[Sify]

Hi everyone,

I was following this guide in order to undervolt my 13900k and I would like a sanity check from you all regarding the use of the setting below as a daily driver.

The hardware in question:

• 13900k (SP P-Core 107, SP E-Core 78)
• Asus Z790 Hero, 0904 BIOS, 16.1.25.2124 Intel ME
• G.Skill Trident Z5 RGB DDR5-6000 BK C30
• iCue H170i Elite Capellix in push-pull configuration

Baseline I started with was XMP1 + Asus MCE disabled. That gave me CB23 score of 37921.

The following the guide I arrived at:
LLC#4
AC_LL 0.22
DC_LL 1.02
IA AC Limit 1700
Power limits on Auto (253W)
CB23 score of 39326.
The above was tested by running P95 for about half an hour and CB23 for an half an hour.

In both situations the temperatures are good (under 90C).

Two questions then:

• Are these okay setting for a daily driver?
• If I remove the power limits (set them to 4095) then the above configuration is no longer stable. It will blue screen. To make it stable without power limits I would need to set AC_LL to 0.37 and then I get CB23 score of 38287. Why is it that I need to raise AC_LL with higher power limits?

 

[Betroz]

Why is it that I need to raise AC_LL with higher power limits?

Because with lower power limits, the CPU downclocks under heavy load in order to maintain the set limit (253W). With power limits removed, your CPU tries to run at default allcore of 5500, but doesn't get enough vcore because of too low AC_LL. Your undervolt is not stable for such heavy loads, but it may be for gaming.

 

[Sify]

Because with lower power limits, the CPU downclocks under heavy load in order to maintain the set limit (253W). With power limits removed, your CPU tries to run at default allcore of 5500, but doesn't get enough vcore because of too low AC_LL. Your undervolt is not stable for such heavy loads, but it may be for gaming.

Right. So the worst case scenario here would be a group of cores running at max frequencies but not pulling enough power to get power throttled. What's the best way to test it? Single core loads? Or does testing groups makes sense too?
One thing I know is that testing each of the 2^24 subsets of cores for an hour is not possible.

 

[Betroz]

What's the best way to test it?

There is no ONE way. People use different tools for stability testing. What do you use your PC for? Gaming? With power limits removed, your cooling would be an issue anyways for Prime95 small ffts AVX2 loads and other such heavy loads. Heck even Cinebench R23 (depending on vcore required).

Asus Realbench 2.56 stresstest set to 32GB profile if you have 32GB RAM, and let it run for 2-4 hours. You may need to set a bigger swapfile for it to run. Playing Battlefield 2042 for 2-3 hours is a good stability test aswell. For RAM testing I use TM5 with anta ABSOLUT profile and Y-cruncher VST.

 

[Sify]

There is no ONE way. People use different tools for stability testing. What do you use your PC for? Gaming? With power limits removed, your cooling would be an issue anyways for Prime95 small ffts AVX2 loads and other such heavy loads. Heck even Cinebench R23 (depending on vcore required).

Asus Realbench 2.56 stresstest set to 32GB profile if you have 32GB RAM, and let it run for 2-4 hours. You may need to set a bigger swapfile for it to run. Playing Battlefield 2042 for 2-3 hours is a good stability test aswell. For RAM testing I use TM5 with anta ABSOLUT profile and Y-cruncher VST.

Thanks for the answer.

I am specifically wondering how to best test a scenario where Vcore might be not enough, but its hidden because if I run all cores the power throttling hides the issues.

 

[Betroz]

I am specifically wondering how to best test a scenario where Vcore might be not enough, but its hidden because if I run all cores the power throttling hides the issues.

A powerlimit of 253W is what I use myself for my 13900K cooled by LF 360 AIO. Since you have a push/pull config you probably have a little more headroom, but not much more. With the 253W limit, games and most normal loads will get you 5500 on the p-cores, but more heavy stuff will power throttle yes.

Go into BIOS and check the SP value for your 13900K and post it here. I have a SP91 chip (SP-100 on the pcores). So I can't really undervolt mine either much. One tip though : disable HyperThreading in BIOS. You will loose performance in rendering and stuff, but not in games.

 

[Sify]

An update.

First of all - I did not directly answer your question regarding the use of this machine. It will be used for both gaming (as I like gaming), but also for work. I do graphics programming as a job and compiling code is part of it. I also do rendering, but for me that is happening on the GPU so not directly relevant here.

The SP scores - I posted in the original post - SP P-Core 107, SP E-Core 78).

And finally - I run a whole suite of tests and benchmarks to make sure things are stable and so far they are. Here is a full list:

IBTx16 (Stress lever: Maximum, Xtreme stress mode) - ok
P95 1h - ok
CB23 MC 30min - ok
CB23 SC 30min - ok
OCCT (Large, Extreme, Steady, Auto thread settings, 1h) - ok
OCCT (Large, Extreme, Variable, Auto thread settings, 1h) - ok
OCCT (Large, Extreme, Steady, Advanced thread settings - physical and virtual + core cycling every 15 minute, 24 * 15min = 6h) - ok
Realbench (1h, Up to 64gb) - ok
Geekbench 6 x10 - 1 - ok, 2 - ok, 3 - ok, 4 - ok, 5 - ok, 6 - ok, 7 - ok, 8 - ok, 9 - ok, 10 - ok

3Dmark - Speed Way Stress Test - ok (passed)
3Dmark - Time Spy Stress Test - ok (passed)
3Dmark - Time Extreme Spy Stress Test - ok (passed)
3Dmark - Port Royal Stress Test - ok (passed)
3Dmark - Night Raid Stress Test - ok (passed)
3Dmark - Wild Life Stress Test - ok (passed)
3Dmark - Wild Life Unlimited Stress Test - ok (not passed)
3Dmark - Wild Life Extreme Stress Test - ok (passed)
3Dmark - Fire Strike Stress Test - ok (passed)
3Dmark - Fire Strike Extreme Stress Test - ok (passed)
3Dmark - Fire Strike Ultra Stress Test - ok (passed)
3Dmark - Fire Strike Ultra Stress Test - ok (passed)
3Dmark - Speed Way Benchmark - ok (passed, 5277 < 5332 avg)
3Dmark - Port Royal Benchmark - ok (passed, 12910 < 13395 avg)
3Dmark - Time Spy Extreme Benchmark - ok (passed, 10233 < 10317 avg)
3Dmark - Time Spy Benchmark - ok (passed, 19546 < 19876 avg)
3Dmark - Fire Strike Ultra Benchmark - ok (passed, 12588 < 12758 avg)
3Dmark - Fire Strike Extreme Benchmark - ok (passed, 23554 < 23776 avg)
3Dmark - Fire Strike Benchmark - ok (passed, 38365 < 39178 avg)
3Dmark - Wild Life Extreme Benchmark - ok (passed, 43739 > 32704 avg, yay)
3Dmark - Wild Life Benchmark - ok (passed, 119532 > 111426 avg, yay)
3Dmark - Night Raid Benchmark - ok (passed, 93629 > 91472 avg, yay)
3Dmark - CPU Profile Benchmark - ok (passed)