Just in case 
Overclocking "Raptor Lake refresh" - 14900k,14700k, 14600k etc results, bins and discussion
33 reading
Nizzen
15,921 - 15,940 of 23,546 Posts
As @ahoypolloi said, the V/F table in ASRock boards is inaccurate anyway.
You can find the correct fused VID for a given frequency yourself. Just set these in the BIOs
TVB voltage optimization=disabled
IA AC/DC = 0.01
Set all core Pcore frequency
Look for the VID in HWiNFO at idle
My post comparing the real V/F vs ASRock's readout
Joined
·
171 Posts
Wow - that fit is tighter than a virgin twink. Nice planning, and nice patience too!
Joined
·
796 Posts
HWiNFO just got an update to version 8.02 that adds:
Added monitoring of VR VOUT, IOUT and POUT via SVID (Intel).
www.hwinfo.com
e: Some quick testing. I was just loading Gray Zone warwafe and idling the lobby without any background FPS restriction. I like this because the heat of the GPU gives me a better stability indication than just running CPU tests by itself and it is a very CPU intensive game.
Anyway. I compared LLC5 (AC_LL 50) with LLC7 (AC_LL 80) with the offsets modified accordingly to make sure it's stable. In both cases I have around 70A current in the game and 90W CPU power, however with LLC5 IccMax triggers basically all the time while LLC7 rarely triggers IccMax.
The mystery remains.
Added monitoring of VR VOUT, IOUT and POUT via SVID (Intel).
Versions History | HWiNFO
Upcoming cahnges and version history of HWiNFO 32/64. Latest version, Upcoming changes and Full version history.
e: Some quick testing. I was just loading Gray Zone warwafe and idling the lobby without any background FPS restriction. I like this because the heat of the GPU gives me a better stability indication than just running CPU tests by itself and it is a very CPU intensive game.
Anyway. I compared LLC5 (AC_LL 50) with LLC7 (AC_LL 80) with the offsets modified accordingly to make sure it's stable. In both cases I have around 70A current in the game and 90W CPU power, however with LLC5 IccMax triggers basically all the time while LLC7 rarely triggers IccMax.
The mystery remains.
Joined
·
332 Posts
I tested flipping back to a lower acll with higher v/f curve points (for the same net voltages) and found that in both cases with CEP enabled it was activating MASSIVELY and tanked scores in the low acll case. In the “raised acll and dropped v/f to compensate” case it was not tanking r15 scores in the same way. Interesting, but perhaps not surprising.
The sticks are the G.Skill Trident Royal 2x 16 GB dual-rank Sammy b-die 4400 MHz 1.50v 17-17-18-38 kit. They do 4200 16-16-16-32 w/ fully tuned secondaries, tertiaries, tRFC (260), etc. on my previous setup - 12900KS, Z690 MSI Edge - w/ DRAM 1.600v.Are you sure the mounting pressure is exactly the same on both chips? I'd remount them one more time and make sure. Also maybe test 1.4V SA and see if it boots 4400. This is honestly a really tough decision, and the answer is also complicated. Do the games you play care more about RAM or CPU, what primary timings can your sticks do at the maximum stable frequency of each chip, are you willing to daily 1.4V+ SA to push that slower IMC, can you maybe run 1T on the slower frequencies... etc etc.
TL;DR: If you really want to be serious about it, tweak both chips to their limit and test performance in SoTTR / games that you play.
I re-mounted each 14900KS twice with identical results for both CPUs both times:
DDR4, Gear 1:
14900KS P-SP 121
- highest post: 4300 MHz
- highest bootable: 4300 MHz
- VDDQ: 1.500v (BIOS default)
- VCCSA: tried the following: 1.325, 1.34x (BIOS default, 1.328 HWInfo), 1.35, 1.375, 1.400
14900KS P-SP 115
- highest post: 4600 MHz
- highest bootable: 4533 MHz
- VDDQ: 1.500v (BIOS default)
- VCCSA: 1.34x (BIOS default, 1.328 HWInfo)
I'm leaning towards selling the P-SP 121 and keeping the P-SP 115. My reasoning - and I could be wrong - is that I can achieve the same CPU clocks with the lower P-SP chip but just with requiring a little more VCore which may not be a big deal due to soon going direct-die but, on the other hand, the higher P-SP CPU can't seem to achieve the higher DDR4 RAM clocks regardless of VCCSA voltage (at least the ones I tried mentioned above)...but I still haven't decided 100%, maybe 75% decided.
P.S. Maybe with the higher P-SP chip it's a case of finding the VCCSA & VDDQ "sweet spot" but I'm not going to test 100 different VCCSA/VDDQ combinations to try and find a "sweet spot" which may not even exist.
Joined
·
447 Posts
The plot thickensHWiNFO just got an update to version 8.02 that adds:
Added monitoring of VR VOUT, IOUT and POUT via SVID (Intel).
Versions History | HWiNFO
Upcoming cahnges and version history of HWiNFO 32/64. Latest version, Upcoming changes and Full version history.
e: Some quick testing. I was just loading Gray Zone warwafe and idling the lobby without any background FPS restriction. I like this because the heat of the GPU gives me a better stability indication than just running CPU tests by itself and it is a very CPU intensive game.
Anyway. I compared LLC5 (AC_LL 50) with LLC7 (AC_LL 80) with the offsets modified accordingly to make sure it's stable. In both cases I have around 70A current in the game and 90W CPU power, however with LLC5 IccMax triggers basically all the time while LLC7 rarely triggers IccMax.
The mystery remains.
Would you mind telling what ac/dc ll's and if you use set vcore or just auto while doing this and at what multi?
ICCmax is definitely weird here too.
Joined
·
1,262 Posts
It's not a tough decision because testing both of these CPUs in games will have no tangible difference. They'll both be excellent (and neither will be significantly better than an overclocked 13600K). You'll perhaps see differences in max FPS, but those will be numbers that are in far excess of the monitor refresh rate and therefore meaningless. Scenarios that bring gaming performance down to a level where there might be perceivable difference might never appear, and if they do, it'll be in ~10 years from now (i.e. when an enthusiast will have moved on more modern systems).Are you sure the mounting pressure is exactly the same on both chips? I'd remount them one more time and make sure. Also maybe test 1.4V SA and see if it boots 4400. This is honestly a really tough decision, and the answer is also complicated. Do the games you play care more about RAM or CPU, what primary timings can your sticks do at the maximum stable frequency of each chip, are you willing to daily 1.4V+ SA to push that slower IMC, can you maybe run 1T on the slower frequencies... etc etc.
TL;DR: If you really want to be serious about it, tweak both chips to their limit and test performance in SoTTR / games that you play.
No need to bottom them out, don't know if it will cause any problems with them being that tight, but they don't need to be. Like another use already said, "two fingertips tight" has been plenty enough for me to get max RAM OC and good CPU temps.
Joined
·
721 Posts
On my 12th gen CPU I've too observed that a droopy LLC with high AC LL (just high enough to maintain load voltage within the CPU's built-in VF curve) apparently doesn't trigger IccMax as much.
I think either the IccMax calculation works before the voltage added by AC LL, or somehow it is so sensitive it is seeing VRM overshoot that might be higher with a less droopy LLC.
If it's the latter case, then it might be important to keep it at the spec value since it will very effectively avoid potentially damaging current spikes to the CPU. I'd be inclined to think so, since I've always thought that IccMax works on peak/instantaneous current anyway.
Joined
·
2,973 Posts
Intel Marketing: "Buy a KS because it's capable of 2x the power." LOL Didn't miss a chance for better segmentation of SKU's.
This changes every day now though. They had the baseline profile yesterday.
Joined
·
2,973 Posts
the KS very well could be binned to handle higher voltages and currents cause not all silicon have the same ability to tolerate currents and voltages. We don't know what are all the characteristics intel considers while binning for different SKUs.
Uh, you think they binned 2x the power handling capability?
edit: and the Extreme profile doesn't retain this same 2x power difference either. But the tech sites and techtubers will re-run benchmarks with whatever the motherboard defaults to which will result in a bigger performance gap between the K and KS SKU's.
Spec says that IccMax is used to limit maximum current in VRM decoupling stage so it is a current between decoupling capacitors/inductors and CPU, VRM controller can't see it but CPU (as a direct consumer) can:
Updated picture with VR TDC/I_trip domains:
CPU tries to not reach this limit at any cost so it aggressively reduces frequency on all cores (unlike PL1/PL2 where you often may see some cores running at +1/-1 ratio).
So if you have MB with bad decoupling (cheap low-current chokes/capacitors) then set lower IccMax.
And IMO: all that "13/14th gen instability" drama comes from incorrect/unlimited IccMax by default, once you set it to reasonable value - CPU will boost fine to max clocks at light loads (games) and reduce clocks for AVX-like loads, call it "automatic AVX-offset", setting it to unlimited must be implemented as a hardware switch on motheroards, just to be foolproof out of the box
Joined
·
2,973 Posts
Where did you get the 2x number from?
PL 1/2 is just 26% more on the KS
Joined
·
721 Posts
The information on AC/DC loadline settings and notes here are going to confuse people as hell.
It's saying that the loadline configured in BIOS must match VRM loadline (which can have a loadline lower than 1.1 mOhm), but if DC LL must in turn also match AC LL, then due to the way AC LL works (at least as implemented in most motherboards), the CPU will always be heavily overvolted in most usage scenarios (which in turn would be... out of spec operation, according to Intel datasheets).
I don't think it's correctly explained, or it's explained in a way that only Intel engineers have a real idea of what this actually means.
Unless I'm misinterpreting the table?
Like I said, marketing. As you are pointing out, it's only 26% for the extreme profile, but 2x for the performance profile. But if the motherboards default to the performance profile, benchmarks will get run with the KS enjoying 2x the power once the CPU falls down to PL1.
I guess it would depend on TAU and the length of the bench as to whether this impacts the results now that I think about it.
Is there any mention of Tau anywhere?
I might have to amend my statement if TAU is sufficiently long.
Joined
·
3,291 Posts
Intel states : "Never exceed 400A", so not just a recommendation when they say "never". Why sell unlocked CPUs' at all then.... Clearly the higher clockspeeds are only meant for LIGHT loads. Not even newer games (UE5 games) comes into this category.
Joined
·
721 Posts
AFAIK, Tau should still be 56s (although in some cases, e.g. 65W CPUs and possibly the baseline profile, it might be 28s). Since it's a time constant and not a fixed duration, exactly how long the processor remains above PL1 depends on the difference between the CPU's reported power consumption and PL1.
Mine is locked to 345A and runs with 6100-6200 Mhz in gaming. 7Zip full load can run >6000 / 4700 with that limit. Single core can boost to 6400 with 100W. So... a KS totally makes sense without using it as power burner.
15,921 - 15,940 of 23,546 Posts
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
-
?
