[SOLVED] Is it ok to undervolt by offset?

[Alvesss154]

My CPU (Ryzen 5 5600g) doesn't boost nicely with stock voltages, (4.168ghz at full load) so I tried undervolting it by offset (-0.036v) and now it can go up to 4.230ghz at full load, I've left everything else on stock and I wanna know if that slight undervolt can damage anything, like my CPU. Ps: before edc was at 95% at full load, now it reaches 100% Mobo a520m s2h

 

[tennis2]

Hello fellow 5600G owner!

Yes, voltage offset is the preferred method. Make sure it's stable and that you aren't inducing clock stretching:

Clock stretching at 11:00

 

[drea.drechsler]

My CPU (Ryzen 5 5600g) doesn't boost nicely with stock voltages, (4.168ghz at full load) so I tried undervolting it by offset (-0.036v) and now it can go up to 4.230ghz at full load, I've left everything else on stock and I wanna know if that slight undervolt can damage anything, like my CPU. Ps: before edc was at 95% at full load, now it reaches 100% Mobo a520m s2h

Undervolting won't damage anything but it might leave the system unstable. And then there's the clock stretching phenomenon so you have to test performance and not just look at clocks to determine if it's working for you.

Can you enter PBO settings in your A520 BIOS? I'm curious because I thought A520's were "overclock locked" and couldn't.

 

[Alvesss154]

Undervolting won't damage anything but it might leave the system unstable. And then there's the clock stretching phenomenon so you have to test performance and not just look at clocks to determine if it's working for you.

Can you enter PBO settings in your A520 BIOS? I'm curious because I thought A520's were "overclock locked" and couldn't.

Surprisingly I can, but some options are missing, such as CPU core ratio. To do the undervolt I put vcore on normal and dynamic voltage on -0.036v.

 

[drea.drechsler]

Surprisingly I can, but some options are missing, such as CPU core ratio. To do the undervolt I put vcore on normal and dynamic voltage on -0.036v.

If you have PBO settings then look for Curve Optimizer. Use that to undervolt by using negative settings. If it is complete you should be able to do it for individual cores.

 

[Alvesss154]

If you have PBO settings then look for Curve Optimizer. Use that to undervolt by using negative settings. If it is complete you should be able to do it for individual cores.

I can see some pbo settings but it's deeply hidden into the bios, unfortunately I can't see curve optimizer.
Are there any problems if I just mantain the vcore setting as it is now (undervolted)?

 

[drea.drechsler]

...
Are there any problems if I just mantain the vcore setting as it is now (undervolted)?

The only serious problem is system instability, meaning the system could crash in certain types of workloads. It's best to test for that with something like Cinebench for 30 min's or so to be sure it stays stable.

Clock stretching alone is a minor problem since it means it's not performing as well as you might think it is. You have to run proper benchmark to determine that...and again, Cinebench is very good for that.

 

[Alvesss154]

Assuming it's not in clock stretching the only serious problem is system instability, meaning the system could crash in certain types of workloads. It's best to test for that with something like Cinebench for 30 min's or so to be sure it stays stable.

Clock stretching alone is a minor problem since it means it's not performing as well as you might think it is. You have to run proper benchmark to determine that...and again, Cinebench is very good for that.

I've just finished the benchmarks here, multi score went from ~10350 to ~10570 points and single core from ~1414 to ~1410 points
Single core score dropped a little bit, but still very close, so if I'm not mistaken that's a technical draw.
I still don't know if clock stretching is happening or not, I tried to give my PC equal conditions on both stock and undervolted benchmarks, what do you think?

 

[drea.drechsler]

I've just finished the benchmarks here, multi score went from ~10350 to ~10570 points and single core from ~1414 to ~1410 points
Single core score dropped a little bit, but still very close, so if I'm not mistaken that's a technical draw.
I still don't know if clock stretching is happening or not, I tried to give my PC equal conditions on both stock and undervolted benchmarks, what do you think?

Even the +2% multi score change is not enough to be noticeable, but it is in the right direction. And the direction the scores moved is pretty much as would be expected with lowering voltage across the board: with less voltage it's less eager to boost single core as high, but can hold multiple cores boosting longer since it heats up the cores less.

Keep something in mind: gaming depends on single core boosting for maximum performance. That's because games have only one or two threads that are heavily loaded and even heavily multi threaded games the other threads don't significantly affect frame render time.

FWIW, for gaming performance spend more time overclocking the iGPU and memory. That will get you much more benefit.

 

[Alvesss154]

Even the multi score change is not enough to be noticeable, but it is in the right direction. And the direction the scores moved is pretty much as would be expected with lowering voltage across the board: with less voltage it's less eager to boost single core as high, but can hold multiple cores boosting longer since it heats up the cores less.

Keep something in mind: gaming depends on single core boosting for maximum performance. That's because games have only one or two threads that are heavily loaded and even heavily multi threaded games the other threads don't significantly affect frame render time.

FWIW, for gaming performance spend more time overclocking the iGPU and memory. That will get you much more benefit.

Thank you for all the replies and sorry for bothering you that much, but I still have one last question.
Initially I was using the stock cooler and it was pretty bad, my clocks didn't go as high and dropped fairly quickly, so I changed to an 120mm aio.
The new cooling solution works as a charm, temps went from 85 to 65 degrees on full load. CPU clock improved but I noticed that they were not as good as they should be, the clocks were on par with similar systems that didn't even have an aio, just the stock cooler.
When benchmarking the "spike" of clocks were similar to what is was before with the stock cooler, and edc and power was still the same (95a and 62w).
Now that it's undervolted clock speeds are higher, EDC maxing out, but power still at 62w.
I don't know if it's my motherboard, because I thought with a better cooling solution my PC would boost higher and I wouldn't need to undervolt.
Do you know why my system is a bit "worse" than most others with similar specs?
Ps: on stock configuration or not, it's kinda hard to hit 4450 mhz on games as I see in most people 5600g

 

[tennis2]

I still don't know if clock stretching is happening or not, I tried to give my PC equal conditions on both stock and undervolted benchmarks, what do you think?

Watch the video I linked. 11:00 to 12:00...just one minute of your time.

I thought with a better cooling solution my PC would boost higher and I wouldn't need to undervolt.

One of these things will limit frequency:

1. Temp
2. Power Draw
3. Frequency Cap (this will vary based on how many cores are loaded, ie mine will do 4.6GHz on a single core, but only 4.1GHz with all cores loaded)
4. EDC/TDC

Most likely you're at the frequency cap.

 

[Alvesss154]

Watch the video I linked. 11:00 to 12:00...just one minute of your time.

One of these things will limit frequency:

1. Temp
2. Power Draw
3. Frequency Cap (this will vary based on how many cores are loaded, ie mine will do 4.5GHz on a single core, but only 4.1GHz with all cores loaded)
4. EDC/TDC

Most likely you're at the frequency cap.

Thank you for the replies too, I don't think I was at cap, because as I said, even with the new cooler the CPU was not as good as it should be in comparison with other systems, and when I undervolted the clocks became "normal".
I've been pondering what could cause that for a few months and haven't reached a conclusion yet, maybe silicon lottery?
There's a guy that has the same mobo and CPU and achieves higher clocks with stock cooler, I don't have the link with me now unfortunately

 

[Alvesss154]

Thank you for the replies too, I don't think I was at cap, because as I said, even with the new cooler the CPU was not as good as it should be in comparison with other systems, and when I undervolted the clocks became "normal".
I've been pondering what could cause that for a few months and haven't reached a conclusion yet, maybe silicon lottery?
There's a guy that has the same mobo and CPU and achieves higher clocks with stock cooler, I don't have the link with me now unfortunately

Ps: I did watch the video, thanks for sending it to me, very informative.

 

[tennis2]

When benchmarking the "spike" of clocks were similar to what is was before with the stock cooler, and edc and power was still the same (95a and 62w).
Now that it's undervolted clock speeds are higher, EDC maxing out, but power still at 62w.

Very likely that since you have an A520 board, the manufacturer may have set a 65W limit to protect the weak VRMs. My 5600G does 75W CPU Package Power in Prime95 Small FFT @ 4.0GHz on all 6 cores. I'm above 70W with 3+ cores loaded.

 

[drea.drechsler]

Thank you for all the replies and sorry for bothering you that much, but I still have one last question.
Initially I was using the stock cooler and it was pretty bad, my clocks didn't go as high and dropped fairly quickly, so I changed to an 120mm aio.
The new cooling solution works as a charm, temps went from 85 to 65 degrees on full load. CPU clock improved but I noticed that they were not as good as they should be, the clocks were on par with similar systems that didn't even have an aio, just the stock cooler.
When benchmarking the "spike" of clocks were similar to what is was before with the stock cooler, and edc and power was still the same (95a and 62w).
Now that it's undervolted clock speeds are higher, EDC maxing out, but power still at 62w.
I don't know if it's my motherboard, because I thought with a better cooling solution my PC would boost higher and I wouldn't need to undervolt.
Do you know why my system is a bit "worse" than most others with similar specs?
Ps: on stock configuration or not, it's kinda hard to hit 4450 mhz on games as I see in most people 5600g

Better cooling won't make your CPU boost to a higher max clock than stock but it will hold higher mid-range boost clocks, longer, under load.

You can see it hitting max clocks best when using HWInfo64 as your monitoring program with polling period set fairly low 200-500mS. Look at individual core clocks to catch the max clock speed of each core. Do this with nothing running, just mousing around, moving Windows around or maybe running a Defender quick scan for Viruses. As soon as you start running a heavier processing load it will warm up the CPU a little and it won't hit max clocks very much if at all.

Ryzen's extremely dynamic boost algorithm with multiple cores makes it nigh on impossible to gauge performance by looking at clocks alone. You have to use a benchmark program. The max clocks it's rated at (4.4Ghz for a 5600G) it only gets on light, bursty workloads on a single core at a time. Those max clocks have very little impact on performance...it's the mid-range clocks it settles in to as the core heats up that matters. But even under a heavy load it's dithering clocks constantly up and down by maybe 200Mhz in order to keep cores cool.

And last thing: you gotta take many of the reports you read online with a big dash of salt. You've no idea just how they determined it or what they did to get it. So don't compare clocks, compare benchmark scores to determine if your system works well. There are a number of published Cinebench 20 scores from reliable review sites, those are the best thing to look for.

 

[Alvesss154]

Better cooling won't make your CPU boost to a higher max clock than stock but it will hold higher mid-range boost clocks, longer, under load.

You can see it hitting max clocks best when using HWInfo64 as your monitoring program with polling period set fairly low 200-500mS. Look at individual core clocks to catch the max clock speed of each core. Do this with nothing running, just mousing around, moving Windows around or maybe running a Defender quick scan for Viruses. As soon as you start running a heavier processing load it will warm up the CPU a little and it won't hit max clocks very much if at all.

Ryzen's extremely dynamic boost algorithm with multiple cores makes it nigh on impossible to gauge performance by looking at clocks alone. You have to use a benchmark program. The max clocks it's rated at (4.4Ghz for a 5600G) it only gets on light, bursty workloads on a single core at a time. Those max clocks have very little impact on performance...it's the mid-range clocks it settles in to as the core heats up that matters. But even under a heavy load it's dithering clocks constantly up and down by maybe 200Mhz in order to keep cores cool.

And last thing: you gotta take many of the reports you read online with a big dash of salt. You've no idea just how they determined it or what they did to get it. So don't compare clocks, compare benchmark scores to determine if your system works well. There are a number of published Cinebench 20 scores from reliable review sites, those are the best thing to look for.

Thank you very much, that's pretty enlightening. And again, sorry for taking your time.