[SOLVED] How to undo AI Suite III CPU Overclock ?

[mtm_kikio]

Hello,

I used AI Suite III 5-Way Optimization to overclock Ryzen 5 5600X (Default Clock Speed: 3.7GHz, after Optimization 4.6GHz).
I want to undo everything and went to BIOS to reset all settings using 'F5' and 'F10'.

BIOS showed everything back to default (Please refer to Pic 1).
But When login into Windows 10, it back again to 4.6GHz (Please refer to Pic 2).
AI Suite setting also back to default, but the clock speed still at 4.6GHz ? (Please refer to Pic 3)

I've tried to uninstall AI Suite 3 software but it didn't help,
How can I change everything back to default?

CPU: AMD Ryzen 5 5600X
Motherboard: ROG STRIX B550-I GAMING

Pic 1:

Pic 2:

Pic 3:

 

[Colif]

uninstall AI suite? and reset bios again.

thats what I did.

run https://docs.microsoft.com/en-us/sysinternals/downloads/autoruns and use it to stop ASio.sys & ASio2.sys from running at startup

 

[mtm_kikio]

Thank you for your reply Colif.
I've tried your suggestions, but this time it switching between 3.7Ghz to 4.6Ghz, something still forcing my CPU to overclock, I think some AI config still active.
Do you have another suggestion?

 

[Colif]

i guess its not cam 0verclocking? only mention it as i see the choice in 2nd picture above

it has to be something loading at startup with windows.
Try a clean boot and see if it changes anything - make sure to read instructions and make sure NOT to disable any Microsoft services or windows won't load right - https://support.microsoft.com/en-au/help/929135/how-to-perform-a-clean-boot-in-windows

if clean boot fixes it, it shows its likely a startup program. You should, over a number of startups. restart the programs you stopped to isolate the one that is to blame.

its likely another Asus program. Unless you use them, I would just uninstall all Asus programs, most things that come with motherboards besides drivers can be ignored..

 

[MonsterMaxx]

RMB on the menu and launch Task Manager. Hit the Startup tab. See if there's anything in there that you don't want starting. Reboot.

If that's not it, it's in your bios.

 

[johnsoner13]

4.6 ghz is default boost clock, no?

 

[mtm_kikio]

@johnsoner13, the base clock speed of 3.7 GHz and a max boost clock speed of 4.6 GHz for Ryzen 5 5600x.

@MonsterMaxx, what is RMB?

@Colif, I've tried your suggestion to clean boot, but the clock speed still stuck at 4.6GHz. I also tried to Re-install Windows 10 and Cleared CMOS but still the same.
One thing I noticed when changing TPU in BIOS from "Keep Current Settings" to "TPU 1" or "TPU 2" the clock speed will change to 4.2GHz(for TPU 1) and 4.3GHz(for TPU 2).

 

[Colif]

rmb = right mouse button. He basically told you to do what i did, stop anything at startup that might cause it. It won't help.
try running hwinfo and tell us what your average effective clock is - I made this for an unrelated reason but it might help - https://forums.tomshardware.com/threads/how-to-use-hwinfo-to-track-sensor-values-on-ryzen.3693704/
its possible NZXT monitor is just as dumb as task manager is at actually knowing how fast you are going at any one moment
Win 10 thinks my 3600XT is always over 4ghz when often its not even close
both of these can't be right

Only HWINFO & AMD Ryzen Master are normally even close to accurate.

 

[mtm_kikio]

@Colif Is this normal? Clocks speed on CPUID Hardware Monitor also keep switching.

Is it okay to leave it like this? or better to lock it at 3.7GHz(using Ryzen Master or Bios)?

 

[Colif]

that is normal, Ryzen CPU are constantly waking cores to do an action and then sleeping them again.

it shows NZXT is looking at the wrong methods to track actual usage. HWINFO changed from just watching the clock x multiplier to actually tracking usage
Average Effective Clock in HWINFO is closer to actual average speed.

Your CPU isn't oc anymore, its back to normal.

 

[mtm_kikio]

Colif
Thank you.

 

[Karadjgne]

Look at the last pic at the very bottom. CLOCKS. Core #0 4.6GHz. Core #1+ 3.5GHz. Apply that to pic #2 and pic #3. Those readings were only of the fastest clock, not the entire cpu. While you are actively doing something like moving the mouse around etc, it's got 1 core boosted to max turbo to deal with the load. The rest are idling at below base speeds.

That's how Ryzens work. They sit low until needed, then boost as high as power limits and temps allow. A single core is well within power limits and the cpu isn't thermal throttling, so that core can boost to max. Pushing 3 cores, you might see all 3 at max, if the load is low enough to keep temps below 60°, but get to 62-64° and you'll see 4.7/4.7/4.6. A Ryzen will start lowering clocks in individual cores, which lowers temps, so other cores can maintain better performance, which is better overall than those 3 cores all being 4.7, but temps climbing to 70°+ (Intel anyone?)

What you can do with Ryzens is dial in ram timings, subtimings, fclock for maximum use per clock. Fps/test results go up and speeds don't change, temp doesn't change, efficiency of each clock changes. It's why you'll see cinebench scores higher than yours, but clocks are lower. They are pushing higher temps, which lowers clocks but the effectiveness of the ram is higher. With better cooling or a little voltage tweaking, they could raise clocks a little and get even higher scores.

For instance my 3700x stock got 3723 in CB R20. At 80+°. Tweaked the ram and set a -0.2v offset and jumped to 4893. At 62°C. At 4.2GHz all cores boosting on auto. Set a 4.4GHz manual OC and got 5101 at 80+°C. So got rid of all that, went back to the offset and auto settings and much cooler temps gaming. And getting almost identical fps to the locked core OC.

And no PBO or Ryzen 200MHz auto OC.

I believe Asus and all the others are so acclimated to Intel OC and it's setup that it's not really sunk in yet that Ryzens left alone to work like Ryzens get much better overall performance than kicking everything as high as possible.

 

[Colif]

Clocks is the old way of measuring speed, it only looks at baseclock x multiplier, not actual usage. its how CPU-Z and NZXT & Task Manager rank how fast CPU is actually going, not taking into account the fact Ryzen uses race to idle approach

getting them to agree would be nice

hwinfo and Ryzen master are closest but they will never be exactly same unless AMD shares their methodology.

 

[mtm_kikio]

Look at the last pic at the very bottom. CLOCKS. Core #0 4.6GHz. Core #1+ 3.5GHz. Apply that to pic #2 and pic #3. Those readings were only of the fastest clock, not the entire cpu. While you are actively doing something like moving the mouse around etc, it's got 1 core boosted to max turbo to deal with the load. The rest are idling at below base speeds.

That's how Ryzens work. They sit low until needed, then boost as high as power limits and temps allow. A single core is well within power limits and the cpu isn't thermal throttling, so that core can boost to max. Pushing 3 cores, you might see all 3 at max, if the load is low enough to keep temps below 60°, but get to 62-64° and you'll see 4.7/4.7/4.6. A Ryzen will start lowering clocks in individual cores, which lowers temps, so other cores can maintain better performance, which is better overall than those 3 cores all being 4.7, but temps climbing to 70°+ (Intel anyone?)

What you can do with Ryzens is dial in ram timings, subtimings, fclock for maximum use per clock. Fps/test results go up and speeds don't change, temp doesn't change, efficiency of each clock changes. It's why you'll see cinebench scores higher than yours, but clocks are lower. They are pushing higher temps, which lowers clocks but the effectiveness of the ram is higher. With better cooling or a little voltage tweaking, they could raise clocks a little and get even higher scores.

For instance my 3700x stock got 3723 in CB R20. At 80+°. Tweaked the ram and set a -0.2v offset and jumped to 4893. At 62°C. At 4.2GHz all cores boosting on auto. Set a 4.4GHz manual OC and got 5101 at 80+°C. So got rid of all that, went back to the offset and auto settings and much cooler temps gaming. And getting almost identical fps to the locked core OC.

And no PBO or Ryzen 200MHz auto OC.

I believe Asus and all the others are so acclimated to Intel OC and it's setup that it's not really sunk in yet that Ryzens left alone to work like Ryzens get much better overall performance than kicking everything as high as possible.

Thank you for explaining in detail but I'm sorry I'm new to overclocking stuff, I do not fully understand your explanation.
Could you kindly explain more about:

1. How do you dial ram timing, subtimings, and fclock? What software do you use?

2. "Tweaked the ram and set a -0.2v offset" How do you tweak it? and what is offset?

3. "went back to the offset and auto settings " are you saying you are using default setting here?

Thank you.

 

[Karadjgne]

1. No software. Your ram has a listing in bios that shows its timings, that's the 16-18-18-39 2T (primary) numbers, followed by 30+ (secondary and Tertiary) other numbers. If you look up videos on Ryzen ram tweaking, there's a few of those that can be changed lower.

Each of those numbers represents a thing. Think of it as the amount of time it takes data to enter a door, cross a room, exit another door. Thats the primary. Secondary and Tertiary would be how much time it takes to turn the first door handle, time waited before actually opening the door, time to move you through the doorway, speed of closing the door, how hard the door is closed etc.

So all that needs to be in harmony, smooth and swift, or you run into the door because you didn't wait long enough, you slam your foot in the door because you closed it too fast.

Tweaking ram is the same. The faster that you can get the data into the ram, through the ram and on its way to the cpu when demanded, the faster your cpu can operate per clock. It's not waiting for the bus.

Ryzen cpus are not a single die (that's the block itself) under the lid like Intel. Because of the way cores radiate heat when working, pack them in tight and they heat up everything next to them. Ryzen splits up into 3-6 dies, that allows the heat better transfer through the lid, without affecting its neighbors as much. Runs cooler. But each of those dies has a freeway connecting them to allow for data to move. That's governed by Infinity Fabric. IF works best at a 1:1:1 ratio with fclock:mclock:uclock. It's speeds are set by the Data Rate of your ram.

Ram is Dual Data Rate, DDR, so with 3600MHz ram, the Data Rate is 1800MHz. So you'd want IF to be 1800:1800:1800. Auto settings usually apply that, but not always, there's variables and chances fclock will change, like the way vcore voltages lower and dip as clock speeds change. So manually setting fclock locks it at exactly half of your DDR. Trying to set it faster or slower messes with the ratio and makes it unstable.

2. Voltage use creates heat. Higher the voltage of the cpu, more heat it makes. If you look in HWInfo, you'll see that vcore per core changes with use, a single core might demand 1.5v, but realistically only needs 1.3v. All cores working boosts to a lower limit but all cores at 1.5v will end up damaging the cpu, too much instant heat. So I changed vcore (auto) to (manual) for the 4.4GHz set OC, just as has always been done with OC's, but didn't like the heat result because that sets all cores to 4.4GHz permanently, even if running CSGO which only uses 2 cores, the other 6 doing nothing, no work. That's a ton of wasted voltage, effort and excess heat.

So by further change to (offset) (negative) (0.2v) and leaving boost at (auto), I get a single core boost at 1.3v instead of 1.5v, and if all cores are boosted, they still max out at 1.3v instead of 1.5v. So CSGO use sees 2 cores at 4.4GHz 1.3v, and 6 cores at 3.5GHz resting at 0.9v. Much less heat, maximum boost. Same affect as the hard OC. Maximum fps.

The specific size of the negative offset can/does change. It all depends on what your individual cpu actually needs, vs what it demands. Could be 0.2v or 0.3v (that's max setting) or more likely 0.15v etc.

3. Yes. PBO works very well with low end cpus. It really doesn't do anything useful for cpus that are already seeing use close to power limits. So PBO I disabled, and no 'Auto OC' which adds upto 200MHz to the boost clocks. That's additional stuff added that isn't necessary. A 3700x boosts to 4.4GHz, no point trying to force it to boost to 4.4GHz when it already does.

But most things other than the ram timings and vcore I left at factory defaults. Ryzens factory overclock themselves, that's boost. All I've done is tweak the process so that boost is more efficient and overall that means more affective. Makes the cpu work faster and stronger without working any harder, in affect the voltage changes make it actually work less hard than default.

Ryzens take a very Human approach to things. You understand the expression 'Work Smarter, not Harder', that's their philosophy. All a Ryzen OC should be is streamlining the boost, making the cpu work smarter. Set OC limits just makes the cpu work harder. And at the end of the day, harder = hotter.

 

[mtm_kikio]

1. No software. Your ram has a listing in bios that shows its timings, that's the 16-18-18-39 2T (primary) numbers, followed by 30+ (secondary and Tertiary) other numbers. If you look up videos on Ryzen ram tweaking, there's a few of those that can be changed lower.

Each of those numbers represents a thing. Think of it as the amount of time it takes data to enter a door, cross a room, exit another door. Thats the primary. Secondary and Tertiary would be how much time it takes to turn the first door handle, time waited before actually opening the door, time to move you through the doorway, speed of closing the door, how hard the door is closed etc.

So all that needs to be in harmony, smooth and swift, or you run into the door because you didn't wait long enough, you slam your foot in the door because you closed it too fast.

Tweaking ram is the same. The faster that you can get the data into the ram, through the ram and on its way to the cpu when demanded, the faster your cpu can operate per clock. It's not waiting for the bus.

Ryzen cpus are not a single die (that's the block itself) under the lid like Intel. Because of the way cores radiate heat when working, pack them in tight and they heat up everything next to them. Ryzen splits up into 3-6 dies, that allows the heat better transfer through the lid, without affecting its neighbors as much. Runs cooler. But each of those dies has a freeway connecting them to allow for data to move. That's governed by Infinity Fabric. IF works best at a 1:1:1 ratio with fclock:mclock:uclock. It's speeds are set by the Data Rate of your ram.

Ram is Dual Data Rate, DDR, so with 3600MHz ram, the Data Rate is 1800MHz. So you'd want IF to be 1800:1800:1800. Auto settings usually apply that, but not always, there's variables and chances fclock will change, like the way vcore voltages lower and dip as clock speeds change. So manually setting fclock locks it at exactly half of your DDR. Trying to set it faster or slower messes with the ratio and makes it unstable.

2. Voltage use creates heat. Higher the voltage of the cpu, more heat it makes. If you look in HWInfo, you'll see that vcore per core changes with use, a single core might demand 1.5v, but realistically only needs 1.3v. All cores working boosts to a lower limit but all cores at 1.5v will end up damaging the cpu, too much instant heat. So I changed vcore (auto) to (manual) for the 4.4GHz set OC, just as has always been done with OC's, but didn't like the heat result because that sets all cores to 4.4GHz permanently, even if running CSGO which only uses 2 cores, the other 6 doing nothing, no work. That's a ton of wasted voltage, effort and excess heat.

So by further change to (offset) (negative) (0.2v) and leaving boost at (auto), I get a single core boost at 1.3v instead of 1.5v, and if all cores are boosted, they still max out at 1.3v instead of 1.5v. So CSGO use sees 2 cores at 4.4GHz 1.3v, and 6 cores at 3.5GHz resting at 0.9v. Much less heat, maximum boost. Same affect as the hard OC. Maximum fps.

The specific size of the negative offset can/does change. It all depends on what your individual cpu actually needs, vs what it demands. Could be 0.2v or 0.3v (that's max setting) or more likely 0.15v etc.

3. Yes. PBO works very well with low end cpus. It really doesn't do anything useful for cpus that are already seeing use close to power limits. So PBO I disabled, and no 'Auto OC' which adds upto 200MHz to the boost clocks. That's additional stuff added that isn't necessary. A 3700x boosts to 4.4GHz, no point trying to force it to boost to 4.4GHz when it already does.

But most things other than the ram timings and vcore I left at factory defaults. Ryzens factory overclock themselves, that's boost. All I've done is tweak the process so that boost is more efficient and overall that means more affective. Makes the cpu work faster and stronger without working any harder, in affect the voltage changes make it actually work less hard than default.

Ryzens take a very Human approach to things. You understand the expression 'Work Smarter, not Harder', that's their philosophy. All a Ryzen OC should be is streamlining the boost, making the cpu work smarter. Set OC limits just makes the cpu work harder. And at the end of the day, harder = hotter.

Karadjgne thank you very much for explaining in detail.
I'll try to read bit more about these, understanding it, and then try implement so I can get the most out of my CPU.

 

[Karadjgne]

Welcome to the New World. Trust me when I say that having to 'un-think' a decades worth of OC practice was a pain. You have the advantage there, it's all new 🤣

 

[mtm_kikio]

Welcome to the New World. Trust me when I say that having to 'un-think' a decades worth of OC practice was a pain. You have the advantage there, it's all new 🤣

Btw could you share screenshot of your settings?

 

[Karadjgne]

Yes and no. My bios will look different, my ram is different.

Pretty much everything is factory default optimised settings except for docp = xmp setting, and using a offset mode negative (-) and the ram subtimings.

It's as basic a change for very good performance as I've ever seen. Normally there'd be 20 things to tinker with and tweak but amd has these cpus already pushing limits so it's just a matter streamlining according to what's there.

Auto actually works realistically for a change.