Just in case 
Overclocking "Raptor Lake refresh" - 14900k,14700k, 14600k etc results, bins and discussion
33 reading
Nizzen
2,841 - 2,860 of 23,546 Posts
The VID you see in HWiNFO, or similar, are not comparable to the values from the SP rating screenshots, or the vid list
There are many factors/settings that skew the value you get from any software (including, but not limited to, AC_LL and DC_LL)
You could end up returning a perfectly good chip. But unfortunately the only way to know would be do some proper testing.
Theres that number again…33… maverick also stopped gunning at the fifth gen fighter with only 33 rounds left haha
Joined
·
1,081 Posts
What is the relationship between a cell's vid and vcore?
For instance if the vid for 5.8 is 1.4v, is that the voltage the vcore will be set to by the bios?
How much margin for error is there? If the cell for 5.8 is 1.4v, could it run at a lower voltage?
Its black magic. I can't claim to fully understand it, I'm not sure anyone outside of the board makers and/or Intel know for certain.
The factory fused VID (the values in SP score screenshots and the list) are the starting point. The motherboard then modifies it in various ways.
For the voltage it requests the VRMs to send the CPU (if on adaptive voltage) it'll take the starting/fused VID, add a little based on your chosen loadline setting (MSI boards do this, Asus boards perhaps, but that may also be modifiable with a setting), add some more based on the AC_LL setting (and how much current draw at that time), and perhaps then remove some if TVB voltage optimisations are enabled and your temps are low.
For the VID you see reported in HWiNFO, you start with all of the above, and then the value is modified based on the DC_LL setting (and how much current draw there is at the time).
But of course, you can override the VID by using a fixed voltage. And also, the starting point VID isn't always the VID from one of the cores, in some (rare) situations (high ring clocks, low core clocks) you can end up with the ring VID being higher than the CPU VID, and it'll use that as a starting point instead.
Its the above that makes the SP rating stuff so valuable (for an initial indication), as Asus use some tricks to figure out the fused VID, I presume, by removing all of the above factors to get an unmolested value.
There is a manual way to try to extract the fused VID values on an MSI board (which involves setting very specific BIOS settings, then manually setting core speeds and checking the value displayed in the BIOS), if this is as accurate as Asus SP readouts, I cannot say for sure, but it's all MSI users have.
As for how this all relates to Vcore; the difference between the requested VID and the Vcore comes down to which loadline setting you have (different settings will allow the Vcore the CPU sees droop below the requested VID by different amounts, depending on how 'strong' a setting you choose).
TL;DR The VID readouts in HWiNFO (and therefore the "CPU Package Power", unless you have properly calibrated DC_LL) are generally worthless, and almost always not comparable.
So I've been testing the thing in post 2, I was too optimistic/afraid of too high power draw so I tried setting 1.28, which naturally failed even starting the test. CPU-Z bench went fine though.
Kept creeping up 10mv and retesting. Finally passed 10 runs back to back at 1.33Vset.
Setting anything less than 1.33 either crashes Cinebench or throws whea errors almost as soon as the run starts. I even tried sneaking up on 1.325 since 1.33 seemed as it would have finished the whole 10 minutes by itself, but that was just as unstable as the 1.32Vset.
As I'm writing this, I thought I was gonna do the full 10 minute loop, and started the run and reset hwinfo and BAM I got 1 whea error. Now I don't know if that was just because I pressed the reset button just as the test was starting, or I maybe need a small bump to fully stabilize it. I'm trying the full loop again without fidgeting with hwinfo to see.
On another note: I've only got one EPS12v cable plugged in to the motherboard, never had an issue since the most I've run was an 13700k with an OC, but normally ran with HT off/Ecores on so rather low power draw. Does this beast necessitate another EPS12v cable for this kind of power draw? Or am I grasping at straws here? TBF the previous 14900k had a peak power draw during R23 a 367w stock, it didn't crash.
EDIT: Not bothering with any more of this testing, as after that error it errors out every time now. I bumped Vcore up to 1.34Vset now, and it still throws a WHEA. As a final attempt I'm gonna set 1.35, but I'm not gonna torture it any further until I see some feedback from any of you guys here.
Kept creeping up 10mv and retesting. Finally passed 10 runs back to back at 1.33Vset.
Setting anything less than 1.33 either crashes Cinebench or throws whea errors almost as soon as the run starts. I even tried sneaking up on 1.325 since 1.33 seemed as it would have finished the whole 10 minutes by itself, but that was just as unstable as the 1.32Vset.
As I'm writing this, I thought I was gonna do the full 10 minute loop, and started the run and reset hwinfo and BAM I got 1 whea error. Now I don't know if that was just because I pressed the reset button just as the test was starting, or I maybe need a small bump to fully stabilize it. I'm trying the full loop again without fidgeting with hwinfo to see.
On another note: I've only got one EPS12v cable plugged in to the motherboard, never had an issue since the most I've run was an 13700k with an OC, but normally ran with HT off/Ecores on so rather low power draw. Does this beast necessitate another EPS12v cable for this kind of power draw? Or am I grasping at straws here? TBF the previous 14900k had a peak power draw during R23 a 367w stock, it didn't crash.
EDIT: Not bothering with any more of this testing, as after that error it errors out every time now. I bumped Vcore up to 1.34Vset now, and it still throws a WHEA. As a final attempt I'm gonna set 1.35, but I'm not gonna torture it any further until I see some feedback from any of you guys here.
Whats your cb23 score at that? Like 45000-46000k?
Joined
·
13,203 Posts
You need to have both cables plugged in. Failure to do this can drop the 12v input values to the CPU (vcore is derived from this source rail) which could make things even worse. We've seen this on RTX and AMD video cards, when the PCIE voltage from the power plugs was reading too low.
Are you setting Actual VRM Vcore Voltage, manual mode, and Loadline Cal level 6?
You can also see your live vcore, IF the VRM is the same type as used on maximus boards, by downloading OCTool and using the Raw VRM option, this will show you your die sense voltage and be much more accurate than what hwinfo or cpu-z will show you. Do keep in mind this will drop your benchmark scores slightly so don't use this tool when doing score compares. You can get OCTool from the ROG forums in the Raptorlake resources thread by shamino.
And if you have trouble passing after, see what happens if you drop the ring ratio down to x45. Might help a little. Wont stop a bad chip from being bad though.
Joined
·
1,126 Posts
Is it your psp 114?
Thanks, I'm gonna go find that second cable before pushing any further then.
This makes perfect sense as I used to run my 1080ti with just one cable using the dreaded daisy chain end as the second PCI-E 12v. Ever since I installed a second cable from the PSU I could run much higher OC.Yes to all three.
I was running just hwinfo at the beginning, then I started to have OCTool open afterwards, as the smaller steps in the readout is easier to keep track of. But I don't know f this is die sense or not, but the reasing is very similar to the one in hwinfo, only more granular. I don't car one bit about the scores atm, just trying to find vmin etc. I can run the bench alone at a later time if I want to see BIG number. xD
###
It crashed during render pass 24 with 2:22 left on the timer, no WHEA though. Will refrain from testing any more until I find and install that second EPS12v cable.
Joined
·
1,326 Posts
yea, p114, e62 , sp97
Joined
·
1,081 Posts
I am completely blind to my cpu's vid table. My asus board doesn't have SP score, nor does it provide vid information in any of the menus I've scoured. It is maddening that the board is obfuscating/hiding the information to me, and despite the hundreds of options that can be set, it is not giving me access to the most important.
Which brings up an interesting observation. If my board doesn't provide this information, then probably the vast majority of motherboards don't; it is only a small percentage of consumers who have access to this information. The $600 motherboards are probably not very common, and afaict, the "Extreme Tweaker" information is only available on those high end boards.
All I can do it use XTU to change offset, tweak clocks, and test for stability. I really want to know if my 14900K is a junk bin. I would be happy with average, because I feel that 5.7 all core would be obtainable with tweaking, I just don't want junk.
I have been reading this thread for days and I'm at the point of wanting to buy another mobo just to have this information, but that defeats the purpose of the "upgrade"; if I'm buying a new board I would just wait til 15th.
Any suggestions on how to determine the quality of my cpu without extreme tweaker information?
Joined
·
2,973 Posts
The voltage is definitely exceeding the intel spec for the current the cpu is drawing.
This is not to you specifically, but there are people who swear that pushing 300w+ has degraded their cpu, and many people consequently are afraid of pushing watts because of them. People who said that might not be wrong, however they do not mention the voltage and current as they think that is irrelevent to degradation which is wrong. Power metric alone can only be used to assess cooling capacity.
Power is not the grim reaper to CPUs. It is VOLTAGE AND CURRENT.
According to intel ( 1520 - 400*1.1 = 1.080v ) which should roughly be 432w, which is under extreme config spec. This doesnt mean 432w is safe for all current/voltage configs.
Intel specs everything at TJmax which is 100c, and lower is better for reliability.
This is 1.52v set LLC3
The below pic is under intel electrical Spec and the cpu is pulling 330W.
Joined
·
1,081 Posts
How do you get this information to fill in? I have the "AI tweaker" version similar to yours, but my values are not filled in, and I don't have an "AI features" menu.
Is there a setting in the bios that unlocks this information?
Attachments
I'm not aware of any claimed accurate method on an Asus board that doesn't have SP rating.
Although perhaps just doing the same thing that is claimed to work on MSI boards will also work on an Asus board. All the same settings should be available, just under different names (and that the AC_LL and DC_LL settings should be 0.01, not 1, and the Loadline should be 1 not 8):
I have no way to test, or confirm. Ideally someone with an Asus board and a CPU with a known VID table could try it out, and let us know if the values line up.
You could try it yourself, see what values you end up with. Should be able to get a decent indication of its accuracy by comparing to the list. If it looks like you have an unrealistically good or bad CPU, then I guess the method doesn't work. But if you end up somewhere in the middle of the table, then I'd guess it does.
Using the MSI method above, my 14900K would be 30th on the list, a different one I tested would be ~50th, so to me the MSI method seems accurate.
Joined
·
2,973 Posts
apologize if u had said this already but have u flashed the bios?
You have a TUF motherboard right? afaik Only Strix boards and above have these features, as only the Maximus series have the features my Strix is missing. That's segmentation for ya.
Joined
·
1,081 Posts
So, in my quest for knowledge, the question of TDP comes up. How much is being used at a given setting?
I have been quoting momentary peak, but there appear to be two values in Core Temp. (momentary and average?)
When I am referencing TDP, should I use momentary or average?
This screenshot is from my all core 56x run. In the first minute before everything gets hot, the momentary maxes around 280ish, but it creeps up over time. How much TDP is this considered?
![Font Material property Parallel Pattern Number]()
I have been quoting momentary peak, but there appear to be two values in Core Temp. (momentary and average?)
When I am referencing TDP, should I use momentary or average?
This screenshot is from my all core 56x run. In the first minute before everything gets hot, the momentary maxes around 280ish, but it creeps up over time. How much TDP is this considered?
Joined
·
13,203 Posts
And PLEASE FOR THE LOVE OF GOD, IGNORE the 1.72v VID setting. Any VID past 1.520v is set by OFFSET mode (a register on the VRM) which on Z390 used to be off by default (on Gigabyte and afaik, MSI), but on by default on Asus. On Z490 and newer, everyone has offset mode enabled at all times. OFFSET mode HAS NOTHING TO DO WITH THE CPU's FUSED VID OR DOES IT HAVE TO DO WITH "OFFSET ADAPTIVE VOLTAGE". The max native VID is STILL MAX 1.520v. Offset mode allows AC Loadline to boost the native VID by a certain factor depending on the ACLL value (and on Z390, by how much current--basically the complete inverse of loadline calibration). Note CAREFULLY that AC Loadline's function was changed DRASTICALLY on Z490 and newer---it is NO LONGER the inverse of LLC. it is NO LONGER affecting VID by the formula (given by Elmor on Z390), with respect to VID only, of VID_Native + ((ACLL mohms * Loadstep (D1-d0) IOUT) - (DCLL mohms * OUT)) - vOffset... (with, on adaptive voltage, you can replace DCLL mohms with loadline calibration mohms when looking at cpu vcore--). <--note the above formula ignores native VID's influence by temps thermal velocity boost (not core boosting TVB, but VID scaling--note that Asus V/F table is based on TVB being 'disabled' or the CPU being at 100C).
On Z490 and newer, the 'offset mode' allows ACLL to boost native VID past 1.520v again, but the formula on how it functions is drastically different. I posted some tables many months ago, in the 12900K thread, testing ACLL 0.01 mohms up to 1.1 mohms, if someone cares to do a search.
When dealing with FIXED (manual) vcore, again the VID cap is 1.520v, NOT 1.720v, so the formula for not cratering your chip still applies (based on 1520mv - (1.1 mohms * desired amps draw)=max safe vcore--and notice, the higher the temps, the more you will still eat into this guardband EVEN if you are under the max safe vcore limit--lower temps are better.
Your guardband determines how much you can overclock or undervolt. "Degradation" is eating into this guardband obviously.
Joined
·
3,291 Posts
Thanks for the tip, but I prefer to buy hardware with warranty intact, and in Norway that is 5 years.
2,841 - 2,860 of 23,546 Posts
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
