As mentioned in another thread I’ve been considering playing with the bios settings on my new rig to tweak the performance, this morning, with a cry of “wish me luck, I’m going in there” I screwed up my courage and did the deed.

Venturing into the overclocking section of the bios I decided that as I’d paid for 3200 rated memory having it run at 2400 was a waste so I took the memory speed off of auto and set it to 3200. A quick test showed that it ran ok so I decided to go ahead with the main change.

The core voltage on auto was showing as 1.448 volts whilst doing nothing in the bios. I would have liked to monitor it under load but it’s the one time Ubuntu has let me down - I cannot find a utility outside of the bios that will report the real time value. A quick bit of research indicated that -0.1 volts was a safe change I set the values from auto to offset, negative, 0.1 and rebooted.

So far the system has been stable, is running about 5 degrees cooler and the WUs don’t appear to be taking noticeably longer. Over the past few days I set up a spreadsheet of credits / hour (credits / decoy for Rosetta) by application that I’ll monitor over the next few days to get a better measure of any loss in performance. If that is minimal I might go further but this is an encouraging start.

As mentioned in another thread I’ve been considering playing with the bios settings on my new rig to tweak the performance, this morning, with a cry of “wish me luck, I’m going in there” I screwed up my courage and did the deed.

Venturing into the overclocking section of the bios I decided that as I’d paid for 3200 rated memory having it run at 2400 was a waste so I took the memory speed off of auto and set it to 3200. A quick test showed that it ran ok so I decided to go ahead with the main change.

The core voltage on auto was showing as 1.448 volts whilst doing nothing in the bios. I would have liked to monitor it under load but it’s the one time Ubuntu has let me down - I cannot find a utility outside of the bios that will report the real time value. A quick bit of research indicated that -0.1 volts was a safe change I set the values from auto to offset, negative, 0.1 and rebooted.

So far the system has been stable, is running about 5 degrees cooler and the WUs don’t appear to be taking noticeably longer. Over the past few days I set up a spreadsheet of credits / hour (credits / decoy for Rosetta) by application that I’ll monitor over the next few days to get a better measure of any loss in performance. If that is minimal I might go further but this is an encouraging start.

Sounds good!!!

No noticeable difference on the crude measure of credits / hour so I might try going further in a few days if it continues this way.

As mentioned in another thread I’ve been considering playing with the bios settings on my new rig to tweak the performance, this morning, with a cry of “wish me luck, I’m going in there” I screwed up my courage and did the deed.

Venturing into the overclocking section of the bios I decided that as I’d paid for 3200 rated memory having it run at 2400 was a waste so I took the memory speed off of auto and set it to 3200. A quick test showed that it ran ok so I decided to go ahead with the main change.

The core voltage on auto was showing as 1.448 volts whilst doing nothing in the bios. I would have liked to monitor it under load but it’s the one time Ubuntu has let me down - I cannot find a utility outside of the bios that will report the real time value. A quick bit of research indicated that -0.1 volts was a safe change I set the values from auto to offset, negative, 0.1 and rebooted.

So far the system has been stable, is running about 5 degrees cooler and the WUs don’t appear to be taking noticeably longer. Over the past few days I set up a spreadsheet of credits / hour (credits / decoy for Rosetta) by application that I’ll monitor over the next few days to get a better measure of any loss in performance. If that is minimal I might go further but this is an encouraging start.

This is for your Ryzen 5 3600?
I just did a quick cursory search and saw an example of that processor 3.6GHz base clock successfully overclocked to 3.9GHz <but not under any load> with a VCore of 1.225V.
I'd suggest you have even more headroom than you think.

Rather than titling this thread as "underclocking", think of it more as undervolting.
I've got zero experience with Ryzens, but one approach I've seen is to set the voltage manually at, say, 1.30V or whatever gives you a temperature your current cooling capability provides that you're comfortable with, and see if you run successfully.
If it's ok, step up the overclock repeatedly until it fails, then step back one level. The temperature won't go up with a fixed/manual voltage.

There's no need to do this quickly. Give it a day or two at each level so you know it'll cover all eventualities with whatever you do with your PC (I'm currently giving it a whole week)

---

As mentioned in another thread I’ve been considering playing with the bios settings on my new rig to tweak the performance, this morning, with a cry of “wish me luck, I’m going in there” I screwed up my courage and did the deed.

Venturing into the overclocking section of the bios I decided that as I’d paid for 3200 rated memory having it run at 2400 was a waste so I took the memory speed off of auto and set it to 3200. A quick test showed that it ran ok so I decided to go ahead with the main change.

The core voltage on auto was showing as 1.448 volts whilst doing nothing in the bios. I would have liked to monitor it under load but it’s the one time Ubuntu has let me down - I cannot find a utility outside of the bios that will report the real time value. A quick bit of research indicated that -0.1 volts was a safe change I set the values from auto to offset, negative, 0.1 and rebooted.

So far the system has been stable, is running about 5 degrees cooler and the WUs don’t appear to be taking noticeably longer. Over the past few days I set up a spreadsheet of credits / hour (credits / decoy for Rosetta) by application that I’ll monitor over the next few days to get a better measure of any loss in performance. If that is minimal I might go further but this is an encouraging start.

This is for your Ryzen 5 3600?
I just did a quick cursory search and saw an example of that processor 3.6GHz base clock successfully overclocked to 3.9GHz <but not under any load> with a VCore of 1.225V.
I'd suggest you have even more headroom than you think.

Rather than titling this thread as "underclocking", think of it more as undervolting.
I've got zero experience with Ryzens, but one approach I've seen is to set the voltage manually at, say, 1.30V or whatever gives you a temperature your current cooling capability provides that you're comfortable with, and see if you run successfully.
If it's ok, step up the overclock repeatedly until it fails, then step back one level. The temperature won't go up with a fixed/manual voltage.

There's no need to do this quickly. Give it a day or two at each level so you know it'll cover all eventualities with whatever you do with your PC (I'm currently giving it a whole week)

Yes, it's the R5-3600. Rather than setting a fixed voltage I'm going down the line of offsetting the automatic value and my next step will probably be moving from -0.1v to -0.15v which would be the equivalent to your 1.30v setting.

I'm happy with the temperature drop so far but a bit more would be nice. My main driver is to reduce the power consumption but I'm not sure how to monitor that directly (in the same way as I don't know how to monitor the vCore voltage or the core frequencies under load).

I'm new to all this and a bit concerned what happens when I do go a step too far. I'm hoping that with dropping the power I'll see the performance drop off before the machine becomes unstable and stops running.

Hmm, this is very interesting! I might have to do this with my Ryzens.
Keep in mind, though, depending on the workunits, credit and decoy count may be far different. It's hard to gauge performance on here unless workunits are the same across all machines. Which is unfortunate.

Hmm, this is very interesting! I might have to do this with my Ryzens.
Keep in mind, though, depending on the workunits, credit and decoy count may be far different. It's hard to gauge performance on here unless workunits are the same across all machines. Which is unfortunate.

I’m splitting them by application and comparing them before and after against itself and against my R5-2600 so I’ve figures for Rosetta, WCG (OPI, MIP and MCM) and tn-grid (fma, avx and sse) so it should be clear where the WUs have changed.

I’m seeing a 35-50% improvement between the R5-2600 and the R5-3600 and the R5-2600 was consistent across the before and after whilst the R5-3600 showed, at most, a 1% drop which I’ll live with for the power drop suggested by the 5 degree drop in temperature..

I would have liked to monitor it under load but it’s the one time Ubuntu has let me down - I cannot find a utility outside of the bios that will report the real time value

Look into the zenpower driver and zenmonitor monitor for monitoring Ryzens and Threadrippers under Linux. Works very well for me on both my 3950X and 2920X.

https://github.com/ocerman/zenpower/
https://github.com/HonsW/zenmonitor

Rather than titling this thread as "underclocking", think of it more as undervolting.
I've got zero experience with Ryzens, but one approach I've seen is to set the voltage manually at, say, 1.30V or whatever gives you a temperature your current cooling capability provides that you're comfortable with, and see if you run successfully.
If it's ok, step up the overclock repeatedly until it fails, then step back one level. The temperature won't go up with a fixed/manual voltage.

There's no need to do this quickly. Give it a day or two at each level so you know it'll cover all eventualities with whatever you do with your PC (I'm currently giving it a whole week)

Yes, it's the R5-3600. Rather than setting a fixed voltage I'm going down the line of offsetting the automatic value and my next step will probably be moving from -0.1v to -0.15v which would be the equivalent to your 1.30v setting.

I'm happy with the temperature drop so far but a bit more would be nice. My main driver is to reduce the power consumption but I'm not sure how to monitor that directly (in the same way as I don't know how to monitor the vCore voltage or the core frequencies under load).

I'm new to all this and a bit concerned what happens when I do go a step too far. I'm hoping that with dropping the power I'll see the performance drop off before the machine becomes unstable and stops running.

On the plus side I don't think you can do anything wrong by undervolting (as opposed to overvolting) that a reboot and a tweak in the BIOS can't solve.
I'm glad there's a suggestion for a monitor utility above as, as well as knowing nothing about Ryzens I know nothing about Ubuntu either.

An article that may be of some use to you is here AMD Ryzen 3000 Undervolting Offset vs. Override | Vcore Voltage - the video in it talks through the article.

The only thing I'd mention is the offsets are particularly sensitive. There's a massive difference between 0.10 and 0.15.
Looking at a couple of videos, voltage changes can be stepped through at 0.00625V at a time, so there are 8 steps between 0.10V and 0.15V offsets
By which I mean, if you try 0.15V and it's not successful in some way, you can step back to, say, 0.14, 0.13 etc until you find you're at a point you're happy with
Similarly, if 0.15V is successful, but you think you're not far away from the max you can get away with, step up to 0.16, 0.17
(Obvs these should be 0.14375, 0.13750, 0.13075 and 0.15625, 0.16250 0.16925 etc but that might be too pernickety - it's up to you)

Edit: Apparently it's quite common for the Auto voltage to come up massively and inappropriately high. You're definitely doing the right thing to try and mitigate it

---

I would have liked to monitor it under load but it’s the one time Ubuntu has let me down - I cannot find a utility outside of the bios that will report the real time value

Look into the zenpower driver and zenmonitor monitor for monitoring Ryzens and Threadrippers under Linux. Works very well for me on both my 3950X and 2920X.

https://github.com/ocerman/zenpower/
https://github.com/HonsW/zenmonitor

Many thanks Sir, that looks to be exactly what I need.

Rather than titling this thread as "underclocking", think of it more as undervolting.
I've got zero experience with Ryzens, but one approach I've seen is to set the voltage manually at, say, 1.30V or whatever gives you a temperature your current cooling capability provides that you're comfortable with, and see if you run successfully.
If it's ok, step up the overclock repeatedly until it fails, then step back one level. The temperature won't go up with a fixed/manual voltage.

There's no need to do this quickly. Give it a day or two at each level so you know it'll cover all eventualities with whatever you do with your PC (I'm currently giving it a whole week)

Yes, it's the R5-3600. Rather than setting a fixed voltage I'm going down the line of offsetting the automatic value and my next step will probably be moving from -0.1v to -0.15v which would be the equivalent to your 1.30v setting.

I'm happy with the temperature drop so far but a bit more would be nice. My main driver is to reduce the power consumption but I'm not sure how to monitor that directly (in the same way as I don't know how to monitor the vCore voltage or the core frequencies under load).

I'm new to all this and a bit concerned what happens when I do go a step too far. I'm hoping that with dropping the power I'll see the performance drop off before the machine becomes unstable and stops running.

On the plus side I don't think you can do anything wrong by undervolting (as opposed to overvolting) that a reboot and a tweak in the BIOS can't solve.
I'm glad there's a suggestion for a monitor utility above as, as well as knowing nothing about Ryzens I know nothing about Ubuntu either.

An article that may be of some use to you is here AMD Ryzen 3000 Undervolting Offset vs. Override | Vcore Voltage - the video in it talks through the article.

The only thing I'd mention is the offsets are particularly sensitive. There's a massive difference between 0.10 and 0.15.
Looking at a couple of videos, voltage changes can be stepped through at 0.00625V at a time, so there are 8 steps between 0.10V and 0.15V offsets
By which I mean, if you try 0.15V and it's not successful in some way, you can step back to, say, 0.14, 0.13 etc until you find you're at a point you're happy with
Similarly, if 0.15V is successful, but you think you're not far away from the max you can get away with, step up to 0.16, 0.17
(Obvs these should be 0.14375, 0.13750, 0.13075 and 0.15625, 0.16250 0.16925 etc but that might be too pernickety - it's up to you)

Edit: Apparently it's quite common for the Auto voltage to come up massively and inappropriately high. You're definitely doing the right thing to try and mitigate it

Thanks for that, a very interesting article - a shame that they chose such a ridiculously low fixed voltage for their comparison, I’d have thought maybe 1.2v would have been more appropriate.

I would have liked to monitor it under load but it’s the one time Ubuntu has let me down - I cannot find a utility outside of the bios that will report the real time value

Look into the zenpower driver and zenmonitor monitor for monitoring Ryzens and Threadrippers under Linux. Works very well for me on both my 3950X and 2920X.

https://github.com/ocerman/zenpower/
https://github.com/HonsW/zenmonitor

Many thanks Sir, that looks to be exactly what I need.

Interesting, it’s reporting a vCore of between 1.219v and 1.263v, mostly to the lower end of the scale, which is far more reasonable than the 1.445v (-0.1v) it was reporting in the bios.

The effective CPU frequency is fairly steady at 3,900 with a max of 3,925 so no damage to performance so far.

If I'm not mistaken, the BIOS shows the "ask" or VID voltage. Not what the Vcore really is under load. I like the zenpower driver because it uses the direct telemetery (SVI2) off the VRM's that supply the Vcore and Vsoc voltages like the Windows utilities like HWinfo. You can talk apples to apples when comparing systems with Windows users.

If I'm not mistaken, the BIOS shows the "ask" or VID voltage. Not what the Vcore really is under load. I like the zenpower driver because it uses the direct telemetery (SVI2) off the VRM's that supply the Vcore and Vsoc voltages like the Windows utilities like HWinfo. You can talk apples to apples when comparing systems with Windows users.

I wish I’d had it before I started to give me a more complete picture - every Ryzen system should have it :-)

If I'm not mistaken, the BIOS shows the "ask" or VID voltage. Not what the Vcore really is under load. I like the zenpower driver because it uses the direct telemetery (SVI2) off the VRM's that supply the Vcore and Vsoc voltages like the Windows utilities like HWinfo. You can talk apples to apples when comparing systems with Windows users.

I wish I’d had it before I started to give me a more complete picture - every Ryzen system should have it :-)

Only if they are interested in those kinds of things, tweaking takes skill and time and sure alot of crunching people can do that but not everyone wants too.
Do I think AMD could make a deal and add it to their software for a price...sure but I'm not sure over 25% of the people would find it helpful and at some point there is a breakover price point making it not cost effective.

If I'm not mistaken, the BIOS shows the "ask" or VID voltage. Not what the Vcore really is under load. I like the zenpower driver because it uses the direct telemetery (SVI2) off the VRM's that supply the Vcore and Vsoc voltages like the Windows utilities like HWinfo. You can talk apples to apples when comparing systems with Windows users.

I wish I’d had it before I started to give me a more complete picture - every Ryzen system should have it :-)

Only if they are interested in those kinds of things, tweaking takes skill and time and sure alot of crunching people can do that but not everyone wants too.
Do I think AMD could make a deal and add it to their software for a price...sure but I'm not sure over 25% of the people would find it helpful and at some point there is a breakover price point making it not cost effective.

But it's free :-)

If I'm not mistaken, the BIOS shows the "ask" or VID voltage. Not what the Vcore really is under load. I like the zenpower driver because it uses the direct telemetery (SVI2) off the VRM's that supply the Vcore and Vsoc voltages like the Windows utilities like HWinfo. You can talk apples to apples when comparing systems with Windows users.

I wish I’d had it before I started to give me a more complete picture - every Ryzen system should have it :-)

Eventually it will, the standard k10temp driver for AMD and Ryzen processors gets updates in the HWMON branch at every new kernel release. The zenpower and RAPL coding has been brought into the k10temp branch for voltage and power readouts.

So in later kernels, you won't have to do anything extra or use an out of tree driver to see those sensors. You will be able to see them just by invoking sensors in a terminal. You still have to pick your poison in regard to a desktop GUI monitor if you choose to use one.

If I'm not mistaken, the BIOS shows the "ask" or VID voltage. Not what the Vcore really is under load. I like the zenpower driver because it uses the direct telemetery (SVI2) off the VRM's that supply the Vcore and Vsoc voltages like the Windows utilities like HWinfo. You can talk apples to apples when comparing systems with Windows users.

I wish I’d had it before I started to give me a more complete picture - every Ryzen system should have it :-)

Eventually it will, the standard k10temp driver for AMD and Ryzen processors gets updates in the HWMON branch at every new kernel release. The zenpower and RAPL coding has been brought into the k10temp branch for voltage and power readouts.

So in later kernels, you won't have to do anything extra or use an out of tree driver to see those sensors. You will be able to see them just by invoking sensors in a terminal. You still have to pick your poison in regard to a desktop GUI monitor if you choose to use one.

Wonderful. My monitor of choice is PSensor but I'm not precious about it.

If I'm not mistaken, the BIOS shows the "ask" or VID voltage. Not what the Vcore really is under load. I like the zenpower driver because it uses the direct telemetery (SVI2) off the VRM's that supply the Vcore and Vsoc voltages like the Windows utilities like HWinfo. You can talk apples to apples when comparing systems with Windows users.

I wish I’d had it before I started to give me a more complete picture - every Ryzen system should have it :-)

Eventually it will, the standard k10temp driver for AMD and Ryzen processors gets updates in the HWMON branch at every new kernel release. The zenpower and RAPL coding has been brought into the k10temp branch for voltage and power readouts.

So in later kernels, you won't have to do anything extra or use an out of tree driver to see those sensors. You will be able to see them just by invoking sensors in a terminal. You still have to pick your poison in regard to a desktop GUI monitor if you choose to use one.

And more is coming for the Zen2 Ryzen's https://www.techspot.com/news/86498-upcoming-free-software-brings-more-performance-lower-power.html

I saw the announcement of the new utility also. No help for us Linux users. We still have to find the individual CCX stable voltage points through experimentation. Good for Windows users. All automatic.