI x16x2 + 10GbE + m.2). Solved by using hyper card instead of m.2

So, with that all out of the way:
1. ~10 minutes of aida64 at 4.8 @ 1.3v (it eventually found instability, but look at those temps?)
2. 5GHz club :cool:

[Qwinn]

Hmmm. Ok. So out of curiosity, what are your vcache, vinput and vccsa for your 1.17 4.5? LLC? Any other settings you think are working particularly well for you?

[cekim]

Qwinn wrote:
Hmmm. Ok. So out of curiosity, what are your vcache, vinput and vccsa for your 1.17 4.5? LLC? Any other settings you think are working particularly well for you?

Will update in a bit. I am investigating some linux issues at the moment.

[cekim]

Still testing and trying to eliminate variables in linux (windows tends to take longer to show issues. Linux finds them right away, but my issues might be dual GTX980s and poor support for same in linux by nvidia).

But here's where I am right at the moment as an example:
CPU: 4.5GHz
Cache: 4.0GHz
VCore: 1.118v
VCache: 1.1v
SA: Auto
LLC: Auto
DDR: 2800 (128G @ 1.35v)

1. I didn't try to tune down VCache - I've run this same CPU at 5.0/4.5 @ VCache 1.25v, so I suspect it may be able to do better. Linux is rough on the cache (see above - it will find cache instability much faster than windows for reasons I cannot fully explain).
2. Those are BIOS commanded values - I saw, but did not fully debug aida64 reporting slightly higher voltages with these same settings in windows.
3. I may need to bump them to declare 24/7 stability despite 16+ hours of Aida64 earlier because I need it to work in linux or its useless to me.

I'm going to keep working to get this machine to 24/7 status (since I need to get back to work shortly). I'll update this thread with the final settings of this and my other 5960x once I think it is as good as I am going to get it.

[Martsmac1982]

That is incredible, my 5960x seems to want no less than vcore 1.35v to hit 4.455 and even then it's still not stable with 64gb DDR4!

[Gobe]

I have to say that cekim's 1.118 Vcore at 4.5 GHz is nothing short of astounding.

I'm beginning to think that there's no telling with the I7 5960X. I have two of them and one is cherry-picked and is able to run individual cores easily at 4.9 GHz with the other 7 cores disabled, and a few of the cores can individually pull off 5 GHZ. The other CPU runs individual cores between 4.7 and 4.9 GHz (all under regular custom loop). From these numbers, you would think that the cherry-picked 5960X would be hands down a better overclocker. Yet, when I run RB H.264 Handbrake with the radiator in an ice bucket, I can get the "regular" 5960X up to around 4.86 GHz and the cherry-picked 5960X up to 4.90 GHz. Not totally stable for either, might get ten cycles on the bench before it crashes, might not get a single cycle run. Not much difference in the end between the $900 off-the-shelf CPU from Microcenter and the $1200 cherry-picked CPU. The cherry-picked 5960X can match core and cache up to about 4.8 GHz and the off-the-shelf 5960X needs to drop cache a couple hundred MHz, so I guess there is that difference.

Of course, I'm not done looking for sweet-spots in the various settings.

[cekim]

Gobe wrote:
I have to say that cekim's 1.118 Vcore at 4.5 GHz is nothing short of astounding.

I'm beginning to think that there's no telling with the I7 5960X. I have two of them and one is cherry-picked and is able to run individual cores easily at 4.9 GHz with the other 7 cores disabled, and a few of the cores can individually pull off 5 GHZ. The other CPU runs individual cores between 4.7 and 4.9 GHz (all under regular custom loop). From these numbers, you would think that the cherry-picked 5960X would be hands down a better overclocker. Yet, when I run RB H.264 Handbrake with the radiator in an ice bucket, I can get the "regular" 5960X up to around 4.86 GHz and the cherry-picked 5960X up to 4.90 GHz. Not totally stable for either, might get ten cycles on the bench before it crashes, might not get a single cycle run. Not much difference in the end between the $900 off-the-shelf CPU from Microcenter and the $1200 cherry-picked CPU. The cherry-picked 5960X can match core and cache up to about 4.8 GHz and the off-the-shelf 5960X needs to drop cache a couple hundred MHz, so I guess there is that difference.

Of course, I'm not done looking for sweet-spots in the various settings.

With some more time in linux, I am seeing the old "there is windows stable and then there is linux stable". Specifically with this chip that will run happily in windows up to 4.8/4.6 (core/cache) and run less happily to 5.0/4.6 but indefinitely without load, linux + speedstep + adaptive has proven to be much more elusive.

Again and again though, what I find is that 100% load for hours at a time appears to be this chip's "easy". The hard part is all the gymnastics of speed-step.

Also not surprisingly, the cache appears to play a critical role here (not surprising given the number of gates/area it represents - it is massive).

I am fairly confident that 1.18 is NOT going to be sufficient in linux for 24/7 stability despite aida64 giving me the thumbs up (16+ hours)

My other chip with an AIO cooler (so, I can't push it as hard - but it lives in a cool/dry basement normally), is stable in Aida at 4.5/4.5 1.2/1.3 so far. Haven't tried linux yet...

[Martsmac1982]

Forgive my ignorance but what on earth are you guys using to cool your chips at 4.5+Ghz at such low voltages as you are using??? (my clocks wont stay stable above 4.4Ghz with anything less than 1.36v vcore which is ridiculous compared to 1.18v ish, i mean thats not even in the range of "I just got lucky in the silicon lottery with my chip" 1.18v at over 4Ghz is mental.

[cekim]

Martsmac1982 wrote:
Forgive my ignorance but what on earth are you guys using to cool your chips at 4.5+Ghz at such low voltages as you are using??? (my clocks wont stay stable above 4.4Ghz with anything less than 1.36v vcore which is ridiculous compared to 1.18v ish, i mean thats not even in the range of "I just got lucky in the silicon lottery with my chip" 1.18v at over 4Ghz is mental.

I don't think it is an issue of cooling, but frankly, your guess is as good as mine. Could just be processor batch or dumb luck. I started from the assumption of my 4970k haswell - set to 1.3v to do anything fun and maybe shave a little off after tuning. So, I added cooling capacity and voltage like everyone else...

This odyssey of low voltages started with:
1. I got a 5930k that hit 4.6 effortlessly at 1.35 (the as rock auto OC 4.6 that I hit accidentally). I then found that chip would run to 5.1 at 1.37-1.38 (though far from stable)

2. then captflippy posted his 1.117 4.5GHz and I had some time to play around this past week, so play I did.

Cooling on this thing is an EK mono-block plus a 420 (3x140) rad (aka: lots, but no ice or phase change).

[Martsmac1982]

Yeah thats what amazes me so much. I run with 3 540mm Rads and 1 pump in a single loop (although It shares the loop with 4x Titan X GPU's) and my temps (and clocks) are pathetic compared with yours on my 5960X back to the drawing board with my OC, good work though sir! Kudos to you!

[cekim]

Martsmac1982 wrote:
Yeah thats what amazes me so much. I run with 3 540mm Rads and 1 pump in a single loop (although It shares the loop with 4x Titan X GPU's) and my temps (and clocks) are pathetic compared with yours on my 5960X back to the drawing board with my OC, good work though sir! Kudos to you!

The chip will do what the chip will do, my only contribution is many hours figuring out what that is. Most of them were fun/productive and do keep in mind I am still hunting for something I'd call rock-solid. Windows is happy and I can't crash it, but linux + speedstep + cache are giving me trouble.

I have discovered that linux is doing "its own thing" WRT speed-step. That is, if I turn off speed-step and c-states in the bios, linux is still busy tuning the clocks. The only way I stop that entirely is intel_pstate=disable at boot-time which gives me only all-or-nothing control of the clocks (min or max, no auto).

That (all or nothing) may actually be why windows is stable though, so at the end of the day, that might be the final answer is good temps and voltage + no stepping for max stability. I don't like that answer - I WANT IT ALL!!!! 🙂