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

[Voodoo Rufus]

Hmm. 12600K running iGPU for HTPC duty, in an HDPlex passive case. I'd totally do it if it wouldn't be a tremendous waste of CPU power.

 

[Zantal]

Finally after a month of tinkering I managed to finish my "stable" overclock.

Why "stable"?

Because I haven't tested with other workloads, mainly with what I use daily which is gaming (Guild Wars 2 or the occasional Raging in Star Citizen) and Developer stuff in Unity3D.
To get it stable in Unity3D was quite difficult, It would be stable in Cinebench even after 10 minutes but 10 seconds of Texture compression would kill my system instantly.

Also I upgraded to an AIO, the NH-D15 is at it's limit if you don't overclock, any overclock will get it outside of it's rated operating temperature quickly and the cooling performances will suffer even more

Final stable overclock is 53x on ALL P-Cores and 42x on ALL E-Cores, Ring set at 46x with down bin disabled, otherwise it's 36x max when E-Cores are even slightly active.
Temperatures during Cinebench barely hit 100C on some cores, and it doesn't throttle, this summer it will be an entirely different story with +15C ambient, but it should still be stable when throttling

Some quirks of my Asus board (Z690-P Prime D4):
Trying to set voltages is a pain in the A**, V/F curve cannot be touched or it won't even boot (even just setting a + 0.001v will fail)
Voltages are a pain on Alder Lake in general, Ring E-Cores and P-Cores share the same voltage regulator, overclock one of those too much and it will set a higher voltage for everything
LLC was already set at Level 4, but needs Level 6, with Level 5 you can get the occasional crash during Load

To get to my final value of 1.45/1.47 (HWInfo Reading) I set adaptive voltage with offset.
Extra Turbo Voltage will only work if you set a higher value than the original V/F Curve.

The voltage is calculated based on the SVID setting, I found it works best with "Best Case Scenario" that generally has a lower voltage curve, and added 0.065v on top of that
I also Tried "Intel's Fail safe" setting which will happily set your vcore to 1.75v at this overclock, yeah I lost a couple of years of my life there.
Even if you specify a voltage ovverride it is not guaranteed it will enforce it, just to try it out I set 1.1v and it would still get to 1.3 under load (this was before overclocking)
I might be missing something or some key option somewhere but after a month my patience with this board is spent.

My E-Cores are now stable at 4.2ghz, anything above that will just not work, you will get to boot and do stuff ,but cinebench will never complete and trying to get them stable with higher voltages will overvolt your P-Cores that in turn run hotter and will throttle sooner.

Setting E-Cores to have their multiplier Synced in bios will cause them to be highly unstable
Ring 46x is the max I was able to get with E-Cores active, 47x will boot and work until it decides it's not working anymore, the only way to get it more stable is feed more voltage into it, but as I said before
it shares voltage with all the cores, so not worth it to get it too high, I have seen people though get it stable at 48x with E-Cores disabled.

This overclock will pull 360W from the cpu according to HWInfo, Both power sockets needed.


Regarding memory overclock, as noted by everyone else Max memory controller frequency on z690 is around 2000Mhz
I thought I had some good sticks that I managed to overclock to 4000MT Gear1 CR1 with low timings of 15-14-14-28 @1.5v
It passed 15 minutes of OCCT AVX memory test, and I thought it would be ok, but after getting some cheeky crashes in GW2 I decided to investigate further and found out they will crap out randomly
So loosened timings to 16-16-16-36 and it would still give me an error randomly, sometimes during the first 30 seconds, other times it would go on and on without errors

It will find errors at XMP rating aswell, which is odd, if it's not the IMC I will change these sticks with a couple of G.Skill rated at 4000MT C16 that I am keeping a close eye on

Also another final note regarding this motherboard and sensor values.

The current stable release of HWInfo (7.14) shows a sensor called IMC VDD which is super high at 1.65v, this is a bug.
IMC VDD is a sensor present on DDR5 boards, they fixed it in the beta version and that sensor now correctly displays DRAM voltage

Forum post for reference here https://www.hwinfo.com/forum/threads/missing-sensor-value-in-beta.7752/

This is what was achieved, if anyone wants me to try other benchmarks let me know
That cinebench score is after 1 run only, It might hold it above 30k for 2 runs but even water cooling has its' limits and it will throttle after a while

 

[Brando]

Looks like I'm behind the times. I just read my first alder lake review and apparently intel is crushing it right now. Unless I'm missing something it looks like there's no real downside to throwing a 12700k into a ddr4 motherboard and re-using my gskill 3200 b-die since apparently ddr4 ram is better than ddr5 in most things I do because ddr5 runs in gear 2 or something like that? I don't really want to get windows 11 yet but couldn't I just disable the e cores to get around the core scheduling thing and buy myself some more OC headroom for the 8 P cores? Unless there's something I'm not getting it looks like a good upgrade

 

[Woomack]

DDR4 is not better but it's hard to clearly say it's worse right now. It's just that most reviews show it as worse, focusing on picked tests/benchmarks. In most tests, it doesn't make a significant difference what RAM is in use as long as DDR4 runs at 3600+ Gear 1 or 4600+ Gear 2, and DDR5 runs at 5200+ CL40 or less.
Keep in mind that most people have no idea how to set DDR5 and they run at XMP/auto settings. Also, nearly all available DDR5 are standard series based on Micron IC and highly limited in OC. On top of that, most motherboards have problems with BIOS.

For sure if you go DDR4 way then you won't lose right now, and when DDR5 will be really worth it then probably will be new chipsets and the next-gen CPUs.

There are already the first BIOS updates with scheduling fix for new CPUs, and there are software solutions too. You can disable E cores if you want.

 

[Brando]

DDR4 is not better but it's hard to clearly say it's worse right now. It's just that most reviews show it as worse, focusing on picked tests/benchmarks. In most tests, it doesn't make a significant difference what RAM is in use as long as DDR4 runs at 3600+ Gear 1 or 4600+ Gear 2, and DDR5 runs at 5200+ CL40 or less.
Keep in mind that most people have no idea how to set DDR5 and they run at XMP/auto settings. Also, nearly all available DDR5 are standard series based on Micron IC and highly limited in OC. On top of that, most motherboards have problems with BIOS.

For sure if you go DDR4 way then you won't lose right now, and when DDR5 will be really worth it then probably will be new chipsets and the next-gen CPUs.

There are already the first BIOS updates with scheduling fix for new CPUs, and there are software solutions too. You can disable E cores if you want.

Thanks, I'm not sure why I said ddr4 is better I just saw it win in a couple game benchmarks but overall it looks like it's a tiny bit slower though not enough to care. I think I may go for it. This is what I was referring to https://www.techpowerup.com/review/intel-core-i9-12900k-alder-lake-ddr4-vs-ddr5/3.html It's a 12900k not a 12700k but close enough to get a feel for it.

 

[Woomack]

Thanks, I'm not sure why I said ddr4 is better I just saw it win in a couple game benchmarks but overall it looks like it's a tiny bit slower though not enough to care. I think I may go for it. This is what I was referring to https://www.techpowerup.com/review/intel-core-i9-12900k-alder-lake-ddr4-vs-ddr5/3.html It's a 12900k not a 12700k but close enough to get a feel for it.

There are a lot of DDR5 haters on the web and those who glorify Samsung B DDR4 (also because they never tried anything else). Most of these people can't afford a new platform with DDR5 and/or simply can't buy DDR5 because it's not available anywhere anymore. As a result, you will find much more comments that DDR4 is better than anything good about DDR5. Even most reviewers from the top websites barely tested 1 DDR5 kit ... simply because there are no samples around.

Right now everything depends on the test/game and settings in which you compare RAM. It's possible to get better results on DDR4 than for example DDR5-5200/5600, but not on DDR4-3600 or lower at XMP settings. It's either DDR4-4000 CR1/Gear 1 and tight timings or 4800+ and more relaxed but still quite tight timings. Both configs require expensive DDR4 based on top IC. If you compare prices then you can find DDR5-5200 even cheaper than top DDR4.

As I already said (and you noticed) it's not worth going for DDR5 with all the problems with availability and higher than expected prices. Maybe in 2-3 months will be better but with the very high demand, I bet that DDR5 price will go high. I already see G.Skill modules listed for $1k per 32GB DDR5-5600/6000 (not available but expected to be in this price when arrive) when ~2-3 weeks ago you could buy Kingston Beast 32GB DDR5-5200 for about $300. DDR5 motherboards also cost more than DDR4.

There is also one more thing. The typical DDR5 kit used in comparisons is 32GB. The typical DDR4 kit is 16GB. I can't imagine having 16GB in my gaming PC anymore. I simply hit 14-16GB or more while playing games with some web browser tabs and other stuff in the background.


One result on 32GB Kingston Beast DDR5-5200 CL40@CL34-39-32 1.27V and some sub-timings tweaking. 12900K should give a higher bandwidth and slightly lower latency.
This setting passed memtest and AIDA64 stability test.

 

[Voodoo Rufus]

The DDR5 availability and pricing is giving me more reasons to just wait for Raptor Lake. Let all the bugs get sorted out on Alder Lake.

 

[Brando]

I definitely wouldn't pay an extra thousand for 2% more performance right now but it should be interesting when the ddr5 equivalent of b-die comes out and starts hitting 10000mhz (who knows).

 

[Woomack]

The current-gen is planned up to DDR5-8000. I don't think we will see more than about DDR5-7000 anytime soon. The highest, listed as potentially available, kits are DDR5-6000 right now. Everything higher is only a news/press release and can be but probably won't be released under expected product numbers or XMP. However, these DDR5-6000 kits suppose to do 7000+ at tight timings.

Samsung in theory released their B-die DDR5. In theory, as it should be available in stores in about 2 months. Count that most G.Skill 6000+ CL36 will be based on Samsung, CL40 at higher frequencies will be probably Hynix. No matter if it's Hynix or Samsung, expected OC seems similar as you can see on QVL of various brands and XMP profiles. The highest frequency that I saw was made on Hynix. Hynix also tolerates the highest voltages.
In some stores like Amazon, you can already order 6000 CL36 and 40 kits that will be shipped in Jan/Feb.

Users have to realize one thing. DDR5 is like quad channel vs dual channel about 2-3 generations ago. So you get higher bandwidth but also higher latency. It's generally faster in multithreading but the same or slightly worse in some single-threaded tasks.
I guess that those who are after the top performance at all costs shouldn't look at Z chipsets but wait for X to hit the stores. Then will be more channels and much better results. If you spend a ridiculous amount of money then better for the top series than mid. Z chipsets were always a middle shelf, just in the last years priced way too high.

 

[Voodoo Rufus]

Considering GPU's have been on GDDR6 for a while, I wonder how long we'll have to wait for desktop DDR6 ram. I can sit happy on my 9th gen for quite a few years I think. Maybe even skip the DDR5 generation?

I kinda wanted to play around in the HEDT space and grab a 10980XE, but the performance just for gaming just isn't there, especially on the dead-end X299 now. AMD owns the space with Threadripper, but I'm not going to waste one of those on casual gaming either.

 

[PolRoger]

It seems that Micron has been running some kind of year end sale/special for some of their Ballistix DDR4 kits. NewEgg/Amazon/Micro Center and other stores are OOS for most SKU's but my local Micro Center still had a single kit of 2x8GB 1R 4400C19 Ballistix Max. I was more interested in 2x16GB 2R kits for my 2-DIMM slot ITX motherboard but they cost ~twice as much and the few remaining kits that were available sold out before I could make a decision on purchasing more memory.

This kit booted right up and runs fine at XMP. With my early Samsung B-die kits 4400 was the max that I could boot into Windows. With this kit I'm able to boot into windows @4533/4600/4800. Both 4900 and 4933 post to BIOS at 4800 speed while 5000/5066/5100 fail to post to BIOS.

4800C19 1.45v DRAM .950v VCCSA:



EDIT: Added AIDA64 memory benchmark...

P49E38 4800C19:

 

[Voodoo Rufus]

Oh joy, another super-binned CPU. I'm kind of a sucker for these.

https://videocardz.com/newz/intel-c...ghz-all-core-frequency-allegedly-in-the-works

 

[Zantal]

I discovered a bug that is likely to be present on all overclocking software (possibly)

Reproduced on Intel XTU and Asus AISuite V3

Changing cache frequency in intel XTU will disable ring scaling based on which cores are active.
When E-Cores are active ring frequency drops down to 3.6ghz for a good reason, as I discovered a while ago, I was running 4.6 ring with E-Cores active and It would crash randomly after a while.

The problem is that the interface in intel XTU or AISuite does not support this behaviour, when you set a value it will be that value, and your program or windows WILL eventually crash.

With AISuite it is on an another level entirely, I never touched ring frequency in there but it would apply it anyway when making modifications

The fun part has yet to come though.

It disables ring downclocking in a way that you can't turn it back on, not even in bios, i even forced ring downclocking in bios and it would still go all the way to 4.7ghz with E-cores active.

The only way is to load default bios settings or another profile (thankfully I saved my stable OC profile a week ago)

And things don't end here, this is unrelated to the above issue but still some interesting quirks about the platform.

My Asus board has the option to Park E-Cores by toggling the "Scroll Lock" button, I was using this method during games to keep the ring frequency at max 4.7
The problem is though that from time to time these cores will become active for no reason, do a tiny bit of work and then go back to sleep, this will cause the ring to momentarily go down to 3.6 and then shoot back up (causing some micro stutters occasionally but barely noticeable)

but wait, there's more!

You apparently can get the ring more stable with E-Cores active in intel XTU because there is a setting to increase cache voltage offset specifically, I tested this a few times with 0.1v offset and it works, the moment I put it back down the app crashes after a few seconds, in the bios such setting is not present, and I was under the impression that all core voltages are linked together anyway.

Alder lake is a nice rebound for Intel, but every day that I am dealing with this cpu it feels like it was kind of . . . rushed?

 

[Culbrelai]

So undervolting is the way to go? My 12900k throttles under a NH-d15 and it seems this is common. Even in my quite cold room in a winter climate. Oof. What's a good low point? 1.2v?

 

[Woomack]

It seems that Micron has been running some kind of year end sale/special for some of their Ballistix DDR4 kits. NewEgg/Amazon/Micro Center and other stores are OOS for most SKU's but my local Micro Center still had a single kit of 2x8GB 1R 4400C19 Ballistix Max. I was more interested in 2x16GB 2R kits for my 2-DIMM slot ITX motherboard but they cost ~twice as much and the few remaining kits that were available sold out before I could make a decision on purchasing more memory.

This kit booted right up and runs fine at XMP. With my early Samsung B-die kits 4400 was the max that I could boot into Windows. With this kit I'm able to boot into windows @4533/4600/4800. Both 4900 and 4933 post to BIOS at 4800 speed while 5000/5066/5100 fail to post to BIOS.

4800C19 1.45v DRAM .950v VCCSA:

View attachment 215156

EDIT: Added AIDA64 memory benchmark...

P49E38 4800C19:

View attachment 215157

Check something like CL18-24-22 or CL18-25-23. Microns like high tRCD and will scale this way up to 5200-5400, depends on luck. Check even 27 as it won't cause significant performance drop while let to set a much higher clock (something like 19-27-23 may still work up to 5200+). Also, they are not really scaling well past 1.65V and usually 1.70V is max worth to set, even for benching. On the other hand, 1.60V is what some brands used in the XMP at DDR4-5000+ so it seems safe for 24/7.
2x8GB is probably Micron E. Better would be 2x16GB Micron B but in both cases it will be single rank. I have 2x16GB Ballistix MAX 4400 C19 that went up to 5500 CL20 on AMD/B550. In general, Micron works better on AMD, Hynix works better on Intel. This is how manufacturers are tuning their mobos/BIOS in the last years. Top mobos OC well all IC.

Btw. I was testing 2x16GB TridentZ Royal 4400/Samsung B dual-rank and it doesn't even boot at XMP on Z690. On Z590 it works fine.

There is also one thing barely anyone noticed. Patriot released Steel series DDR4 (2x8GB) at 3600 CL14-16-16 1.45V and 4000 CL16-16-16 1.45V. I have both kits in tests right now. Would be better if these kits were 2x16GB but Steel series DDR4 used to be quite cheap, so I guess it can be an interesting option for some overclockers/gamers.

I discovered a bug that is likely to be present on all overclocking software (possibly)

Reproduced on Intel XTU and Asus AISuite V3

Changing cache frequency in intel XTU will disable ring scaling based on which cores are active.
When E-Cores are active ring frequency drops down to 3.6ghz for a good reason, as I discovered a while ago, I was running 4.6 ring with E-Cores active and It would crash randomly after a while.

The problem is that the interface in intel XTU or AISuite does not support this behaviour, when you set a value it will be that value, and your program or windows WILL eventually crash.

With AISuite it is on an another level entirely, I never touched ring frequency in there but it would apply it anyway when making modifications

The fun part has yet to come though.

It disables ring downclocking in a way that you can't turn it back on, not even in bios, i even forced ring downclocking in bios and it would still go all the way to 4.7ghz with E-cores active.

The only way is to load default bios settings or another profile (thankfully I saved my stable OC profile a week ago)

And things don't end here, this is unrelated to the above issue but still some interesting quirks about the platform.

My Asus board has the option to Park E-Cores by toggling the "Scroll Lock" button, I was using this method during games to keep the ring frequency at max 4.7
The problem is though that from time to time these cores will become active for no reason, do a tiny bit of work and then go back to sleep, this will cause the ring to momentarily go down to 3.6 and then shoot back up (causing some micro stutters occasionally but barely noticeable)

but wait, there's more!

You apparently can get the ring more stable with E-Cores active in intel XTU because there is a setting to increase cache voltage offset specifically, I tested this a few times with 0.1v offset and it works, the moment I put it back down the app crashes after a few seconds, in the bios such setting is not present, and I was under the impression that all core voltages are linked together anyway.

Alder lake is a nice rebound for Intel, but every day that I am dealing with this cpu it feels like it was kind of . . . rushed?

We are beta testers for some years already. Intel (and AMD) used to test new generations for at least 18 months before release. Now they aren't even sure how everything works and nothing has full software support. When the 11th generation was released, most manufacturers started to work on the 12th gen and dropped any support besides any critical issues for Z590 motherboards. Z690 is supposed to be better in everything but every 1-2 weeks we hear about some issues or things that weren't planned. Actually, E cores work better than Intel was planning. On the other hand, there is no full BIOS and software support for all new things. We have seen 2-3 BIOS updates for some motherboards in the last 2 weeks. One with new microcode, one fixing some stability issues, and one with DDR5 stability and performance improvements.

So undervolting is the way to go? My 12900k throttles under a NH-d15 and it seems this is common. Even in my quite cold room in a winter climate. Oof. What's a good low point? 1.2v?

Typically, 1.25V seems optimal, but depends on the used cooler. Some motherboards already have improvements in BIOS and for example, ASRock Z690 Extreme keeps lower voltages (up to about 1.3-1.35V) and the CPU doesn't throttle under load.
Funny thing is that even the smallest Noctua NH-L9i can handle these CPUs without significant performance drop caused by throttling. I was testing 12600K with that cooler in the last days and in AVX tests, the CPU could still run between ~4.2 and 4.5GHz while in mixed load/gaming tests it was between 4.5-4.9GHz. It's a significant improvement compared to the last generation.

 

[PolRoger]

Check something like CL18-24-22 or CL18-25-23. Microns like high tRCD and will scale this way up to 5200-5400, depends on luck. Check even 27 as it won't cause significant performance drop while let to set a much higher clock (something like 19-27-23 may still work up to 5200+). Also, they are not really scaling well past 1.65V and usually 1.70V is max worth to set, even for benching. On the other hand, 1.60V is what some brands used in the XMP at DDR4-5000+ so it seems safe for 24/7.
2x8GB is probably Micron E. Better would be 2x16GB Micron B but in both cases it will be single rank. I have 2x16GB Ballistix MAX 4400 C19 that went up to 5500 CL20 on AMD/B550. In general, Micron works better on AMD, Hynix works better on Intel. This is how manufacturers are tuning their mobos/BIOS in the last years. Top mobos OC well all IC.

Btw. I was testing 2x16GB TridentZ Royal 4400/Samsung B dual-rank and it doesn't even boot at XMP on Z690. On Z590 it works fine.

There is also one thing barely anyone noticed. Patriot released Steel series DDR4 (2x8GB) at 3600 CL14-16-16 1.45V and 4000 CL16-16-16 1.45V. I have both kits in tests right now. Would be better if these kits were 2x16GB but Steel series DDR4 used to be quite cheap, so I guess it can be an interesting option for some overclockers/gamers.

I went back took another look and I was able to boot to BIOS and get into Windows @5200.

5000/5066/5100 on this board will still fail to post and requires a CMOS CLR. I thought that perhaps maybe one time I got the board to post @5066? but I'm not so sure now as most other times I've tried it just fails to post. 5200 seems to be working and is repeatable. Auto settings must be quite loose as AIDA64 memory benchmark is showing worse results than @4800.

 

[Woomack]

Play with sub timings, I guess that under auto are too relaxed settings. On AMD or Z590 I had like 5GB/s higher bandwidth and lower latency at 5200+ than at 4800. Also if you can boot with tRCD 22 then you have a pretty good kit. Most Microns won't even boot so high at 22. If you get problems with stability then set 24 and tighten other timings.

 

[wtsds]

I think I mentioned it in some posts. Max Gear 1 on ASRock and MSI is 4000. Biostar has problems at 3600+. Max CR1 depends on the motherboard but with Gear 1 will be also 4000. I wasn't testing ASUS and from Gigabyte I have only DDR5 mobo. I think that PolRoger could make 4000 Gear 1 on his Gigabyte.
I guess that you could push Gear 1 to something like 4100-4133, but there is one problem. At least on ASRock there is no memory ratio for 4100-4133 so the next ratio is for 4200. At 4200 I couldn't make it boot with Gear 1.

I had no time to check high memory clock on Z690 with Gear 2, but on Z590, 3733 Gear 1 was slower than 4600+ Gear 2 and I could push memory to 5000+. Barely anyone is testing RAM at high clock and Gear 2 because most are reading posts/articles of clueless people who without tests repeat that Gear 2 and high clock is bad. On Intel, performance loss is not as significant as on AMD. I don't want to say what is better or worse but it's worth to test high clock with Gear 2 and compare. There are always tests/games that react better to low clock/tight timings and high clock/more relaxed timings.
High clock/Gear 2 is a good option for all who have memory kits based on Hynix A/D or Micron B. In both cases, can run them higher and at lower voltages than Samsung.

This is my particular case. I have a memory Adata xpg D50 2x16gb 4133mhz cl 19-23-23-45 and another profile with 18-19-19-39 at 4000mhz. On the gigabyte board I can't get into windows with 4000mhz with these timings. I can by 1.62v of ram, 1.45 SA vddq 1.5v and blue screen before opening windows.

What did I do? I went up the memory frequency as far as it went to know the limit. With these 19-23-23-45 I can get 4600mhz with 1.5v ram, 1.25v SA and 1.3Vddq

The best I've come up with so far is 18-23-23-43 4600mhz with tighter secondary and tertiary timings with 1.55v of ram.

But I can't boot above 4600mhz, no matter how much slack I give my latencies.

I would like to know if someone like Gigabyte board managed to boot at 4800mhz and could share their bios configuration so I can get a better idea.

 

[Woomack]

I wonder what IC is in this ADATA kit. Samsung would have a hard time even boot at 4600. Hynix would probably need more relaxed timings. Micron, depending on IC quality, is supposed to run higher without big problems so can be limited by the motherboard (or BIOS).

 

[PolRoger]

I wonder what IC is in this ADATA kit. Samsung would have a hard time even boot at 4600. Hynix would probably need more relaxed timings. Micron, depending on IC quality, is supposed to run higher without big problems so can be limited by the motherboard (or BIOS).

I believe that the kit is Micron B-die with a 2x16GB 1R configuration. He is from Brazil and was posting over on a OCN thread that is mostly about ADL and running/tuning Samsung B-die DDR4 @4000(+) Gear 1. He is running on a Gigabyte Z690 Aorus Elite DDR4 motherboard.

https://www.overclock.net/threads/o...emory-overclock.1794799/page-73#post-28914819

https://www.overclock.net/threads/o...emory-overclock.1794799/page-73#post-28914890

https://webapi3.adata.com/storage/downloadfile/datasheet_xpg_spectrix_d50_ddr4_rgb_dram_20210517.pdf