Overclock.net banner
4381 - 4400 of 4920 Posts

·
Registered
Joined
·
650 Posts
Mine (1.305v ) is for Real Bench 2.56. So we are comparing apples to apples.
Doesn't sound right. You should be seeing a Vmin (Vdroop) under heavy load. What does the Vcore do in Cinebench R23? Can you post some pics of how your setup in the BIOS?
 

·
Optimal Pessimist
Joined
·
3,174 Posts
I see a big vdroop. In order to be stable at LLC5 I have to jack VF6 offset to +70 mv. so my VF6 point is 1.429v. that's a 124mv droop, which is the right amount for LLC5 + TVB voltage optimization. It is a lottery and this is a real stinker.

EDIT: Maybe my criteria for stability is too high. It is ultimately 112/112 AVX (AVX2 disabled) without WHEA.
 

·
Optimal Pessimist
Joined
·
3,174 Posts
Exploring this further as a weak IMC, I am able to reduce the core voltage by about 30mv at the cost of raising VCCsa and VCCio to 1.16v and 1.19v, 51/47 @ 1.279v core RB 2.34 stable for an hour run. I ran P95 112/112 AVX (no AVX2) for a half hour no WHEA. Still not great, but better. Changing Power phase control from extreme to Power Phase Response Ultra fast seems to have helped too.
 

·
Registered
Joined
·
405 Posts
Exploring this further as a weak IMC, I am able to reduce the core voltage by about 30mv at the cost of raising VCCsa and VCCio to 1.16v and 1.19v, 51/47 @ 1.279v core RB 2.34 stable for an hour run. I ran P95 112/112 AVX (no AVX2) for a half hour no WHEA. Still not great, but better. Changing Power phase control from extreme to Power Phase Response Ultra fast seems to have helped too.
I would try to rise acll to 0.5 and reduce vf#6 offset.
 

·
Optimal Pessimist
Joined
·
3,174 Posts
Already ahead of you. I adjusted ACLL for a VF6 offset of 0 while maintaining the same vcore under load. this turned out to be ACLL = 0.16. Also, I think I might be stable at 1.35v RB load 52x with VF8 offset of -0.1 with this ACLL. Only ran it a little bit on RB though. This chip is very touchy about Vccsa and Vccio. Maybe able to do a little bit better, but don't really know whether it is worth the effort and reduction in chip life.
 

·
Registered
Joined
·
405 Posts
A higher acll allows a lower vf offset...
For me works fine to use acll = LLC (mohm) divided by 2.
For llc 4 I use 0.5.... For llc5, 0.4...
 

·
Optimal Pessimist
Joined
·
3,174 Posts
A higher acll allows a lower vf offset...
For me works fine to use acll = LLC (mohm) divided by 2.
For llc 4 I use 0.5.... For llc5, 0.4...
Like I said I adjusted it for 0 offset at 51x with stable minimum load voltage. there is reason for me to do otherwise.
 

·
Registered
Joined
·
405 Posts
Negative offset and lower vcore at idle.
 

·
Optimal Pessimist
Joined
·
3,174 Posts
Negative offset and lower vcore at idle.
Vcore at idle doesn't mean much to me. At idle all that matters to me is that a single core isn't pulling more than, say 15-20A. Negative offset? You will probably need one at some VF point. What is important is that, for a given frequency, it is stable across all loads given VID = VF +ACLL*Current behavior. I guess I try to pick a VF point that is an inflection point and has the worst behavior - zero VF offset with surrounding points requiring zero or less than zero offset. If you set the AACLL such that the offset is zero for that point, then that seems to work for the VF points below it (which don't really need adjusting), and then you can adjust the points above it. For me that is VF6. Setting the ACLL to 0.16 mohm and VF6 offset = 0 gives me stable VF6 and below (all zero offset), and I adjust VF7 offset to -0.1. I could just as easily do this procedure with LLC4 with a larger ACLL and potentially better guard band.
 

·
Registered
Joined
·
650 Posts
Are you using Sync all cores @GeneO?

My setup:
Sync all cores 51x
Adaptive voltage - Auto, Auto
SVID - Best case (which is AC/DC LL 0.01/0.01)
LLC 5
Voltage Optimization - Enabled
VF curve - No change

I thought this was stable but apparently it's not, the Vdroop is massive and hits instability. Even going to LLC6 does not help. It appears the only way to stabilize is adding voltage to the VF Curve for 5.1 (VF #6).

The problem going to a higher AC LL of 0.50 (with Sync all cores) like Roberto suggests is that it draws excessive voltage (1.450V). That is 116mv over the CPU internal VID for 5100 and not necessary. There must be a balance in the AC LL that is needed for stability and it appears you have already worked this out (0.16).
 

·
Registered
Joined
·
3,810 Posts
For me on sync all cores 5.2 core 4.6 cache it's ACLL 0.10, LLC4 (slight droop) with -0.150 offset (v/f curve / advanced offset at x52) with the rest of the idle X8 up to x48 curve on Auto VID. It never runs x51 but I have it set to -0.120 which is what it needs if I sync all cores to x51. This results in a load voltage of 1.354v (VR VOut and verified with a multimeter on the readout points of the board). It's rock solid there. And no crashes in light loads or switching loads. AVX workloads go a bit higher to 1.376-1.380v but still well within spec and cooling. It's hot here now like, 33c, room temp around 25-26c with the A/C running, and it runs high 50's to low 60's in games and low 80's in extreme stress tests like Prime95 Small FFT AVX On.

RAM is at 4400 17-17-17-36-340-2T with tight secondaries at tWR 12, tCWL 16, tRRD_s & l 4, tFAW 16, tRTP 8, tWTR 7/2, tREFI 48500 and RTL/IO 66/66/6/6. ODT 80-40-40. VCCIO is 1.25v, VCCSA is 1.35v. It cannot go lower, the IMC needs this for 4400.

Only thing I really hate about this CPU is the cache. It will not in any way do anything over x46 without hammering the core voltage through the roof. For x47 it needs 1.390v and for x49 it needs 1.456v to even stand a chance and that is only stable under load, idle and switching loads crash it with that high of a cache.
 
  • Rep+
Reactions: Astral85

·
Registered
Joined
·
50 Posts
Quick question. I'm working on going direct die with the rockitcool kit on my z490-e strix and a Optimus foundation block. I'm a bit worried about fitment and proper contact. Do I need to worry about it?
 

·
Facepalm
Joined
·
10,019 Posts
Discussion Starter · #4,393 ·
Exploring this further as a weak IMC, I am able to reduce the core voltage by about 30mv at the cost of raising VCCsa and VCCio to 1.16v and 1.19v, 51/47 @ 1.279v core RB 2.34 stable for an hour run. I ran P95 112/112 AVX (no AVX2) for a half hour no WHEA. Still not great, but better. Changing Power phase control from extreme to Power Phase Response Ultra fast seems to have helped too.
Ultra fast should be the same or even possibly slightly worse than Extreme.

Extreme is the most aggressive setting.
"Ultra fast" is only on a user defined setting set to the highest level. Extreme should set everything to the highest level for you.
 

·
Registered
Joined
·
9 Posts
Overclocking my 63 score CPU. Help if possible. I just re applied thermal compound and moves my h100i v2 rad to the front as an intake rather than at the top as an exhaust. I'll include a link to my spreadsheet detailing the things I have tried. As well as pics of the f steps and what not.

As you can see from my spreadsheet so farm my best cinebench r23 score has been on auto everything. I mostly use this beast for gaming and x265 encoding.

Spreadsheet, my 5.2 vid idle is at the bottom of the spreadsheet

Currently I am at 1.26 vcore bios, and cores are set (2x50)(6x49)(10x48)(10xA) , though I don't think I ever see 2 of them at 5.0ghz, not sure how that thing works. In the spreadsheet you can see my stable voltage at 4.9, might be better now that I redid the cooling.

Looking for some help and suggestions.
 

Attachments

·
Optimal Pessimist
Joined
·
3,174 Posts
Ultra fast should be the same or even possibly slightly worse than Extreme.

Extreme is the most aggressive setting.
"Ultra fast" is only on a user defined setting set to the highest level. Extreme should set everything to the highest level for you.
I was wondering whether it was the same, hence "seems to have". I changed it while I was getting random WHEA so just some randomness. Thanks.
 

·
Facepalm
Joined
·
10,019 Posts
Discussion Starter · #4,396 ·
I was wondering whether it was the same, hence "seems to have". I changed it while I was getting random WHEA so just some randomness. Thanks.
That stuff takes some major iron ration eating to test.
Let me know if you're able to determine if it actually makes a repeatable difference, even if it takes days to do so (assuming you still care about such a thing).
I would think the fastest way to test such a thing would be LinX residuals, but what do I know....
 

·
Registered
Joined
·
1,220 Posts
Well I keep getting these WHEA errors when I play Metro Exodus Enhanced, tried reducing cache to stock (from 47x to 43x), increasing vcore from 1.20 to 1.25v and no success... I guess this is the instability where i9 9900k, i9 10900k suffers and i9 11900k are immune?
2514398
 

·
Facepalm
Joined
·
10,019 Posts
Discussion Starter · #4,399 ·
Well I keep getting these WHEA errors when I play Metro Exodus Enhanced, tried reducing cache to stock (from 47x to 43x), increasing vcore from 1.20 to 1.25v and no success... I guess this is the instability where i9 9900k, i9 10900k suffers and i9 11900k are immune? View attachment 2514398
That's the good old parity error. Yep.
Either increase your loadline calibration (higher LLC actually helps reduce parity errors, but it can increase the chances of CPU Cache L0 errors if you reduce the vcore when you raise the LLC strength), reduce the ring ratio, or increase your cpu vcore.
Apparently 'certain' RAM tertiary timings can affect this but that's @cstkl1 's thing. I don't know what does.
 
  • Rep+
Reactions: owikh84

·
Registered
Joined
·
1,220 Posts
That's the good old parity error. Yep.
Either increase your loadline calibration (higher LLC actually helps reduce parity errors, but it can increase the chances of CPU Cache L0 errors if you reduce the vcore when you raise the LLC strength), reduce the ring ratio, or increase your cpu vcore.
Apparently 'certain' RAM tertiary timings can affect this but that's @cstkl1 's thing. I don't know what does.
I tried increasing voltage didn't fix, atm I use LLC6 with 1.20v in BIOS resulting in 1.14v-1.15v in full load, you think I should increase to LLC7? But I would like the same load voltage so I would have to reduce the vcore in BIOS a bit to have the same 1.14v at load, definitely reducing ring ratio and increasing CPU vcore didn't fix. I have a big stuttering when the error occur while in game but I can still play fine for a long time before the 2nd WHEA error. It doesn't crash the game or anything, just a stutter.
 
4381 - 4400 of 4920 Posts
Top