Undervolting Ryzen 3XXX CPU

[holyprof]

My Ryzen PC is annoying me since I bought it last summer. Yes, the performance is great, everything is faster and smoother BUT ... as much as I edit fan profile curves, add more and bigger case fans, the stock CPU cooler makes too much noise even at idle with CPU load below 0.5%. My old Intel rig was so silent (in idle) I could hear the GPU water pump noise (and it's quite silent). Idle temp on the Ryzen CPU was touching 55-57ºC in the first months. Later AMD released the new AGESA and idle temps fell a bit, to 48-53ºC (my old i5-4690K is happily idling at 28ºC, in the same room).

So I did a google search about undervolting and found this article

https://www.gamersnexus.net/guides/3494-amd-ryzen-3000-undervolting-offset-override

Following the advice from it, instead of hard undervolting I configured the offset instead, putting -0.05V offset applied to CPU Vid.
I didn't try benchmarking but the PC is just as smooth and fast as before, but now my Ryzen is at 40-48ºC with a browser with 5 tabs open and CPU load of 2%.
Maybe if I finally replace the stock cooler with something better, results will be even more satisfying, but I think that better cooler would make bigger difference at high loads, not at idle. Also the current Coronavirus situation raised the prices of everything and stocks are almost non-existent so new cooler is out of the question for the foreseeable future.

Now my question is: Is this a placebo effect, am I deluding myself, or is such gain possible with such a small offset in VID?

 

[NoJuan999]

I have a 3700x with a Scythe Ninja 5 cooler and my CPU idles at 30-32C (ambient temp is ~20-21C).
I leave my CPU voltage on Auto and have PBO enabled.

So yes, a good cooler can lower idle temps.

 

[northvisit]

I have cooler CPU (3600) and I get about 3-4 ºC decrease in temperature in stress test in full load (about 75=>71-72 ºC ) with NH-U12A using -0,0875 offset voltage. With stock cooler effect could be bigger. I would assume (not checked) idle temps to drop less, but VCore is generally bigger in low load situation so it could be as well other way. Have you validated that negative offset value 0.05V works without performance drop with some test run like CB20. Do it without any background tasks as those cause easily variation on results. If CB results drop any meaningway way then there is CPU automatic effective downclocking happening to about 3,6kHz occasionally. Increase the negative value until you start to see performance drop and then go back to lower it until no drop any more. I started to see slight perf drop with bigger that -0,100 V values so I use -0,0875V offset to have a bit margin. This limit varies on each CPU chip.

 

[sanchom]

I have a Ryzen 9 3950x and I set 1 volt and a frequency of 4040 to 16 cores (32). I ran all the tests, at full load 100 watts. In idle time, only 2 cores operate at a frequency of 500, the minimum consumption.

 

[moproblems99]

My Ryzen PC is annoying me since I bought it last summer. Yes, the performance is great, everything is faster and smoother BUT ... as much as I edit fan profile curves, add more and bigger case fans, the stock CPU cooler makes too much noise even at idle with CPU load below 0.5%. My old Intel rig was so silent (in idle) I could hear the GPU water pump noise (and it's quite silent). Idle temp on the Ryzen CPU was touching 55-57ºC in the first months. Later AMD released the new AGESA and idle temps fell a bit, to 48-53ºC (my old i5-4690K is happily idling at 28ºC, in the same room).

So I did a google search about undervolting and found this article

https://www.gamersnexus.net/guides/3494-amd-ryzen-3000-undervolting-offset-override

Following the advice from it, instead of hard undervolting I configured the offset instead, putting -0.05V offset applied to CPU Vid.
I didn't try benchmarking but the PC is just as smooth and fast as before, but now my Ryzen is at 40-48ºC with a browser with 5 tabs open and CPU load of 2%.
Maybe if I finally replace the stock cooler with something better, results will be even more satisfying, but I think that better cooler would make bigger difference at high loads, not at idle. Also the current Coronavirus situation raised the prices of everything and stocks are almost non-existent so new cooler is out of the question for the foreseeable future.

Now my question is: Is this a placebo effect, am I deluding myself, or is such gain possible with such a small offset in VID?

You likely lost performance with it as with both of my chips immediately downclock if using offsets.

 

[agentnathan009]

Unless you are strapped for cash, I'd spring for one of the new Arctic Cooling AIO coolers (they are not too expensive and they should be much quieter, especial for that higher powered CPU). I have the first gen 120 AIO and it keeps my 2600X around 35-40 at idle, and about 60-65 under full load.

 

[holyprof]

I have cooler CPU (3600) and I get about 3-4 ºC decrease in temperature in stress test in full load (about 75=>71-72 ºC ) with NH-U12A using -0,0875 offset voltage. With stock cooler effect could be bigger. I would assume (not checked) idle temps to drop less, but VCore is generally bigger in low load situation so it could be as well other way. Have you validated that negative offset value 0.05V works without performance drop with some test run like CB20. Do it without any background tasks as those cause easily variation on results. If CB results drop any meaningway way then there is CPU automatic effective downclocking happening to about 3,6kHz occasionally. Increase the negative value until you start to see performance drop and then go back to lower it until no drop any more. I started to see slight perf drop with bigger that -0,100 V values so I use -0,0875V offset to have a bit margin. This limit varies on each CPU chip.

In the article I linked, they say it depends more on the motherboard than the CPU itself. They found the acceptable negative offset to be between 0.05 and 0.1V (using the same chip). All of that with performance loss of less than 2% (and even 2% gain in some scenarios, due to lower temps so the CPU boosts higher).

I have a 3700x with a Scythe Ninja 5 cooler and my CPU idles at 30-32C (ambient temp is ~20-21C).
I leave my CPU voltage on Auto and have PBO enabled.

So yes, a good cooler can lower idle temps.

From the more than 10 Ryzen-compatible cooler reviews, the ninja was what attracted me the most. Better than most of the competition for approximately half the price. Also, since I record sound and work using videoconferencing now, noise from the PC is even more annoying - the damn condenser microphone catches all.
Problem is, I can't buy it anywhere in my country (the official Scythe importer went bankrupt or closed doors I don't know) and I don't feel comfortable ordering from abroad and risking of getting a half-smashed package with heatsink blades torn and broken fan.
I turned logging on, and for about a minute, the CPU idled at 39ºC, something I've never seen before! It's been always in the high forties (48 and above). I'm happy for now.

I have cooler CPU (3600) and I get about 3-4 ºC decrease in temperature in stress test in full load (about 75=>71-72 ºC ) with NH-U12A using -0,0875 offset voltage. With stock cooler effect could be bigger. I would assume (not checked) idle temps to drop less, but VCore is generally bigger in low load situation so it could be as well other way. Have you validated that negative offset value 0.05V works without performance drop with some test run like CB20. Do it without any background tasks as those cause easily variation on results. If CB results drop any meaningway way then there is CPU automatic effective downclocking happening to about 3,6kHz occasionally. Increase the negative value until you start to see performance drop and then go back to lower it until no drop any more. I started to see slight perf drop with bigger that -0,100 V values so I use -0,0875V offset to have a bit margin. This limit varies on each CPU chip.

Will test CB20 soon and let you guys know the result. Will try offset of AUTO, -0.05, -0.075 and -0.1 (if the CPU boot at all and no bluescreen).

 

[agentnathan009]

In the article I linked, they say it depends more on the motherboard than the CPU itself. They found the acceptable negative offset to be between 0.05 and 0.1V (using the same chip). All of that with performance loss of less than 2% (and even 2% gain in some scenarios, due to lower temps so the CPU boosts higher).


From the more than 10 Ryzen-compatible cooler reviews, the ninja was what attracted me the most. Better than most of the competition for approximately half the price. Also, since I record sound and work using videoconferencing now, noise from the PC is even more annoying - the damn condenser microphone catches all.
Problem is, I can't buy it anywhere in my country (the official Scythe importer went bankrupt or closed doors I don't know) and I don't feel comfortable ordering from abroad and risking of getting a half-smashed package with heatsink blades torn and broken fan.
I turned logging on, and for about a minute, the CPU idled at 39ºC, something I've never seen before! It's been always in the high forties (48 and above). I'm happy for now.


Will test CB20 soon and let you guys know the result. Will try offset of AUTO, -0.05, -0.075 and -0.1 (if the CPU boot at all and no bluescreen).

After reading this thread I decided to do a negative CPU voltage offset on my 2600X. I ended up at -0.1 and found that it runs cooler and boosts higher. Previously at stock settings it would only boost all cores to 3.9 Ghz give or take. With the offset it boosts to 4.16 GHz on all cores and sometimes boosts up to 4.1 on most cores somewhat frequently.

 

[Cranky5150]

Getting a better cooler is your best bet my friend, not only for future expansion but, for lower temps and a more stable running rig. There is a old saying that nowadays is kinda true but : "there is no replacement for displacement" LOL

 

[holyprof]

Ended up buying the only cooler I could get my paws on in my local shop - dirt cheap Hyper 212 Black Edition. Idle temps are down to 30-32 ºC (and ambient went up a bit so it's even better). 5 cinebench 20 multicore runs in a row, and max CPU temp is 66 ºC. Only thing i can hear now is the GPU water pump, no way around it. The rest of the system is so quiet i need to look at fans to see them spinning because I can't hear them. Vcore offset = -0.05V.
Thanks for all the advice.

 

[moproblems99]

Ended up buying the only cooler I could get my paws on in my local shop - dirt cheap Hyper 212 Black Edition. Idle temps are down to 30-32 ºC (and ambient went up a bit so it's even better). 5 cinebench 20 multicore runs in a row, and max CPU temp is 66 ºC. Only thing i can hear now is the GPU water pump, no way around it. The rest of the system is so quiet i need to look at fans to see them spinning because I can't hear them. Vcore offset = -0.05V.
Thanks for all the advice.

Did you lose any clocks with the offset?

 

[holyprof]

Did you lose any clocks with the offset?

There is a small (within margin of error) drop in CB20 multi-core test. Single core went up though, I thought it would be just the opposite.
Strangely, with PBO turned on, single-core CB20 went down, but the multi-core went up.
I can't track boost frequency well, but peak power draw is as follows:

• 120W with PBO (CPU temp 73ºC during CB20 MT test)
• 90W with everything on auto (CPU temp 64ºC)
• 90W with -0.05V offset, but average power draw is lower, maybe around 80W (during CB20 MT test). CPU temp hangs around 62ºC.

All of that while keeping scores within margin of error, but the system is so quiet now that I need to look at it to tell it's on. For someone that does several hours of videoconferencing every day, records video and audio, this was a godsend. I wanted to make the PC quieter while keeping performance. New cheap cooler + offset = win!

I should try bigger offset to amplify the effect so it becomes noticeable. Too lazy though, maybe I'll try it later today.

 

[agentnathan009]

There is a small (within margin of error) drop in CB20 multi-core test. Single core went up though, I thought it would be just the opposite.
Strangely, with PBO turned on, single-core CB20 went down, but the multi-core went up.
I can't track boost frequency well, but peak power draw is as follows:

• 120W with PBO (CPU temp 73ºC during CB20 MT test)
• 90W with everything on auto (CPU temp 64ºC)
• 90W with -0.05V offset, but average power draw is lower, maybe around 80W (during CB20 MT test). CPU temp hangs around 62ºC.

All of that while keeping scores within margin of error, but the system is so quiet now that I need to look at it to tell it's on. For someone that does several hours of videoconferencing every day, records video and audio, this was a godsend. I wanted to make the PC quieter while keeping performance. New cheap cooler + offset = win!

I should try bigger offset to amplify the effect so it becomes noticeable. Too lazy though, maybe I'll try it later today.

When I turned PBO on I ended up with higher temps, more noise and lower clocks. When offset voltage 0.1 temperature went down, clocks went up and noise went down. I attempted 0.15 offset and I think I encountered a stability issue. Haven’t further tested to verify.

 

[northvisit]

When I turned PBO on I ended up with higher temps, more noise and lower clocks. When offset voltage 0.1 temperature went down, clocks went up and noise went down. I attempted 0.15 offset and I think I encountered a stability issue. Haven’t further tested to verify.

Remember to check the performance as well with negative offset values. If zen 2 chips detect too low voltage it will clock down (clock stretching) "automatic". It is not visible in monitored clock speeds, but can be seen in lower performance. So in reality real clock might be lower than what you see. Some chips accept over -0.1 V while some chip maybe nothing.

 

[holyprof]

With PBO, the sustained all-core clock went from 3.9-4.2 to 4.0-4.325 when running the CB20 MT test. Without it, clocks are quite a bit lower, but scores aren't that much lower. All in all, overclocking is not worth with today's CPUs that dynamically auto-overclock themselves.

 

[ltcdata]

Will test CB20 soon and let you guys know the result. Will try offset of AUTO, -0.05, -0.075 and -0.1 (if the CPU boot at all and no bluescreen).

Hi! I'm in the same boat as you.
I did exactly that as a test.
My config is:
Windows 10 pro x64 1909, 18363.815.
Ryzen 3700x, wraith cooler (high speed enabled)
Asus x570 tuf gaming wifi, Bios version 1407, AGESA 1.0.0.4
16gb Corsair Vengeance LPX (8x2) memory 3000mhz 16-20-20-38 1,35v, XMP profile,
(i couldn't get better memory here in Argentina)
Latest AMD drivers installed (2.04.04.111), AMD Ryzen BALANCED power plan selected.

And these are my test results:

cinebench score - pbo off - vcore in bios- cores frequency as reported by RyzenMaster -, vcore avg reported by RM cpuz bench - temp by ryzenmaster during CB20 load

cinebench20 4590 pbo off vcore without offset cores 3,85ghz vcore 1,1 cpuz 5340/512 temp 71°C
cinebench20 4631 pbo off vcore offset -0,05 cores 3,87ghz vcore 1,22 cpuz 5389/509 temp 67°C
cinebench20 4665 pbo off vcore offset -0,08 cores 3,92ghz vcore 1,25 cpuz 5444/510 temp 67°C
cinebench20 4286 pbo off vcore offset -0,1 cores 3,98ghz vcore 1,30 cpuz 5353/509 temp 67°C

Every cinebench run was left in a loop 10 minutes, to get stable values.

It seems that using an offset of -0.08 volt in vcore gives me better performance.

So... this is not a placebo effect. In fact, my CB20 scores went higher with some negative offset, in special with -0,08volts, and that score is stable (i let it run for 1 hour continuously).

What gives?

 

[holyprof]

My own scores, temp reads from HWInfo64, fans on silent preset (700 rpm idle, around 1000 rpm at full load) except last run.
Cooler: Hyper 212 Black Edition.

CPU voltage	fan profile (case + CPU)	CB20 multi	CB20 single	Peak CPU temp.	Peak CPU power
Auto	silent	4802	511	64 ºC	90 W
Offset -0.05V	silent	4813	506	64 ºC	90 W
PBO	performance (vaccumcleaner-like noise)	4982	508	73 ºC	118 W

Except for benchmarking, all results look same to me. I won't trade dead-silent PC and 30W of power saving for a 4-5% less rendering time in the few apps that can push the CPU to 100% on the 16 threads.

 

[ltcdata]

All my results are with custom fan profile:
Ambient temp aprox 20°C, cpu fan minimum speed 1300rpm, idle temp 35-40°C, fan starts to ramp up at about 60°C (during the run, averages 2100rpm), 2500rpm at 75°C, max speed above that, and i have 3 80mm intake fan, 2 fixed at 500rpm, and another one governed by cpu temp, starts at 600rpm when cpu =>60°C and can go up to 2000rpm at >75°C, so, intake fans are very silent.
There is one 80mm exhaust fan, fixed at 500rpm also.
So, all my scores are with a very silent PC, you can hear a little whoosh when the cpu fan spins at 2100rpm, but i have my case with sound-dampening material so you can barely hear it.
If i set all fans to turbo, the scores do not go higher, they remain the same. I will try to do another run in silent mode and report, to see if there is any thermal constraint.

From the looks of your score, your CB20 is almost the same in auto and with offset, and in my case CB20 scores go up a little. Different silicon i guess.

I do a lot of RAW photo processing and 4k video transcoding so multicore performance at 100% load in all cores is important to me and every little gain is appreciated.

 

[agentnathan009]

Hi! I'm in the same boat as you.
I did exactly that as a test.
My config is:
Windows 10 pro x64 1909, 18363.815.
Ryzen 3700x, wraith cooler (high speed enabled)
Asus x570 tuf gaming wifi, Bios version 1407, AGESA 1.0.0.4
16gb Corsair Vengeance LPX (8x2) memory 3000mhz 16-20-20-38 1,35v, XMP profile,
(i couldn't get better memory here in Argentina)
Latest AMD drivers installed (2.04.04.111), AMD Ryzen BALANCED power plan selected.

And these are my test results:

cinebench score - pbo off - vcore in bios- cores frequency as reported by RyzenMaster -, vcore avg reported by RM cpuz bench - temp by ryzenmaster during CB20 load

cinebench20 4590 pbo off vcore without offset cores 3,85ghz vcore 1,1 cpuz 5340/512 temp 71°C
cinebench20 4631 pbo off vcore offset -0,05 cores 3,87ghz vcore 1,22 cpuz 5389/509 temp 67°C
cinebench20 4665 pbo off vcore offset -0,08 cores 3,92ghz vcore 1,25 cpuz 5444/510 temp 67°C
cinebench20 4286 pbo off vcore offset -0,1 cores 3,98ghz vcore 1,30 cpuz 5353/509 temp 67°C

Every cinebench run was left in a loop 10 minutes, to get stable values.

It seems that using an offset of -0.08 volt in vcore gives me better performance.

So... this is not a placebo effect. In fact, my CB20 scores went higher with some negative offset, in special with -0,08volts, and that score is stable (i let it run for 1 hour continuously).

What gives?

Hi! I'm in the same boat as you.
I did exactly that as a test.
My config is:
Windows 10 pro x64 1909, 18363.815.
Ryzen 3700x, wraith cooler (high speed enabled)
Asus x570 tuf gaming wifi, Bios version 1407, AGESA 1.0.0.4
16gb Corsair Vengeance LPX (8x2) memory 3000mhz 16-20-20-38 1,35v, XMP profile,
(i couldn't get better memory here in Argentina)
Latest AMD drivers installed (2.04.04.111), AMD Ryzen BALANCED power plan selected.

And these are my test results:

cinebench score - pbo off - vcore in bios- cores frequency as reported by RyzenMaster -, vcore avg reported by RM cpuz bench - temp by ryzenmaster during CB20 load

cinebench20 4590 pbo off vcore without offset cores 3,85ghz vcore 1,1 cpuz 5340/512 temp 71°C
cinebench20 4631 pbo off vcore offset -0,05 cores 3,87ghz vcore 1,22 cpuz 5389/509 temp 67°C
cinebench20 4665 pbo off vcore offset -0,08 cores 3,92ghz vcore 1,25 cpuz 5444/510 temp 67°C
cinebench20 4286 pbo off vcore offset -0,1 cores 3,98ghz vcore 1,30 cpuz 5353/509 temp 67°C

Every cinebench run was left in a loop 10 minutes, to get stable values.

It seems that using an offset of -0.08 volt in vcore gives me better performance.

So... this is not a placebo effect. In fact, my CB20 scores went higher with some negative offset, in special with -0,08volts, and that score is stable (i let it run for 1 hour continuously).

What gives?

lower voltage leads to lower temps which CPU uses that extra thermal headroom to raise clock frequencies. Auto overclocking entails the CPU/motherboard having to manage voltage, current (amps), thermal limits, etc to make on the fly overclocking. If you CPU is running cooler, it can raise clock frequency, or lower if CPU is exceeding thermal limitations.

However, offsetting too much can incur a performance loss. You might see higher frequency spikes (higher frequencies showing up in HWinfo) but lower effective clock speeds as the clock speeds are adjusted in milliseconds all of the time by CPU for auto overclocking.

offsetting voltage can lower temperature a little and fan noise and may incur a small performance penalty depending on how much voltage offset you apply.

 

[holyprof]

From ltcdata's post and from what i've read in articles, -0.075 V seems to be the sweet spot for negative offset undervolting for most of the Ryzen 3700X CPUs. I only tested with -0.05V to be on the safe side. Maybe tomorrow I'll test with -0.075 V (and gain 1-2% additional CB20 points).

 

[ltcdata]

Yes.
With PBO enabled and a very conservative profile (110PPT 65TDC 90EDC) i can hit 4820 in CB20, but at the expense of higher noise and temperatures (average 77°C after 10 runs).
For me, the sweet spot of noise, temperature and longevity of the cpu is PBO OFF and offseting to -0,08. It can boost in single-dual core loads to 4,3ghz without problems, and even if all core it remanis at about 3,86ghz, my CB20 scores are almost 100 points higher than stock, so better for me, and with an average temp of 65°C.

 

[holyprof]

Update: tried voltage offset of -0.075V but the PC became unstable. Dialled back to -0.0625V, will try to run some CB2 benches tomorrow. And compare to -0.050V results. But I guess it won't gain me anything.

 

[tabascosauz]

Update: tried voltage offset of -0.075V but the PC became unstable. Dialled back to -0.0625V, will try to run some CB2 benches tomorrow. And compare to -0.050V results. But I guess it won't gain me anything.

Every CPU's different, it's hard to draw conclusions amongst 3700X units given how wild the binning is for this process. In order to sustain a -0.075V undervolt (maximum stable undervolt without losing performance compared to default), I have to run Turbo LLC to minimize droop and get the full load Vcore right where I want it, on the edge of stable territory. It really doesn't make much of a difference.

If you're not after keeping thermals controlled in a confined space like I am, and instead want to undervolt to increase performance, you would probably be better off running stock settings, enabling PBO, and tweaking the PBO advanced parameters to try and pull off the "EDC trick". Keeping the PPT and TDC parameters stock at 88 and 60 respectively, then putting a slight reduction on EDC can provide a fair bit of a single-thread and all-core boost clock increase. Stock EDC is 90A, mine works best at 83A, with the +200MHz override and a 2X scalar. But others have settled on a variety of EDC numbers between 80A and 90A, it just depends on what your chip wants.

• Undervolted with Turbo LLC: 4890-4950pts, with all-core at 40.3-40.8x.
• Stock: 4900pts, with all-core at 40.3-40.5x.
• Balls-to-the-wall PBO: 4900-4950pts, with all-core at 40.8x. But a hell of a lot of heat and noise.
• Stock + EDC trick: 4950-4980pts, with all-core at 41-41.5x and single-core boost up to 4.43GHz. Same heat/noise as stock except slightly more under full load.

 

[holyprof]

Every CPU's different, it's hard to draw conclusions amongst 3700X units given how wild the binning is for this process. In order to sustain a -0.075V undervolt (maximum stable undervolt without losing performance compared to default), I have to run Turbo LLC to minimize droop and get the full load Vcore right where I want it, on the edge of stable territory. It really doesn't make much of a difference.

If you're not after keeping thermals controlled in a confined space like I am, and instead want to undervolt to increase performance, you would probably be better off running stock settings, enabling PBO, and tweaking the PBO advanced parameters to try and pull off the "EDC trick". Keeping the PPT and TDC parameters stock at 88 and 60 respectively, then putting a slight reduction on EDC can provide a fair bit of a single-thread and all-core boost clock increase. Stock EDC is 90A, mine works best at 83A, with the +200MHz override and a 2X scalar. But others have settled on a variety of EDC numbers between 80A and 90A, it just depends on what your chip wants.

• Undervolted with Turbo LLC: 4890-4950pts, with all-core at 40.3-40.8x.
• Stock: 4900pts, with all-core at 40.3-40.5x.
• Balls-to-the-wall PBO: 4900-4950pts, with all-core at 40.8x. But a hell of a lot of heat and noise.
• Stock + EDC trick: 4950-4980pts, with all-core at 41-41.5x and single-core boost up to 4.43GHz. Same heat/noise as stock except slightly more under full load.

Thanks for the advice, I applied undervolting mostly to reduce noise in idle and during light loads (while keeping performance - if it goes up it's a free bonus).
When encoding videos, I can live with the noise but I can also live with time being 1hour and 10 minutes instead of 1 hour and 9 minutes. So I'll dial back offset to -0.05V and keep PBO off and all the rest on auto.

 

[Max(IT)]

without open a new thread I will try this one.
Having recently jumped ship from Intel to a new AMD gaming rig, I'm still confused about some behavior of Matisse CPUs.
I found that, at least during hot summer temperature, the best solution for performance is to slightly undervolt my CPU. A setting between -0.0750 and -0.0625 V are providing better performance (in both Cinebench R20 ST/MT and CPU Z bench) while keeping temperature under control. If I go farther than that, single core performance are impacted (not so much but a little bit definitely yes), I if apply less undervolt MT perfomance are impacted (most probably for thermal limits, since the CPU is reaching 80/83°).
Now I have a question regarding LLC, something I was able to use on Intel platform but I'm not sure about AMD, so I'm asking for advice.
On Intel platform while undervolting I was rising LLC to the maximum (or a notch less) to provide more Voltage under load. Now I have an Asus ROG Strix B550-F Gaming board, with a decent VRM section, and 5 levels of LLC (default AUTO).
Do you advise to set it to 4 or 5 while undervolting ? Or just keep it as it is (AUTO) ?