Discussion Alder Lake - Builders Thread

[AdamK47]

This thread is for those of us that own or are looking to get an Alder Lake CPU.

Bought mine on release day. 12900K. Memory selection was limited to only Crucial DDR5 4800 at the Micro Center I went to. Bought two 2 x 8GB kits. Motherboard is a Gigabyte Z690 Aorus Master. I did buy a 360mm Lian Li Galahad 360mm AIO. Ended up returning it since the Lian Li Socket 1700 backplate that I bought separately worked well with my NZXT Kraken X73.

 

[coercitiv]

I gave up on building around mITX or mATX due to a combination of both bad prices and availability, so here's my first ATX build in more than a decade. I was aiming for a smaller form factor due to height concerncs, but in the end I decided to invest in a new case that can trade height for width - O11 Dynamic Mini from LIan Li. The upside is this was a very easy build in a flexible case, and I have 2 more NVME slots on the motherboard.

The build is as follows:

• 12700K tempered by my old NH-D14
• MSI Pro Z680-A DDR4
• 2x32 GB 3600 CL16 @ 3900 CR1
• Vega56 Pulse from Sapphire
• Corsair SF750 Platinum

A fun little fact considering the same SSD was used on both systems - i7 8700 & i7 12700K:



The sequential speeds are of less importance and far less relevance considering the move from PCIe 3.0 to 4.0 for the SSD.

Thermal wise it falls exactly where I was expecting, will tune my fan noise for 150W TDP. I'm running an unorthodox cooler orientation with the fan pointing up because I can I know the MSI board overvolts a bit on Vcore and also trashes System Agent voltage with XMP enabled, so I'm currently running a fixed -40mV Vcore undervolt and fixed SA voltage, when I have time I'll do proper undervolting and proper RAM tuning.

MSI UEFI seems unable to disable E-cores, in the sense that the option exists but does nothing in practice. Will probably get fixed on next UEFI revision.

 

[Markfw]

I have a E5- 2683v2 14 core ES, runs at 2.5 ghz. Do you think a 12700k would be faster in heavily multithreaded apps ? Like 28 individual tasks of a similar nature at a time ? I mean just the 8 P cores @5 ghz vs 14 cores@2.5 I think it would win.

Trying to talk myself into replacing it, since I want to own an alder lake to see for myself how they do. I have a coolermaster 612-2, and coolermaster will not provide me the AM4 adapter kit, but its so universal in the mounting kit, that it might work on a socket 1700 ???? It works on socket 1151,1153,1366,2066 with the hardware I have.

Also, power. the Xeon is 22nm and 120 watt TDP. How do you think in reality it will compare to the 12700k at stock, or even if I turn the power down a shade ?

 

[JoeRambo]

MSI UEFI seems unable to disable E-cores, in the sense that the option exists but does nothing in practice. Will probably get fixed on next UEFI revision.

This link contains quite a few beta BIOSes for variuos MSI Z690 motherboards, if You feel adventurous, You could try one for yours.

Beta/MP - Google Drive

On Edge i have no problems disabling, it is counter intuititive in bios - but you do E Core control "Enabled and disable all" item and that's it.

 

[coercitiv]

This link contains quite a few beta BIOSes

No thanks, will wait for a proper release. I know the Beta will fix it, but I'm not in that much of a hurry. (no hurry at all actually, all my work tools work fine as is)

On Edge i have no problems disabling, it is counter intuititive in bios - but you do E Core control "Enabled and disable all" item and that's it.

I also assumed this is limited to my board model or a subset of MSI boards, not all of them.

I used the same setting as you described from the get-go, as the "Active E-cores" option is only meant for partially disabling cores, no value for "zero" there. Doesn't work though, E-cores are working. MSI is aware of it, already patched as per the thread I linked earlier.

 

[JoeRambo]

The only thing that worked for me was this specific item:

 

[coercitiv]

The only thing that worked for me was this specific item:

No dice, I only have the Enabled / Disabled options for that field.

 

[igor_kavinski]

Also, power. the Xeon is 22nm and 120 watt TDP. How do you think in reality it will compare to the 12700k at stock, or even if I turn the power down a shade ?

RapydMark64 - Google Drive

There is a dual socket 24 thread Xeon there and it's slower than 12600K. Maybe benchmark your Xeon and see if it is able to beat the 12600K?

Also, you will need to set PL1=PL2=125W for the 12700K since it can go up to 190W.

 

[AdamK47]

Micro Center finally received a good supply of higher speed DDR5. Picked up 32GB (2 x 16GB) G.Skill Trident Z5 DDR5-5600 CL36. Currently running them at 6000 CL38.

 

[Markfw]

Micro Center finally received a good supply of higher speed DDR5. Picked up 32GB (2 x 16GB) G.Skill Trident Z5 DDR5-5600 CL36. Currently running them at 6000 CL38.

For how much ? scalping even in our FS/FT has been horrible. $800 for 5200 2 x 16 gig ??

 

[AdamK47]

For how much ? scalping even in our FS/FT has been horrible. $800 for 5200 2 x 16 gig ??

$500

 

[AdamK47]

6200 CL36 @ 1.35V seems doable.

 

[OlyAR15]

Just finishing up my build:
12700KF
Gigabyte Z690i Ultra DDR4 mITX board (the only ITX board with DDR4 available right now)
XPG Gammix S70 2TB SSD
I'm reusing an MSI 3090 and 32GB Gskill Ripjaws 3200 from my sim rig.
Cooling with custom water loop: EK LGA1700 and GPU waterblocks, two 240mm Alphacool XT45 rads, D5 pump, and Swiftech microres. Two Phanteks T30 fans to cool the rads. I also built a custom case, which was a real pain: my skills just aren't up to my vision.

So far things seem stable, so far just a mild overclock to 5.1 single core to 4.9 all cores, mild bump to E-cores. Still dialing things in.

One thing I noticed is that I'm getting a ton of WHEA 17 errors unless I set the PCIe to Gen3. Anyone else getting this? I cloned my old drive to this one, so using the old Window 11 install. Had no problems running on the new hardware, just had to install some drivers and everything seems to work fine, but eventually I'll probably do a fresh install.

 

[JoeRambo]

One thing I noticed is that I'm getting a ton of WHEA 17 errors unless I set the PCIe to Gen3.

Try turning ASPM in BIOS off for that particular PCIE root complex and returning back to GEN4.

 

[OlyAR15]

Try turning ASPM in BIOS off for that particular PCIE root complex and returning back to GEN4.

ASPM is disabled by default in the bios. Didn’t stop the errors. May have to wait for an updated bios.

 

[JoeRambo]

ASPM is disabled by default in the bios. Didn’t stop the errors. May have to wait for an updated bios.

Yeah, i found the hard way why it is disabled

 

[JoeRambo]

@OlyAR15 btw can you do the following test:

1) Run CB23 with 8 threads in default config
2) Run CB23 with 8 threads with E cores disabled
3) Run CB23 with 8 threads with E cores and HT disabled?

I am getting nasty penalty of ~10% with HT enabled in quite a few tests and wondering what is going on

 

[coercitiv]

1) Run CB23 with 8 threads in default config
2) Run CB23 with 8 threads with E cores disabled
3) Run CB23 with 8 threads with E cores and HT disabled?

I am getting nasty penalty of ~10% with HT enabled in quite a few tests and wondering what is going on

Win 10, 46x all core multi, PL1 = PL2 = 150W

8/16 + 0
12882

8/8 + 0
13859

8/8 + 4
14216

I'm too tired to repeat the tests and look at this properly, but it does look like something we should look closely into.

 

[JoeRambo]

Yeah... Except the penalty is even more heavy with CB23 for me. (on WIN11) Getting like 15k with 8/8 and low 13+k with 8/16. Seems to get scheduled without regard to the HT when the marketing cores are disabled.

 

[Hitman928]

Yeah... Except the penalty is even more heavy with CB23 for me. (on WIN11) Getting like 15k with 8/8 and low 13+k with 8/16. Seems to get scheduled without regard to the HT when the marketing cores are disabled.

Yikes. Seems like a pretty major bug in how Windows is handling HT with ADL. What do you get comparing 8c16t P to 8c8tP when E cores are enabled for both?

 

[coercitiv]

Yikes. Seems like a pretty major bug in how Windows is handling HT with ADL. What do you get comparing 8c16t P to 8c8tP when E cores are enabled for both?

On my Win10 rig I'm seeing similar scores between 8/8+4 and 8/16+4 when running 8 threads. What I did notice however was a slight drop of ~150points when manually excluding CB23 affinity from the E-cores. This shouldn't happen with CB set to use 8 threads only, but I repeated the process again and again, the score drop is there and there's a clear spike of E core usage when starting the test.

So I guess I'll need to redo the tests in a different setup, maybe run 6 threads instead of 8 on all configs to have free resources every time and watch what happens when I also pin threads manually.

 

[AdamK47]

I've been running at an all PCore 5GHz and all ECore 4GHz. The Uncore @ 4GHz and the memory at 6000 36-36-36-76.

I decided to try out a 125MHz base clock since all of these frequencies are evenly divisible by 125. I set the PCore multiplier to 40, ECore to 32, Ring to 32, and memory to 48.

Didn't think I'd see much of an increase at all, but....

Edit: Apparently this is a "bug" with AIDA64 with bus speeds above 100MHz.

It puts the legitimacy of any AIDA64 memory benchmark result into doubt.

 

[coercitiv]

The Uncore @ 4GHz and the memory at 6000 36-36-36-76.

I've been thinking about this as well, though it seems a weird trade-off. On one side the system will run better when there's any kind of load on the E-cores, but on the other side we lose the 4.6Ghz stock uncore speed for when the E-cores are idle.

The ideal situation would be to increase the "base" uncore speed and still have dynamic shift to 4.6Ghz+ based on core usage... but that's doesn't seem to be an option, not on my end anyway.

 

[AdamK47]

I've been thinking about this as well, though it seems a weird trade-off. On one side the system will run better when there's any kind of load on the E-cores, but on the other side we lose the 4.6Ghz stock uncore speed for when the E-cores are idle.

The ideal situation would be to increase the "base" uncore speed and still have dynamic shift to 4.6Ghz+ based on core usage... but that's doesn't seem to be an option, not on my end anyway.

Locking uncore to 4.0GHz gives me better performance versus letting it bounce between 3.6GHz and 4.7GHz. When gaming I'm going to have the PCores utilized while the ECores handle background processes. Uncore stays at 3.6GHz when I actually do something. It's akin to single core boost clocks. I can see it certain benchmarks, but that's about it. I'd rather lock uncore to 4.0GHz and all PCores to 5.0GHz.

 

[Justinbaileyman]

Hey guys, so I am really on the fence here about buying a 12900k. I am currently running a AMD 3950x on x570 Aurous Master Motherboard and need to know if there is any word as to if there will be one more CPU release for this Am4 socket or not? If not how much of a upgrade would a 12900k be? I have the 12900k in my cart right now so should I go ahead and pull the trigger or wait??

 

[Markfw]

Hey guys, so I am really on the fence here about buying a 12900k. I am currently running a AMD 3950x on x570 Aurous Master Motherboard and need to know if there is any word as to if there will be one more CPU release for this Am4 socket or not? If not how much of a upgrade would a 12900k be? I have the 12900k in my cart right now so should I go ahead and pull the trigger or wait??

side grade if anything. Unless all you do is game. In which case you want a 12700k.