Alder Lake CPUs: Overclocking and general ADL memory/motherboard discussion

[Zantal]

Is there any way to increase the stability of the memory controller?
I managed to get 4200MT at gear 2 but is highly unstable at 4000 gear 1, windows will boot but most programs or games will crash.
If I could get it to work at gear 1 this system would be perfect. all this was possible after I updated the bios, anything outside xmp parameters wouldn't even boot before.

Also would you guys regard 1.45 max on vcore safe?
during full load it stabilizes at 1.38 but during lower loads it can jump up to 1.45

 

[Johan45]

You can try SA/Uncore voltage around 1.2-1.25V

 

[Woomack]

@Zantal, I could run at all auto voltages at DDR4-4600 Gear 2 on Biostar motherboard, without a tuned BIOS. Gear 1 couldn't work at more than DDR4-4000 and even at this clock, the motherboard was sometimes acting weird. It's not guaranteed you can run DDR4-4000 at Gear 1. Sometimes it works, sometimes not. That's why I said, if you can't stabilize high enough clock at Gear 1 then try as high as you can with Gear 2 and maybe it will be easier and give you the same or better results. Can be also worse but you won't know that until you test it.
I don't think I saw anywhere what RAM you are using (or I just don't remember). Depending on the used IC, it may OC better or worse.

I assume that 1.45V is still in the specified range for 12th gen CPUs so should be safe. Of course, if you run it at such high voltage and under load for longer then it may cause faster degradation ... or not. Funny thing is that even Intel doesn't know that. If you run it with power saving options enabled so when it drops to low clock/lower voltage when the CPU is under low load then for sure all will be fine.

I still have to check my TridentZ Royal 2x16GB DDR4-4400 and something that could run at DDR4-5200+ on Z590, to compare performance.

 

[Voodoo Rufus]

How about these sticks, Woomack?

https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820374042?Item=N82E16820374042

 

[Woomack]

I'm almost sure it's a dual-rank Samsung B as it's the only one that fits the XMP. It should be the best option if you want to run it at Gear 1 and CR1.

Btw. I made some quick tests with my 2x16GB TridentZ Royal and it doesn't work at XMP/4400 on ASRock Z690 Extreme. The motherboard has an option to set full XMP or only main timings. However, it doesn't work at any auto or manual settings. I know this RAM is demanding and so far could run at XMP only on MSI Z590I Unify.

 

[Zantal]

I think I have hit the limit for my motherboard/cpu capability

Ram is Kingston Fury Renegade 32gb 3600 CL16 (KF436C16RB1K2/32)

been trying to get 4000 MT at gear 1 for a while now, it will complete benchmarks just fine (even finished a Cinebench r23 with 29k score final)
but it will crash every 5/10 minutes in guild wars 2 no matter what I do, games hit the memory controller differently maybe
(continually accessing non contiguous data I suppose)

Increasing voltages will do nothing for stability above 1800 Imc freq.

settled for 3900 gear 1, I tried to run higher frequencies at gear 2, even managed to boot at 4600 but to get it stable I needed to loosen timings to the point where
the increase in bandwidth is negligible and at a much higher latency, besides I am not 100% sure it is stable anyway

(final overclock for ram is 3900-15-15-15-36-51-CR1 edit: this is now 15-14-14-30-44 might get lower lol)

I don't think they designed the IMC to work at anything higher than 1900Mhz, which is logical since it's probably made to work with DDR5 speeds anyway.


Edit:

There is some link between memory controller and E cores clock, E cores will be stable at 4.2 ghz but will start throwing out errors on tests coupled with high
MC frequency, I reduced them to 41x and now I can run 4000MT Gear 1 15-14-14-27 at 1.5v and P cores at 5.2ghz all cores. it's a pity these sticks won't do 14 CAS, it will just not boot not even with added voltage.

Also can someone explain to me why on low load (but still in C0 state) VCore will fluctuate wildly between 1.42 and 1.47 but when running cinebench for example it drops down to 1.38/1.39 top

I wish I could make it run always at max 1.4, I feel like it could run cooler with "light" gaming loads
1.38/1.39

 

[Woomack]

IMC was designed to work with RAM at 1600MHz/3200MT. Everything above that is overclocking and not guaranteed. On my ASRock Z690 Extreme, I have problems after retraining. Once it works fine at 4000 G1/CR1 and another time, after I turn off and on the PC, it crashes. I guess that BIOS is still not tuned well enough. I have similar problems on Gigabyte but DDR5 and anything above 5200.

VCore depends on many factors. Under load, it always drops and that's why you use LLC to keep it stable at the desired value. The same is for AMD and Intel. Find Load Line Calibration in BIOS and check which level gives you the best results (usually 1-3 lower than the max). Then you can set 1.4V and keep it at this voltage regardless of whether it's under load or idle.

 

[Zantal]

Thank you very much for your time.

Hopefully we get enough info in here to compile a better "how to" list for people looking to overclock Alder Lake.

What I found out (and should have known before since schematics clearly show it) is that Core voltages are shared, meaning that if you overclock E-Cores the increase in voltage is reflected on P-Cores as well, even if they do not need it that high.

E-Cores just shouldn't be overclocked past 4.0/4.1

I have one last question to complete my final overclock.

In the Asus bios there is a setting to disable down binning of the ring clock but it comes with a warning saying that if disabled it could potentially overvolt the core.
Disabling down clocking would keep the ring frequency above 3.6ghz, which is the frequency set when E-cores are active (like 90% of the time in every load scenario)

Edit: I think I have gotten this wrong, I guess the frequency will still drop when E-Cores are active and ring frequency is decoupled from P-core frequency
Anyway, it's a shame that e-cores influence ring frequency to such an extent.

What really throws me off in these warnings is the use of the word "Could", I really suffer from anxiety when playing with voltages outside of the safe spec limits
and I don't know what to look for if I were to disable the down binning option and keep the ring at say 4.5/4.6 ghz.

Overclocking the memory and tightening the main timings so far has had a minor but tangible impact in the game I play the most (Guild Wars 2) I can only assume it will benefit from a higher ring clock as well.

 

[EarthDog]

E-Cores just shouldn't be overclocked past 4.0/4.1

Why not?

 

[Zantal]

I have limited testing with this but here is what I am seeing right now

E-Cores and P-Cores and ring share the same voltage controller, so basically if E-Cores want to run at 42x they will override voltage for P-cores and ring frequency if they are set to run at lower voltages, the same is true for P-Cores juicing E-cores up if they require more voltage.

I was struggling to run P-Cores at a lower voltage than 1.45 because of this very reason (for a 52/53x on P-cores) and setting offsets too low would cause The E-Cores to crash, funnily enough they never caused a blue screen when they crashed but it was very apparent when running benchmarks or stress tests, Cinebench R23 would halt after 3-5 seconds and give me an error, I thought it was the memory controller/Ram at first but after bringing down E-Core multiplier to 40 it completes a first run at 5.2 ghz all cores with 29.2k on a partial run (slowly going down as after a while it will thermally throttle a little)

This is all on Air with a Noctua nh-d15, and I am surprised it keeps temps barely under 100C for the first 30 seconds at over 300W lol
If i can get the voltage a little bit lower and still have it stable it will not throttle as much and the performance gained from undervolting should be better than having an extra 100/200 Mhz on the E-cores.

 

[EarthDog]

4.1 GHZ on Ecores was literally set it and forget it at 1.25V for me. That same voltage allows for a 5.1GHz "stable" overclock across all Pcores. Perhaps overclocking with an override or offset is better than auto as it won't 'add voltage' automatically? Not sure. But I don't think that's the limit of the Ecores... my Pcores crap out at that voltage, not my Ecores. I haven't tried 5.2 GHz on the Pcores though... but 1.25V (actual, using a multimeter) is the limit of my 3x120mm AIO during an AIDA64 stress test (CPU/Cache/FPU checked off).

I think you just need to do one at a time. Find your stability point for each (E and P cores) and see where they end up together at the same voltage. Again, in my experience, Pcores seem to run out of steam than the scores at the same voltage.

 

[Zantal]

Why are our experiences so different? What board do you have?
My VID dropped the moment I backed down the E-Cores as they have their own v/f curve and I supposed it was overriding the p-cores. I got them stable once already one at a time but 1.45/1.5 v at 52x and 42x e-cores is just too much for my liking and as I said being on air it throttles really quick at that voltage.
Then when I started messing with downvolting using adaptive voltage and a -0.075v offset my e-cores got unstable quickly. Auto will just follow the v/f function +/- offset

I can't wait to get back home and try 53x at 1.4v and fine tune the vdrop that is currently holding me back to successfully run stable 5.3 all core.

Try running e-cores at 42/43x and see if anything changes.

 

[EarthDog]

I tested that on the MEG UNify. TBH, after finding that 5.1/4.1 GHz 'sweetspot' (for testing), I haven't done too much on the overclocking front. I'm temperature limited though at those settings.

On my next board, I'll give your settings (42x Ecore) a try.

When you say VID, I can't help but think of 'stock' voltage as a VID is what comes from the process. You're changing Vcore. Again, I use an override voltage so nothing should change... but I'll check.

 

[Zantal]

Sorry my bad, i erroneously use vid vcore meaning vcore but I use VID readings to check more or less where I am at in windows.
Anyway setting vcore override will bypass the v/f curve entirely, when using adaptive it's basically default behaviour and you can offset that to undervolt, it was driving me crazy as even when reducing multiplier on pcores I would get the same voltages just because ecores were requesting it.

 

[Zantal]

HWInfo is reporting IMC VDD @ 1.652, I can't find any info about safe operating ranges for this, it scales directly with Dram voltage, at default jedec specs it will be 1.35 which is exactly 0.3 less than my current set dram voltage of 1.49, there seems to be no setting in my bios to unlink it, is this a cause for concern?

 

[Woomack]

I see that naming has been changed in this generation and hwinfo64 also calls some sensors in a different way. Motherboards have it called in a different way in BIOS too. I don't remember if IMC VDD was related to SA or something else. I don't have it under IMC VDD in BIOS on the mobo I have in tests right now. However, hwinfo64 shows 1.20V on my DDR5 mobo, so 1.652V seems a bit too high. I assume it should be still safe as under auto there are some limits that motherboard manufacturers are setting as a safe limit. Well, in the worst case you still have a warranty

Some random comments:
- ASUS Strix Z690-I Gaming is training memory much better than Gigabyte. Still not perfect but does that much faster and with better results.
- 12600K has a bit worse results in memory bandwidth and latency than 12900K so can't compare it directly. 12600K also has lower cache bandwidth but I assume it's because of multithreading.
- ASUS is also marking ratios that don't work. I mentioned it some time earlier that on Gigabyte some memory ratios are not working, so ASUS is actually marking them as greyed-out in BIOS. For example, 5100, 5300, 5467, 5500,5700, and 5733 won't work and when you pick any of them, then will run at a lower frequency. If you pick 5500 then it will work as 5400, when you pick 5733 then it will work as 5600 ...

 

[Zantal]

I see that naming has been changed in this generation and hwinfo64 also calls some sensors in a different way. Motherboards have it called in a different way in BIOS too. I don't remember if IMC VDD was related to SA or something else. I don't have it under IMC VDD in BIOS on the mobo I have in tests right now. However, hwinfo64 shows 1.20V on my DDR5 mobo, so 1.652V seems a bit too high. I assume it should be still safe as under auto there are some limits that motherboard manufacturers are setting as a safe limit. Well, in the worst case you still have a warranty

Also latest HWInfo beta (that specifically supports my motherboard) doesn't even list it in the sensors, and that voltage scales 1:1 with dram voltage with a 0.15 offset, my DDR4 sticks are set at 1.49v, if I set them at xmp IMC VDD will be 1.5v
I have seen some of the higher end motherboards will feature a voltage control name that is similar to what reported here, mine just doesn't have such an option or it is named differently, it is frustrating that there is no info about this voltage or similar not even in intel's spec pdf. I can exclude also that it is related to SA, as I have been going through all my voltages trying to get them down, CPU VCCSA at 1.2 is plenty even for the overclock I am running.

This motherboard is also driving me crazy, you literally cannot force a voltage lower than it's minimum for vcore not even with an offset, I hope they come out with another bios

Update:

I am starting to think IMC VDD is being misread as a different sensor, HWInfo 7.15.4625 officially supports my motherboard and in the position of that sensor value there is Dram voltage reading instead.
If the newest beta for HWinfo supports your motherboard could you try and update aswell and check with me?

 

[Woomack]

On ASUS Z690-I Gaming, hwinfo64 shows 1.20V at auto 5200/XMP when in BIOS is 1.15V. At 5400 manual is 1.20V in hwinfo64 and 1.19V in BIOS. This is on DDR5 so voltages will be probably different. I will check it on DDR4 during the weekend. No time to set another motherboard earlier.

One little update, my Micron-based DDR5 couldn't run below CL40 at 5400 on Aorus Z690 Master while on ASUS passed memtest at 5400 CL36 and now I'm testing at CL34. Weird timings like 34-40-36 and 1.27V. On Gigabyte it was working fine up to 1.4V. On ASUS doesn't like much above 1.3V.

Edit:
5400 34-39-32-70 1.27V passed memtest. If I set tRCD at 38 then it doesn't boot or crashes quickly in the OS.
5600 boots but only at CL42+ and has some stability problems. Maybe I'm missing something.

 

[Zantal]

5400 34-39-32-70 1.27V passed memtest. If I set tRCD at 38 then it doesn't boot or crashes quickly in the OS.
5600 boots but only at CL42+ and has some stability problems. Maybe I'm missing something.

We are missing good DDR5 kits and updated bios, at first I ordered some kits, but soon realized it was not worth it, I think you'll have to wait end of next year for good overclocking capability
After some tinkering with tertiary timings I managed to get a respectable 67000MB/s and 45ns latency all at 1.5v, can't see the other values because I am not going to buy AIDA64.

Also I finally caved in and went with an AIO for the cpu, it is just impossible to sustain an overclock in any benchmark (not even real world work stuff I do) and it constantly hits 100C
Got a 30.050 run in cinebench r23 (partial run, will be lower after a while) at 5.2 and 4.3, next step is getting 5.3 stable at an ok voltage, 5.4 is doable in games (so far only tested guild wars 2 as I literally play only that lol)

Hopefully this weekend I can finish this overclocking project

 

[Woomack]

With RAM is one problem right now. The only kits that are widely available are based on Micron IC so 5600 max (even after OC). It's Kingston Fury Beast 4800/5200 and harder to get, Crucial/no heatsinks 4800. In press releases, other brands are showing 6000+ kits but distribution can't even tell when any DDR5 will be available in stores. Most users don't really care that this or that brand has 6400+ DDR5 when no one can buy them and won't be able for next 2-3 months, if not longer. Until then, prices will go up by next 20-30%. About two weeks ago, it was possible to buy 2x16GB Kingston Beast 5200 for less than Samsung based 2x16GB DDR4-3600/4000 usually cost. In this week, I see that all DDR5 disappeared. I know from Kingston that they will deliver some more soon, but stores will probably raise prices like they did with graphics cards.

Some other little updates:
12600K keeps 4.5GHz+ on all cores in Cinebench R23 ... using Noctua NH-L9i cooler It goes up to 90°C+ at the end but there is no throttling. I know it will throttle in long stability tests with AVX but it's not typical usage. It makes 12600K a great option for HTPC/SFF.

Btw. if you bought ASUS motherboard then you would get AIDA64 license for a year ... I know it's not worth it as ASUS bumped prices way too much.

Btw2. I made my DDR5 kit work at 5600 CL34-44-32 when it doesn't even boot on ASUS at 5600 and most settings at auto, or it works at 5600 on Gigabyte but only at 44-44-44. I have to work on stability as right now it's passing all benchmarks, but is crashing in longer stability tests.

I wanted to take some screenshots yesterday but during tests the PSU blew up and I had nothing free around to replace it. Probably one of the main caps, I didn't check, but there was a "bang" and some smoke.