[igor_kavinski]

Ay my current config going from all cores to 6p/8e took me from 49.2S to 67.3S. Curious to see what you get.

P53 E49 RAM at 8200 MT/s and XMP timings in Win11:

Stock core clocks with XMP RAM timings in Win Server 2019:

It's like Win Server 2019 is optimized for max multicore throughput.

 

[igor_kavinski]

My final best score on Win Server 2019 with P53 E49 Ring 44 NGU 33 D2D 34 8200C38-48-48-85:

 

[someperson12345]

Once I get it in the case I will play around with further undervolting but I have enough info to say I am incredibly happy the gaming performance seems more than fine. I didn't do multiple runs it isn't great methodology it's to prove installing this in the case is worth it. Hopefully it helps or entertains someone else.

my ring doesn't have much headroom and neither does my RAM so I'm currently at:

D2D 33x @ .78V
NGU 32x @ .935V
PL1 reduced to 162W (from 250)
CPU core voltage offset -75mv
latency low 70's

This keeps it from throttling on air even in Y cruncher which hits almost 200W the entire run even at PL1 of 162W.

Some comparisons using hwinfo for CPU power average

Cinebench 2024
179w average /1093 pts (12700K) vs 160w / 2087 pts (265K)

Y cruncher 2.5B
164w average / 114.4s (12700K) vs 197w / 48.5s (265k )

I think those are pretty impressive. If you look at the overall power used to do the same calcs (wattage multiplied by time) in the Y cruncher run 12700K is 18.76KWs (killawatt seconds) of power used while 265K is 9.55KWs. Performance per watt in non gaming so far at least seems good. But gaming sucks right? That was my concern.

Cyberpunk 2077 testing was next, three tests with a 3080

1920x1080 low preset
12700K 121 min, 197 max, 158 average (102W)
265K 197 min, 284 max, 235 average (108W)

1920x1080 ultra preset
12700K 89 min, 165 max, 126 average (102W)
265k 118 min, 179 max, 143 average (94W)

2540x1440 ultra preset
12700K 76 min, 111 max, 88 average (89W)
265k 81 min, 120 max, 94 average (76W)

At 2560x1440 ultra it's ~7% faster than my old rig with almost 15% less power. Not bad.

At low FHD we're talking 48.7% faster at 5.8% more power. W00t w00t.

Time to install this. I've been seeing CPU usage across all cores as high as 77% with Satisfactory and performance is starting to suck that will be my next test.

 

[igor_kavinski]

By the way, thanks @someperson12345 for posting your y-cruncher score. I had never tried that benchmark before and it let me detect an unstable RAM OC when all the other benchmarks passed. Now it's gonna be my first benchmark any time I tinker with RAM.

 

[someperson12345]

By the way, thanks @someperson12345 for posting your y-cruncher score. I had never tried that benchmark before and it let me detect an unstable RAM OC when all the other benchmarks passed. Now it's gonna be my first benchmark any time I tinker with RAM.

No problem, it's amazing software. It found RAM and ring OC errors on the stress test that passed others for me also.

 

[jamesch]

Once I get it in the case I will play around with further undervolting but I have enough info to say I am incredibly happy the gaming performance seems more than fine. I didn't do multiple runs it isn't great methodology it's to prove installing this in the case is worth it. Hopefully it helps or entertains someone else.

my ring doesn't have much headroom and neither does my RAM so I'm currently at:

D2D 33x @ .78V
NGU 32x @ .935V
PL1 reduced to 162W (from 250)
CPU core voltage offset -75mv
latency low 70's

This keeps it from throttling on air even in Y cruncher which hits almost 200W the entire run even at PL1 of 162W.

Some comparisons using hwinfo for CPU power average

Cinebench 2024
179w average /1093 pts (12700K) vs 160w / 2087 pts (265K)

Y cruncher 2.5B
164w average / 114.4s (12700K) vs 197w / 48.5s (265k )

I think those are pretty impressive. If you look at the overall power used to do the same calcs (wattage multiplied by time) in the Y cruncher run 12700K is 18.76KWs (killawatt seconds) of power used while 265K is 9.55KWs. Performance per watt in non gaming so far at least seems good. But gaming sucks right? That was my concern.

Cyberpunk 2077 testing was next, three tests with a 3080

1920x1080 low preset
12700K 121 min, 197 max, 158 average (102W)
265K 197 min, 284 max, 235 average (108W)

1920x1080 ultra preset
12700K 89 min, 165 max, 126 average (102W)
265k 118 min, 179 max, 143 average (94W)

2540x1440 ultra preset
12700K 76 min, 111 max, 88 average (89W)
265k 81 min, 120 max, 94 average (76W)

View attachment 2694074


View attachment 2694075


View attachment 2694076


At 2560x1440 ultra it's ~7% faster than my old rig with almost 15% less power. Not bad.

At low FHD we're talking 48.7% faster at 5.8% more power. W00t w00t.

Time to install this. I've been seeing CPU usage across all cores as high as 77% with Satisfactory and performance is starting to suck that will be my next test.

Do you have Baldur's Gate 3? I find that game is extremely sensitive to CPU core undervolts. Generally you crash super fast on that game with aggressive undervolts, whereas Cyberpunk is pretty lenient since it's more of an all-core load.

Your undervolt seems pretty aggressive for Arrow Lake, it seems like all the one's I've seen prior to yours nobody tries to do more than -30mv, but that's with the 285K.

 

[someperson12345]

Do you have Baldur's Gate 3? I find that game is extremely sensitive to CPU core undervolts. Generally you crash super fast on that game with aggressive undervolts, whereas Cyberpunk is pretty lenient since it's more of an all-core load.

Your undervolt seems pretty aggressive for Arrow Lake, it seems like all the one's I've seen prior to yours nobody tries to do more than -30mv, but that's with the 285K.

I do looks like that will be my weekend! Thanks for the advice I've been trying to determine how to test the undervolt, I am definitely concerned about stability and how to test for it. I've been thinking a lot about what the best way to test an offset across the entire V/F curve is when you dont know where the worst case is at, so this is very useful info. I've been expecting that I may need to decrease it I dont have much time on that config yet. This will all be installed in my new Fractal Design Meshify this weekend and I'll start the in depth testing next week.

 

[Monsicek]

Once I get it in the case I will play around with further undervolting but I have enough info to say I am incredibly happy the gaming performance seems more than fine. I didn't do multiple runs it isn't great methodology it's to prove installing this in the case is worth it. Hopefully it helps or entertains someone else.

my ring doesn't have much headroom and neither does my RAM so I'm currently at:

D2D 33x @ .78V
NGU 32x @ .935V
PL1 reduced to 162W (from 250)
CPU core voltage offset -75mv
latency low 70's

This keeps it from throttling on air even in Y cruncher which hits almost 200W the entire run even at PL1 of 162W.

Some comparisons using hwinfo for CPU power average

Cinebench 2024
179w average /1093 pts (12700K) vs 160w / 2087 pts (265K)

Y cruncher 2.5B
164w average / 114.4s (12700K) vs 197w / 48.5s (265k )

Heya, thanks for posting these, my 265K is on the way. Is 12700K undervolted as well or running stock voltages. Feel free to include anything relevant as well.

 

[someperson12345]

This seems to be too good to be true. Made it past the Nautilus in BG3 with no problems. At this point I'm wondering if it's doing anything so I turned down the fan speed to increase temps and ran a few more. Then I changed the undervolt back and forth a few times and ran Y cruncher both with -75mv and auto to make sure things looked right. Then I did -100 and Y cruncher and BG3 also ran fine (just briefly). Y cruncher results are about 1S better with undervolt (air cooling that is throttling without UV).

-125 booted Windows but would not run Y cruncher.

This is what HWinfo showed on each run

Did I win the lottery? I had no idea what "normal" on these was until jamesch posted so I didnt realize it was that good!

Any other suggestions on how to test stability?

Now off to replace this cooler with a Phantom Spirit 120 evo and hope it's as good as they say

 

[someperson12345]

Heya, thanks for posting these, my 265K is on the way. Is 12700K undervolted as well or running stock voltages. Feel free to include anything relevant as well.

No problem. My 12700K was not tweaked at all, actually probably a little handicapped. I added two more stick of DDR 4 (4x16) to play Satisfactory and it wouldn't run XMP timings any more.

 

[someperson12345]

Guess I just have to spend some time gaming! -100 undervolt failed pretty quickly but I thought -75 was looking good until I decided to finish Satisfactory. ~2 hours in and crash to desktop. Currently testing -55mv. Off to test BG3 some more.

Good news is Satisfactory performance is great at end game with a large factory. Max CPU went from ~77% to ~53% the bad dips are gone and FPS is more consistent. Wattage is about the same.

 

[someperson12345]

-55mv undervolt is stable I am calling this build done. Tested many hours with BG3 and Satisfactory including lots of idle time in each. I'll update OP with my build details.

 

[igor_kavinski]

PL1 reduced to 162W (from 250)

Does this reduction help with the turbo boost and provide slightly higher ST performance?

 

[someperson12345]

Does this reduction help with the turbo boost and provide slightly higher ST performance?

I didn't notice any difference in Y cruncher when changing this I just did it to reduce temps and since it didn't seem to affect my scores I left it. I can do a full Y cruncher 2.5b run now on air cooling without any throttling.

However when applying the under volt I did gain about one second in Y cruncher, I do believe that undervolting does help with allowing it to boost higher. Runs cooler and slightly faster. For my use case I am very happy. I didn't specifically try a ST workload with and without UV but in Y cruncher it was consistent, about one second improvement from the UV.

 

[igor_kavinski]

What default undervolt protection settings did you have to disable in the BIOS for the undervolt?

 

[someperson12345]

None. Under "CPU DLVR Configuration" I just changed Core Voltage Offset to -55 (Core Input Voltage must be set to offset mode).

 

[igor_kavinski]

New microcode coming soon? https://www.msi.com/Motherboard/MAG-Z890-TOMAHAWK-WIFI/support

 

[someperson12345]

Lol I was wondering is this was actually ever going to arrive! Thanks for the heads up I dont see anything yet for my Asrock board I'll keep checking.

 

[igor_kavinski]

Lol I was wondering is this was actually ever going to arrive! Thanks for the heads up I dont see anything yet for my Asrock board I'll keep checking.

Week 10 is the expected date. So maybe around March 15th.

https://www.overclock.net/posts/29429068/

 

[Xclock]

@igor_kavinski what mobo did you use for that tests. is not visible in your tests....