Clevo 2020

I just keep it at 5Ghz, I can apply a 120mv undervolt and keep it stable.

 

Above 5Ghz you also need to do override voltages to get anything stable/sane.

 

Do you find that when you use an override voltage at >5 ghz, you wind up fully stable in Cinebench R20 or Prime95 or AIDA64, but if you load up Battlefield 5, you get a CPU Cache L0 error or a BSOD?

 

I've had to deal with that on several occasions, I think it's due to IMC being hit with more load than R15/R20 etc.

 

Oh no it's not.
It's because AC Loadline is controlling voltages.

Vcore is this:

Vcore=vCPU + (ACLL * dI) - (VRM LL * I) + vOffset.

ACLL=AC Loadline in milliohms (this is usually set to the same value as base VRM Loadline)
VRM Loadline=Loadline Calibration (LLC) in mOhms (this is locked at 1.6 mOhms on 8 core, 2.1 mOhms at 4 and 6 core). Desktops allow you to reduce this setting (reducing vdroop, via "Loadline calibration".

dI=d1-d0. (I don't know what this is. it has something with amps. This formula came from Elmor).
I=Current
vCPU=base VID (@ AC Loadline=0.01 mOhms, DC Loadline=0.01 mOhms). This value will also change depending on temps. At highest turbo, it is -1.5mv/-1C starting at 100C.

AC Loadline is very good at setting voltages during balanced synthetic loads, AVX or Non-AVX (Cinebench R20, Prime95, AIDA64 Stress FPU, etc).
However during mixed loads (Battlefield 5, especially on loading screen, Realbench 2.56, etc), ACLL does not seem to be getting the current load properly, then load voltage either becomes too large (too low vcore=crash) or too high. It seems ACLL cannot react to violent changes in current quickly (to raise or lower operating voltages).

The only way to fix this behavior is to flatten ACLL by setting it to 0.01 mOhms. On desktop boards, you can then increase the LLC to reduce vdroop (Lower mOhms on LLC=less vdroop). On laptops, you will need to increase override voltage or increase vOffset if you reduce ACLL. This is the only way to fix the voltage swings. But this would require a very high idle voltage (up to 1.52v if ACLL is 0.01 mOhms).

 

I managed to bench on adaptive up to around 5.2. It did need the secondary power limits tweaking a lot though.

 

I just raised the voltage by 5mV with the static override, until this was no longer the case. Since with an offset it kept happening.

5.2 with an undervolt, man what were the temps?

 

Override voltage changes the VID (vCPU). Gotta remember this.
I am not sure at what point on the circuit offset is calculated, but I don't think it's added to vCPU but added after AC Loadline does its work, after loadline calibration drops the voltage down.

 

One of the reasons we go by temp and not VID, as you saw back when I tested 1.6 mOhm VR LL vs hw stock 2.1 mOhm.
Going to 5.1Ghz is still within the thermal budget, even 5.2Ghz with cold air. However, for daily and longevity 5Ghz is hard to beat. R0 does run hotter though, but direct die VC mod should offset this.

 

Guys,
These new clevos with 10900k and 2080 super, how do they correlate with a 9900k and 2080? Is there any early tests? I am sitting with my beloved 870km1, I am pretty much interested in it is the time for upgrade or not

 

https://twitter.com/_rogame/status/1259148403226873857

This is the Dellaware R2, no numbers from X170 yet. But this gives you an idea of what to expect, physics are around what 9900Kx scores at 5.2Ghz.

 

omg!

Where did you get your direct die copper place? What sizes are required to do direct die?

 

Original mod by @ole!!!
http://forum.notebookreview.com/threads/p870tm-2nd-cpu-fan-port.812706/page-10#post-10993736

My version, direct die no ILM.
https://imgur.com/a/gyMIBLI

56x56x3mm vapor chamber
https://www.quick-cool-shop.de/en/heatdiffuser/heatdiffuser-quadratic/heat-diffuser-quadratic.html

 

Is this compatible with the P751? Seems to be a small squarish part by dimensions.

 

I wish I could tell you a definite yes or no, but unfortunately it's not as easy. Since you need to measure it on your machine.

If the mounting holes on the P751 CPU heatsink are >56mm across, it should fit. It might not be perfectly centered, depending how the vrm cutout is placed. You definitely want to measure before committing to this mod. VC also adds 3mm to the total package height, which is why you need to run direct die. Since it effectively replaces the IHS and we remove the ILM.
If it's a unified heatsink design, you want to be prepared for contact issues for GPU that way incur from this. Ie. Needing a shim for GPU die.

My advice would be to measure and make a mock-up VC from paper/cardboard, with the dimensions slightly greater than 56.5 x 56.5mm, and go from there.

 

At 5.2 it was at stock IIRC but that was still plenty hot.

 

Voltages above 5Ghz also begin to get out of hand, 5.1Ghz is cool and all.

But running 70c at 5Ghz is kinda more neat to me.

 

I would like to have your cpu and heatsink 5ghz in a laptop way more than good.

 

But brother, you are more than capable of doing the same mod! I remember the work you put into the P870KM1-G upgrade

And my KF is not a binned sample, it's average at best, doing 5Ghz at 1.22v @iunlock has two KS that do 5Ghz at 1.14v.
Here's a typical KS vs K

 

My 9900k can do 5.1 1.34v no cache ratios AVX 0 and pass my sig criteria. 5.2 requires 1.385 AVX 1. I'll have to go back and retest with the better cooling.

CB15 requires 1.263 @ 5ghz no offsets (AVX obviously doesn't matter )

The KS vs K shows KS 4.7@ 1.076, yet I haven't found one that can pass CB20 and RB2.56 Stress 4.5ghz @ 1.1. I'm sure they're out there, but looking over my KS request runs of other owners/sellers, so far I'm 0/18 at last count over the last few months.

CB20 and RB2.56 (both AVX enabled) really pull the watts/heat.

What was the stress test criteria for the KS vs K?

How does your KF fair @ 5ghz with CB20 and RB2.56?

 

My high end heads towards that but my 5Ghz level is better which is why I stay around there.

 

9900K is temp sensitive, in the past Luumi talked about how keeping it around 60c max nets best OC and voltage requirements. I've dabbled with 5.1Ghz for R15 and that extra 100Mhz comes at almost 10c cost, ~70>80c.
For gaming it will be less, but you begin to lose efficiency due to power leakage. And since my temps depend on ambient being within reason, 5Ghz leaves enough thermal room for summer temperatures. But it sure was fun to bench in December, 5.2Ghz at 62c Firestrike, lol.

KS voltages, binning within a binned CPU? And yeah, R20 is much more punishing, due to how long it takes to finish on 8 cores.
From what I recall, that screenshot was tested with an AVX load, I want to say Blender. link to vid

R20 KF at 5Ghz in R20 ~90c - added screens of core spread as well.
https://imgur.com/a/lEuuEjd

 

Thx for the link, I am ordering the peltier immediately How can you solder to the heatsink? I have also a delid guard (derdauer I think). If I am not mistaken with 9900KS I have better changes for the 5ghz... if 10th gen comes out I think I can find for a reasonable price.

 

You mean vapor chamber? The 56x56x3mm one, its what @ole!!! and I used for the mod.

 

....very...very...interesting...hmm...I recognize those KS points...

 

I think i asked you before, but is that BIOS set voltage or LOAD voltage (die-sense)?
Because 4.5 ghz @ 1.1v load voltage is child's play. My 9900k can do that. I mean hell--my 9900k can run FMA3 small FFT Prime95 @ 4.7 ghz @ 1.1v (LOAD voltage). Now 1.1v bios voltage is a completely different story because you have vdroop to contend with at full load.

 

How do I check if avx is enabled on my system? I saw that to disable avx instructions on windows I'd have to run "bcdedit set xsavedisable 1" but have no idea how to check it? Thanks

 

Run CBR-20. With avx disabled you should see lower scores.

 

I think I posted a response to that in Clevo Overclockers (or another forum) with a screen cap? I wanted you to take a look to see if you saw or recommended anything of note?

1.1v BIOS voltage.

Under load HWInfo reports ~1.040 - 1.048 for CB20 and 1.048 RB2.56 Stress since I'm seeking AVX workloads too.

My (current) objective is to set override in BIOS on the P870TM1 to 1.1 @ 4.5 baseline to draw ~ 110w under load to match conditions I reproduced on my 8086k to draw ~ 110w because that is the ideal noise level for me.
I'll probably use @jc_denton excellent groundwork for AC / offset tests to dial it in if possible to lower the overall pull even more.

I'll end up testing 4.6 sometime over the next few days @ 1.1, but I fully expect it to plonk.

If I'm benching, it can be as noisy as it wants to be, but for day to day use, much quieter fans really keep me sane especially (pre-covid) being site to site working in quiet environments.

8086k @ 4.5 all cores draws ~83w = ultra quiet.
8086k @ 4.8+ all cores draws ~110w = quasi quiet for gaming and ultra quiet for everything else day to day usage. This is my new norm and I run my 8086k @ 4.9ghz and it provides the perfect noise level since the BartX upgrade.
I can't recall what it draws @ 5ghz+ but it was noisy enough to scurry back down to 4.9 for noise management.

I delidded my 9900k last week and installed a BartX 4.3mm IHS w/ Silver King and now it is running cooler than before and able (in my desktop) to do 5.1 AVX 0 (Before I was using AVX 1) and 5.2 AVX 1 (I haven't tested AVX 0 yet) on air
atm w/ a Noctua N15 and Gelid Extreme. I'm not using anything extreme or water based atm to reflect air based cooling with fixed ambient temps.

I was going to hold off till I get a few vials of EVGA's new Frozen 2 that are on their way this Thursday to test to see how it compares (and wrap up a few laptop upgrades/repairs) and
once the air clears slide it into the P870TM1 and start working on it (Put the 8086k in my desktop and re-bin it with the BartX it has too vs. the previous stock IHS delid). Right now, my daughter is staring daggers to repair, upgrade and tweak her P377SM-A.

Let me know if you see or recommend anything. I always enjoy your power draw responses.

 

I mean, I wouldn't say no to a killer KS.

 

Be sure to read @ole!!! s original mod, since the direct die frame needs to be modified, as the are SMDs near the socket that make mounting it impossible.

KS is cool and all, but a good P0 scores a bin above it at same frequency.

 

Thx I will. Could you also give me some hints how to solder the vapor chamber square to the TM1 heatsink? What temparature and with what (bismuth)?
thx

regarding the delid guard, we will cut off the tricky parts with dremel

 

mr. fox's original video for soldering shim.

 

Thanks a lot!
which is the max temp where I need to be very cautios not to damage the heatsink? (both tm1 and vapor)
thx!!!

 

Well I can do 4.7 @ 1.021v load cinebench R20 and realbench 2.56 on a certain platform but I haven't tested 4.5 and I assume that's still too high for you?

 

solder paste in video melting point is around 140C i think, but its the same cheap solder paste used by clevo. if you have TM1 heatsink then becareful with the added fins that didn't exist on pre TM heatsinks, those fins are cheaply soldered and will fall off so I'd recommend hold it down with a clamp

 

I think we both know 4.7 @ 1.021v rocks. I would be interested to see how low you can set BIOS voltage (sub 1.1) and your total pull @ LLC 6 traditional delid on air w/ no LM. Maybe see how under ~103w (RB 2.56 Stress 15 run) and 113w (CB20) you can get total draw? Depending on your numbers and exact setup (no esoteric custom water, direct die, etc...), maybe you would be interested in selling? You may have that chip I'm looking for.

 

I am under NDA for that information. This was just a quick test. Can't even tell you what SKU it is. But my 9900k has a 1.080v wall at 4.7 ghz. Any lower will BSOD. This is for SSE2 instructions.

But I am surprised you said no one had a 9900KS that can do this. I saw one person with a delidded 9900KS that was running Prime95 FMA3 small FFT @ 5 ghz @ 1.078v (load). And many people (usually with Maximus XI Apex's) who were doing 1.128-1.140v load. And the Dark has even better transients...

Just sent a PM to two people I know have golden 9900K/S chips to ask what their 4.7 ghz voltage floor is. Assuming I found the right people.

x264 4.6 ghz 1.032v

https://imgur.com/XVr0Vb3

 

Were those in desktops or laptops though?

 

All in Desktops.

Note: Sheriff PM'd me and said that his BIOS can't even set a voltage below 1.1v.

Also as I mentioned before, you can't emulate setting 1.1v via "Override Mode" in a laptop with setting 1.1v on a desktop, with intel spec Loadline calibration being used, because (at 4.7 ghz, or even lower), the desktop will BSOD instantly due to Intel default vdroop being used (1.6 mOhms---for example a 138 amp load will cause the 1.1v to drop to 0.879v, (Math: 1100mv - ( 138 * 1.6 ); no 9900KS can run at that voltage at 4.5 ghz).

On a laptop, setting 1.1v via "override mode" will cause the AC Loadline to save you from a BSOD. That's the entire problem.
On a laptop, instead of having this formula:

1100mv - ( 138 * 1.6 ) = 0.879v, where 138 is amps and 1.6 is Intel default Loadline calibration,
You have this instead:

If CPU default VID is 1100mv:
vCPU=1100mv

Vcore = vCPU + (AC Loadline mOhms * Amps) - (LLC mOhms * Amps) + offset voltage
In this case, ACLL=1.6 mOhms, Amps=138 amps and LLC=1.6 mOhms.
vCPU will rise depending on temps due to Thermal velocity Boost (this depends on the CPU multiplier, explained here):
https://rog.asus.com/forum/showthread.php?106375-MCE-explanations-and-others&highlight=explanations

So even if we disable TVB so the VID is as if the CPU were at 100C, we get this at 138 amps:
Vcore = 1100 + (138 * 1.6) - (138 * 1.6) +0mv offset = 1100mv= 1.1v

So your desktop would be running at 0.879v while the laptop is running at 1.1v.

That's why 0 of 16 9900K's passed this test.


The only way to satisfy the OP's request is to set the desktop into "Offset mode", with a 0mv offset, and set AC Loadline to "Intel fail safe", which is 1.6 mOhms--or set AC Loadline manually to 1.6 mOhms or 160, and then you will be fine.

The problem is, most desktops hide the "overclocking performance menu" from the user, preventing you from changing the CPU's VID. Laptops only use Serial VID. They have no access to the VRM controller, which is why you won't see "Loadline Calibration" on almost any laptop. Loadline calibration requires direct hardware access to the VRM itself. The Vcore you set on a laptop is VID.

Your request won't work because laptops handle "override voltage" completely differently than desktops. Override voltage re-programs the VID and uses the AC Loadline. Desktops don't reprogram the VID. They access the voltage controller directly (in fixed vcore mode, Loadline calibration also accesses the controller directly).

On your laptop, go into your BIOS and go to "Internal VR Settings"--> "Core I/A Domain" and set AC Loadline to "1" and then see what happens if you try setting a 1100mv "override" vcore at 4.5 ghz or 4.7 ghz. (This emulates desktops using Intel default loadline calibration), and most 9900k/KS chips will have a default VID of about 1100mv at 4.5 ghz. This may work at 4.5 ghz if your CPU is good enough, but it's going to be an instant BSOD at 4.7 ghz. Assuming you can load windows.

On the eVGA Z390 dark, this is equal to either 75% or 100% "increased" vdroop loadline setting, depending on if they use a baseline of 2.1 or 1.6 mOhms for 100% vdroop.

 

I went back on my Z390 FTW and tested sub 1.1 (1.075v) @ 4.5 and now pull ~109w on CB20 @ ~1.016v. Because of your great post (and a new set of K/KS data points), I now have to go upend my previous water mark. but looking at the numbers you provided from other users and your great write up, I don't think I'm going to do much better than my 9900k P0. I'll keep looking, though. If you happen to come across a few that are better @ 4.5 total package pull sub 109w CB20 and sub 1.016v under non extreme cooling settings and for sale I'd greatly appreciate it if you would PM me.

I didn't know the exact details in regards to settings and load/pull between desktops and laptops, so that's a major thanks.

My request isn't trying to create parity between my laptop and desktop setting for setting when using the 9900k. That would never fly.

My request is to dial down on as optimal a desktop installed 9900k/kf/ks on its own merits in a desktop setting, classically delidded w/ LM I can find under load @ 4.5ghz all cores with my preset criteria (CB20 and RB2.56) using standard cooling (Air or OTS AIO) and stock thermal grease (or possibly LM) between the IHS and Heatsink. When I think I'm not going to get that much better, then delid and set that chip up with a BartX IHS to use in my 870TM with the ultimate goal using the 870TM's approach to power and heat management with as good as a sample as I can get tested on a desktop board (assuming and knowing all desktop boards are not equal but you make the best out of what you can control and test) for as quiet as possible day to day usage.

Thanks again man!

 

Tuning for quiet daily, the 2nd 5v CPU fan runs ~28dB at 100%, paired with VC the other 3 fans remain switched off. Temps are 40-50c when browsing and watching youtube with that combo, this is at 5Ghz no C-states. You could tune the 5v fan with a potentiometer to run 50% and use c-states to clock down and keep temps in the 30s. But the small fan does have a higher pitch that might annoy some.

 

What annoys me is when two fans are discordant and the tone grates my ears.

 

It's a balancing act, performance ± acoustics.

 

I get by with just the three fans

 

I was going to mod a thick 4th fan into the chassis, but after I tested it at 100% on 5v, I decided not to.

 

Yeah the different size fans are likely to cause odd tones going on.

 

This was more like adding a leafblower to the chassis.

 

Lol well there is that too.

 

We already have 3x12v 60dB fans in there, kinda overkill adding more. But reality is that you need to mod both chassis and the TM1 CPU heatsink in order for it to fit.