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

[Zantal]

'm pretty sure this is about what I can get out of my system but there may be tweaks I don't know about. I have E-cores disabled but heat still keeps me from getting 5.2ghz without throttling. 5.1ghz @ about 1.25v with ram at 3600mhz cas15 1.4v and cpu SA at 1.2v but I can't do 1T for the life of me. I can live with it. I think somebody told me once that 32gb dual ram kits have a hard time with 1T and I couldn't do it on my last system either.

My current b-die won't do 1T unless I run them in gear 2, which is pointless. I think I have read somewhere it's just a b-die thing
And since you also have g.skill those timings make me think you have a lower binned b-die.

My previous kit could do 1T at 4k but I just couldn't get it stable, and it wasn't b-die (never really looked into it) so I just sold them.
It really depends on the sticks.

With e-cores disabled you "should" be able to get 5.2 at 1.3 -> 1.35 (full load reading), for me that was the limit on air on my NH-D15, and you can't get any better cooling than that, still it was hitting over 90C, since I can get about 10C more ambient during the summer I just switched to an AIO.

One solution would be to enable TVB and downclock by 1 or 2 ratio over 60/70C, so you can safely game at low load and when you need to really pump you can do it at a more safe temperature and voltage.
I am running a low unstable voltage for 52x but it never crashes, not even with quick load transients as it goes down to 51x with stable voltage of 1.28
(with E-Cores enabled) and it averages a temp during gaming of 40C with max 60C spikes on a beefy 420 AIO and 75 to 80 during max load

To be 100% stable you would need to add an avx offset (I am aiming for 4.9 during avx that's an offset of 3x at 52 boost) in no way you will be stable during AVX workloads with that voltage, y-cruncher is a good way to test stability since it has a very mixed workload and will have different and quickly varying load scenarios, I was stable in an AVX stress tests, but not in y-cruncher, 1 level higher LLC fixed the issue.

You can skip this step, but I still recommend it as you never know where and how heavily these instructions are implemented and you won't know why you suddenly got a random BSOD

 

[Brando]

I was doing okay at 5.2 GHz at 1.3v for the regular Intel stress test but as soon as I do the memory stress test the heat goes through the roof then thermal throttling. I don't know what they're doing different for the memory stress test but that one crashes me when the other ones don't and I'd rather deal with the worst case scenario now so I can not worry about it too much later. I have it at -3 AVX offset. I'm not super worried about getting 1T on the ram. I don't think it really makes much difference in the scores much less real world performance. Will look into tvb thanks

 

[Woomack]

I'm not sure if it's just my feeling or it is real but I see that Z690 is not optimized for Samsung IC. It can be also related to my motherboards. This is why I have that feeling:
- ASRock Z690 Extreme and Riptide work up to 4000 CL14 CR2/Gear 1 with Samsung IC but max at Gear 2 and any timings, even very relaxed, is 4400 on the Extreme while the Riptide mobo can't even boot at 4200+ with Gear 2. Hynix works up to 4800+ without issues.
- Biostar Z690 has problems with some kits based on Samsung IC at XMP 3600. Manually I couldn't boot at 4000 with Samsung while HyperX/Hynix kit works fine at XMP 4600.
- My ASUS Z690 mobo has no Samsung-based DDR5 on QVL what suggests that ASUS had problems getting kits for tests.
- My Gigabyte Z690 has problems with Samsung-based kits and it just received BIOS fix to work better with G.Skill/Samsung kits. It still works worse than ASUS.
- There are some reports from random users that they have problems with Samsung-based kits on their motherboards. Random users and cases so can be users' fault.
- Samsung IC loses stability or has problems with booting when voltages are too high (but still should be acceptable/safe). The same with DDR4 and DDR5. For example, with DDR5 I see stability loss at more than 1.4V. With DDR4 I can't see stability improvement above 1.55V. No problems like that with Micron DDR5 or Hynix/Micron DDR4.

Re Gigabyte Z690 Master - I could barely boot at 6600 with Samsung IC and the latest BIOS, while it's stable at 6400. The weird thing is that it works at 6400 using an OC profile but it doesn't work when I set the same settings manually.
ASUS Z690I-Gaming runs up to 7000 but is stable at 6600 with the same RAM.

In the end, one stick from my G.Skill TridentZ 5 6000 CL36 kit died last weekend.

 

[Brando]

In that case I think I'll just call it a day on the oc until noctua releases an nh-d16 with a special base made just for alder lake that has at least 5c better temps. It's my understanding that ddr4-3600 cas15 is just about ideal and I got 3600 with cas14. I probably also need less bandwidth due to not using e-cores.

 

[Woomack]

All popular Noctua coolers have LGA1700 mounting kits and I'm not sure if Noctua releases anything larger than the NH-D15 anytime soon. I guess I would know as I have contact with them but maybe I missed something. NH-U12A performs great and it was released in black version maybe 2 months ago. It's pretty much my favorite cooler as it fits most cases and performs not much worse than the NH-D15. Better than these coolers is only something like better series of 280/360 AIO (or custom wc).

 

[Brando]

I could see them tweaking the base for better contact. Not sure if it would help or not.

 

[funnyperson1]

@opt33 May I ask where you got the ASRock Timing Configurator and MemTweakIt applications from? I can't seem to find the updated versions for Z690 anywhere.

A lot of reviewers and other folks are saying there isn't much point to overclocking the 12700/12900K since you can't go very far without hitting a heat and power wall. I haven't done too much messing around, but I tend to agree on this, I think undervolting and optimizing within a fixed envelope might be more interesting. At stock speeds I'm pulling 200W with Prime95 torture test with Package temps in the 80s with a 240mm AIO in a room with ambient below 70F, at 5GHz I was at 88C package and power throttling at 243W. Before messing with XTU I had set my motherboard power limit to its max (265W) and that was throttling at 5.1 but I didn't check if it was power or thermal.

Anyways, I'm not sure if I'll try to uncap things and push it much past 5.1GHz all core, I don't really trust my motherboard to live through 300-400W stability tests and I'm not going to win any screenshot wars with my 240AIO. I might push it a little bit more since my computer is currently in an unheated room and ambient is around 50-60F if I don't turn the space heater on. I haven't used XTU on a modern Intel platform and the configurability is impressive, I might be tempted to find the max stable OC for each core individually. I wouldn't be shocked if one of the preferred performance cores could do 5.5+.

Undervolting is definitely fun though, so far it looks like -100mV is stable at 4.7GHz all cores in Prime95. This runs me around 180W and 78C. I kinda want to find the max stable undervolt and try to increase clock speed from there.

 

[Zantal]

You are correct, it feels this cpu was already pushed to the limits stock, I can get this stable at 5.3 at 1.38v load and it's very close to thermal throttling and it will throttle after a longer stress test
(I have hit 380w for a couple of seconds once)
Not worth it in the long run, in a couple of months it would sure be degraded even if I just game on it, at idle or in a game it goes to 1.42/1.45v which is a lot.
Final overclock is 5.1 at all core and the 2 best cores at 5.2 (but they will go down to 5.1 over 50C with TVB) at 1.288v under load, considering stock it would be 1.25/1.26 at 4.9ghz I call this a win

I hope they don't sell the 12900ks at retail because in the back of my brain there is already that little devil whispering me to buy it as soon as it releases =/

 

[funnyperson1]

I was able to get 5.2/3.8 stable enough for some cinebench runs which was good for top 10 with this CPU on hwbot. I must have upset some folks this morning because I’m pretty sure I was 5th or 7th when I submitted it late last night.

https://hwbot.org/submission/4905799_

Also derbauer made a nice discovery that the high end Asus boards allow overclocking non K SKUs with BCLK, he could get 5.2 with the 12400:

That could be pretty exciting if Intel doesn’t squash it and it ever becomes available on sub $400 boards.

 

[Brando]

I decided to upgrade to win 11 and give the e-cores a chance. I ran intel extreme tuning utility and temps only went up 1-2c vs no e-cores but xtu froze my pc despite being stable without e-cores. The e-cores are at stock speed so I'm not sure what the issue is unless having them on is robbing the p-cores and imc of power and causing instability that way. What should I try next?

 

[funnyperson1]

I decided to upgrade to win 11 and give the e-cores a chance. I ran intel extreme tuning utility and temps only went up 1-2c vs no e-cores but xtu froze my pc despite being stable without e-cores. The e-cores are at stock speed so I'm not sure what the issue is unless having them on is robbing the p-cores and imc of power and causing instability that way. What should I try next?

Maybe a VDroop issue? Might try bumping VCore a tad to see if it stabilizes. I've only been playing with the voltage offset in XTU rather than setting a fixed voltage at a fixed frequency. My CPU seemed happy up to 5.2/3.8 with the stock voltage curve (I think it was around 1.35-1.4V at 5.2/3.8)

 

[EarthDog]

Vcore is shared between e and p cores. Perhaps you need a bit more of that if you're overclocked on the p cores.

 

[Brando]

I tried 1.26v and it lasted longer during the stress test but it was way too hot. Next I tried going back to 1.24v and lowered p cores from 5.1 to 5.0 and made it through an intel xtu memory stress test topping out at 90c. I'm thinking overall performance will be better with 4 extra e-cores than with only p-cores at an extra 100mhz but maybe I'm wrong.

 

[Woomack]

@Brando, it should run stable up to 100°C when the settings are correct. It may throttle but still shouldn't crash in tests. Maybe you are missing something or the motherboard is setting something under the auto.


I had to RMA one G.Skill TridentZ 5 DDR5-6000 kit ... already got the money back and I have another, the same kit. Somehow, the second kit overclocks a bit worse but not much. What's also weird, it doesn't work at XMP on Gigabyte Z690 Master (last BIOS releases were supposed to improve compatibility with G.Skill/Samsung RAM). I don't remember if I was trying XMP stability on the first kit. In short, on Gigabyte I can't stabilize anything 6000+, manual or XMP. There are always single errors in memtest or it's crashing in windows tests.
The same as with the first kit, on ASUS it boots up to 7000 CL38/40 and passes easy tests but isn't stable at 6800/7000. Even at 6600, there are problems to make it pass memtest at relaxed timings. It passes windows tests but not memtest (it's built-in the ASUS BIOS).
It also doesn't react well at more than ~1.37-1.40V. It boots maybe one step higher but isn't fully stable at higher voltages. I see some guys running Samsung at 1.5V but on my kits, high voltages gave me nothing.

 

[EarthDog]

Well see some solid results on adata 6000 chips when we get the next review up.

 

[Woomack]

I guess that Hynix IC, not Samsung.
Somehow I feel it has a lot to do with the PCB and power section design on the RAM itself. I heard about other users having problems with G.Skill too. Nothing confirmed, as always, can be the user's fault too.

 

[EarthDog]

Correct. This kit uses Hynic ICs.

 

[Brando]

90c in the winter is already pushing it imo. Need to have some wiggle room for the summer. I'd like to get down to 80c tbh but it doesn't look like it's happening without water cooling.

 

[Woomack]

Do you need it to run at 5GHz+ all the time? I mean, in my daily PC I see no point to OC CPU. Well, I OC RAM as at XMP it feels like broken but CPU is fine at auto. Actually, I downgraded my gaming CPU because I wanted it to run cooler and I haven't seen the difference in performance. I went down from 5900X to 11700K and next to 5700G ... 8 cores at up to 4.6GHz and 85W.

 

[EarthDog]

Maybe lower things a smidge for summer. This way, your winter clocks won't get robbed.

TBH, my ambient temps (read: inside the house) don't change much between winter and summer. When I first started overclocking, I saw a lot of this mentioned, but since we're heat/ac users, the only worries are when the wife wants windows open and it reaches 80F outside, lol.

My 12900K is stock.... lol