Overclock.net banner
4361 - 4380 of 4920 Posts

·
Registered
Joined
·
405 Posts
I don't think so. He was pretty clear and explicit.
Take a pic of your CPU V/F curves, let me see how bad are you chip... LOL
 

·
Optimal Pessimist
Joined
·
3,174 Posts
So I thought I'd test this. I ran RB 2.64 5 GHz with ACLL = .26 and DCLL either .26 or .76 (I calculate LLC5 is about .76). These run at about 155A @ ~ 1.26v.
So according to you I should see a drop in power of current*(.76-.26)*current/1000 (current*volts) = 12 Watts. I measured these successive iterations between .26 and .76, here are the last two, the others are about the same. The current and power are HWINFO64 taken from the Asus EC chip.

.26 156A 196W
.76 157A 197W

So what would have been expected for .76 would have been around 196-12 = 184W. Always the ,76 run was slightly higher in wattage, not lower.

So basically no difference in reported wattage.

Dang, guess I should look at power reported by the chip. the EC probably has all information to calculate it correctly.
 

·
Optimal Pessimist
Joined
·
3,174 Posts

·
Optimal Pessimist
Joined
·
3,174 Posts
OK, now I believe it. I always look at the EC power and current, which are apparently always correct regardless of AC/DC LL. I never look at package or IA cores power so did not notice a discrepancy. They are indeed about 12w higher with the .26 ACLL than the .76 mohm as predicted for this test.

Now I imagine that also impacts power throttling, which I never hit (I setup to throttle pretty much by temperature). So yeah, should set DCLL = VRM LL for correct CPU calculated power, though EC is always on the money for reports.

Now yet another way to find out what the mohm is for a given LLC - power comparison.
 
  • Rep+
Reactions: RobertoSampaio

·
Registered
Joined
·
405 Posts
OK, now I believe it. I always look at the EC power and current, which are apparently always correct regardless of AC/DC LL. I never look at package or IA cores power so did not notice a discrepancy. They are indeed about 12w higher with the .26 ACLL than the .76 mohm as predicted for this test.

Now I imagine that also impacts power throttling, which I never hit (I setup to throttle pretty much by temperature). So yeah, should set DCLL = VRM LL for correct CPU calculated power, though EC is always on the money for reports.

Now yet another way to find out what the mohm is for a given LLC - power comparison.
Nice !!!!! :)
Now you have the MB power measurement matching to the CPU power measurement.
Are you running "sync all cores" or "by core" ?
Did you set VmaxStress on or off? And how about Voltage optimization?
 

·
Optimal Pessimist
Joined
·
3,174 Posts
Vmax stress off, voltage optimization on, all cores. But it really doesn't matter. It needs 1.3v+ core under load for 51x. 50 and lower are another matter. I am working on them. 52 is not stable but for over 1.36v and I can only test that with no hyperthreading. Just a crappy chip.
 

·
Registered
Joined
·
650 Posts
sp 63. Adaptive LLC 5 with AC//DC loadlines of of 0.26 / 0.26. IMC is weak too. can't do uncore higher than 47 and need a higher IO/SA than I have ever needed for a mild memory OC. Tests with P95 FFT 112/112 with noAVX and noAVX2, Realbench and About the worse predictions I have seen for a 10900k. Have delidded and still at 51 use OCTVB to drop a multiplier at 75c and another at 85c.

View attachment 2514027
Your predictions are better than my SP 63. I don't know if the predictions are accurate anyway. Try sync all cores (51), SVID - Best case, Adaptive voltage (all Auto), LLC5 and let me know the result. I have 5100Mhz stable in that config with 1.270-1.30v (Vmin around 1.24v).
 

·
Optimal Pessimist
Joined
·
3,174 Posts
Your predictions are better than my SP 63. I don't know if the predictions are accurate anyway. Try sync all cores (51), SVID - Best case, Adaptive voltage (all Auto), LLC5 and let me know the result. I have 5100Mhz stable in that config with 1.270-1.30v (Vmin around 1.24v).
I agree, they aren't in general. My 10700k performed better than predicted by quite a bit. This one worse than predicted. That is what I am running at sync all core, 51x, adaptive minimum vcore at load that is stable is 1.3v+
 

·
Optimal Pessimist
Joined
·
3,174 Posts
Your predictions are better than my SP 63. I don't know if the predictions are accurate anyway. Try sync all cores (51), SVID - Best case, Adaptive voltage (all Auto), LLC5 and let me know the result. I have 5100Mhz stable in that config with 1.270-1.30v (Vmin around 1.24v).
So under what type of load do you get that 1.24v vcore?
 

·
Registered
Joined
·
405 Posts
I agree, they aren't in general. My 10700k performed better than predicted by quite a bit. This one worse than predicted. That is what I am running at sync all core, 51x, adaptive minimum vcore at load that is stable is 1.3v+
You can do the following test...
Reduce RING to 43x and test [email protected]@full_load.
After that you can go rising ring until stable. I think all will be fine until 46-47x

If I set ring to 48x I need Vcore=1300mv no matter the CPU frequency.

Another thing you can try is to start with LLC#4, AC_LL=0.5 and DC_LL=1.1 and make changes in V/F#6 (5100).
This ensures sufficient Vcore at idle and when CPU is @load AC_LL provides additional voltage.
 

·
Optimal Pessimist
Joined
·
3,174 Posts
Already did that with LLC4. But I may be on the cusp of ring instability. I am at 47 and can't do 48, so I'll play around with that some, thanks. I did lower it to 46 at one point. The IMC/ring does seem a little weak on this.
 

·
Registered
Joined
·
50 Posts
I'm trying to determine whether or not I should direct die my 10900k. Currently building a pretty aggressive loop for it in order to hit 5.3. I've delidded a 6700k so I understand the process, but I've heard some horror stories about the newer soldered CPU's.

I feel like if I'm building this loop to extract some extra performance out of it that I should take the risk and go ahead and run it direct die, I just want to make sure it's not extremely risky if I take it slow and use a heat gun. Do you have to be somewhat of an idiot to break it during the process? You basically just want to heat it up during the delid and then just be gentle when mounting the cooler, right?
 

·
Registered
Joined
·
592 Posts
You need a direct-die mounting kit, but yeah, it's super simple. I had zero issues de-lidding my 10900K. I heated it up with a heat gun, threw it in the de-lid tool from RockItCool, and it was still pretty tight. Took it back out, hit it with some more heat, and it popped off no problem. I'm not sure how you could screw it up with a de-lid tool. Maybe if you decided to crank on it with zero heat?

Either way, my chip came out perfect. I used tape to seal off the die when I applied liquid metal, and that worked really well too. Overall, it's the best way to reduce temps. All the cooling capacity in the world doesn't matter if you can't transfer the heat fast enough to utilize it.
 

·
Registered
Joined
·
50 Posts
You need a direct-die mounting kit, but yeah, it's super simple. I had zero issues de-lidding my 10900K. I heated it up with a heat gun, threw it in the de-lid tool from RockItCool, and it was still pretty tight. Took it back out, hit it with some more heat, and it popped off no problem. I'm not sure how you could screw it up with a de-lid tool. Maybe if you decided to crank on it with zero heat?

Either way, my chip came out perfect. I used tape to seal off the die when I applied liquid metal, and that worked really well too. Overall, it's the best way to reduce temps. All the cooling capacity in the world doesn't matter if you can't transfer the heat fast enough to utilize it.
Thanks for the info. I'm going to go all in and order the direct die kit from RockItCool I think. With a dual 360mm push pull setup with a D5 running maxed out and a direct die setup I should be able to push it far more than I currently can with my Corsair 280mm AIO setup.
 

·
Optimal Pessimist
Joined
·
3,174 Posts
You need a direct-die mounting kit, but yeah, it's super simple. I had zero issues de-lidding my 10900K. I heated it up with a heat gun, threw it in the de-lid tool from RockItCool, and it was still pretty tight. Took it back out, hit it with some more heat, and it popped off no problem. I'm not sure how you could screw it up with a de-lid tool. Maybe if you decided to crank on it with zero heat?

Either way, my chip came out perfect. I used tape to seal off the die when I applied liquid metal, and that worked really well too. Overall, it's the best way to reduce temps. All the cooling capacity in the world doesn't matter if you can't transfer the heat fast enough to utilize it.
I think it is relatively safe. The solder is somewhat soft snd a heat gun should help though I did mine without one. 11th gen is another story because of components on the pcb.

EDIT: I guess I should say I am relidded, not direct-die.
 

·
Registered
Joined
·
592 Posts
Thanks for the info. I'm going to go all in and order the direct die kit from RockItCool I think. With a dual 360mm push pull setup with a D5 running maxed out and a direct die setup I should be able to push it far more than I currently can with my Corsair 280mm AIO setup.
I'm currently Direct-Die on a EVGA 360 CLC and it works alright. The issue I have is the CLC coldplate was very convex and caused nearly zero contact on the die. After much lapping/sanding, it's better, but still not perfect. The cooling ability of the CLC is just not there either. I have a bunch of Heatkiller stuff coming in and will be building my first custom loop, and retaining the direct-die. I can run 5.2Ghz but I have a bad spike on Core4 due to the coldplate issues and it causes me to hit a Cache L0 fail. The cooler the chip, the less voltage required.
 

·
Optimal Pessimist
Joined
·
3,174 Posts
You can do the following test...
Reduce RING to 43x and test [email protected]@full_load.
After that you can go rising ring until stable. I think all will be fine until 46-47x

If I set ring to 48x I need Vcore=1300mv no matter the CPU frequency.

Another thing you can try is to start with LLC#4, AC_LL=0.5 and DC_LL=1.1 and make changes in V/F#6 (5100).
This ensures sufficient Vcore at idle and when CPU is @load AC_LL provides additional voltage.
No banana.
 
  • Rep+
Reactions: RobertoSampaio

·
Registered
Joined
·
50 Posts
I think it is relatively safe. The solder is somewhat soft snd a heat gun should help though I did mine without one. 11th gen is another story because of components on the pcb.

EDIT: I guess I should say I am relidded, not direct-die.
Ordered the full kit from RockIt. Excited to see how hard I can push this with two GTS 360mm rads with 12 P12's push/pull and a D5 all running fast. I'm at 5.2 now with 1.4v under load (z490-e strix board) with very high temps on a 280mm Corsair AIO push/pull. I'm hoping the dramatic improvement in thermals can let me hit 5.3/5.4 with aggressive voltage and low temps.
 

·
Registered
Joined
·
650 Posts
So under what type of load do you get that 1.24v vcore?
This is the Vmin under very heavy load like Cinebench R23 and Asus RealBench. Run either of these and monitor with HWiNFO. Post the results.
 

·
Optimal Pessimist
Joined
·
3,174 Posts
This is the Vmin under very heavy load like Cinebench R23 and Asus RealBench. Run either of these and monitor with HWiNFO. Post the results.
Mine (1.305v ) is for Real Bench 2.56. So we are comparing apples to apples.
 
4361 - 4380 of 4920 Posts
Top