Question I FINALLY Figured it out - Over/Undervolting my Ryzen 3000

[HOLDMYPC]

I have an A320M board running a ryzen 5 3500 - By default it was running at 1.1V (SoC) - and
i got a Cinebench R23
Multi core: 6003
Single core: 1098
Since its and A320 and Ryzen 3000 - i have no curve or OC
i tried an Overvolt - SoC Offset 100mV+
R23 - MC - 6341
SC - 1136
OV - 200mV+ offset
MC - 5888
SC - 1088
OV - 250mV+
MC - 6375
SC - 1150
Note - My temps went from 80 degrees to 85 degrees (stock AMD cooler) - i live in ambient temps of around 30-35 degree
So i was concerned about the temps
i tried undervolting
UV - 100mV- (1.0V)
And Voila -
MC - 6378
SC - 1148
with temps at 80 degrees and 7W fall in CPU power consumption

Was there a mistake in my methodology - please provide some insights whether this is how i was supposed to do it- i tried asking on reddit communities... no one responded so i tried to tinker myself
- all my other settings were constant throughout testing

 

[Roland Of Gilead]

Hey there,

What CPU do you have? Is it the 3600 you mean? There is a 3500u which is a mobile CPU.

Please list your full PC specs, including PSU, case, fans etc. or make and model of laptop if that's what you have.

What exactly are you trying to achieve? Faster bench scores? Better gaming FPS?

 

[HOLDMYPC]

Hey there,

What CPU do you have? Is it the 3600 you mean? There is a 3500u which is a mobile CPU.

Please list your full PC specs, including PSU, case, fans etc. or make and model of laptop if that's what you have.

What exactly are you trying to achieve? Faster bench scores? Better gaming FPS?

I have a Desktop 3500 - its a 3600 with SMT off sold for cheaper, better CLOCKS and performannce - my rated clocks r 4.1GHz single core but single core never crosses 3.9GHz, after that last undervolt it started occasionally touching 4.0GHz, PSU is a corsair CV550 temps r fine considering stock cooler with max temps hitting 80 degrees in all core, and 67-69 in games

 

[Karadjgne]

Yes, there was a mistake in the methodology. You operated under the assumption that more voltage = higher performance, but when it comes to Ryzen its generally the opposite.

Ryzen boosts are governed by 3 things. Temps, voltages and load. Assuming the load is static, same for all tests, your boosts will be governed by temps and voltages. If you increase voltage, 2 things happen. First is you increase temps, second is you lower headroom since the Ryzen has internal limits and won't allow itself to go beyond, lowering boost amounts to maintain voltages in acceptable range.

Somewhere around 60°C is the optimal temp for a Ryzen. Boosts will lower per core by 50-100MHz as you get closer to @ 80°C. After that, boosts drop off a lot more aggressively.

Ryzen are an efficiency engine. Intel are powerhouse engines. The more efficient you make a Ryzen, the better it performs. By undervolting, you dumped a bunch of heat which allows more headroom, higher boosts per core, for longer periods of time, and so the CB scores went up.

With a more effective cooling solution, and making no other changes, your CB scores will increase, even if temps remain the same, because instead of occasionally hitting 4.0GHz, you'll be constantly 4.0GHz and more than occasionally hitting 4.1GHz.

 

[HOLDMYPC]

Yes, there was a mistake in the methodology. You operated under the assumption that more voltage = higher performance, but when it comes to Ryzen its generally the opposite.

Ryzen boosts are governed by 3 things. Temps, voltages and load. Assuming the load is static, same for all tests, your boosts will be governed by temps and voltages. If you increase voltage, 2 things happen. First is you increase temps, second is you lower headroom since the Ryzen has internal limits and won't allow itself to go beyond, lowering boost amounts to maintain voltages in acceptable range.

Somewhere around 60°C is the optimal temp for a Ryzen. Boosts will lower per core by 50-100MHz as you get closer to @ 80°C. After that, boosts drop off a lot more aggressively.

Ryzen are an efficiency engine. Intel are powerhouse engines. The more efficient you make a Ryzen, the better it performs. By undervolting, you dumped a bunch of heat which allows more headroom, higher boosts per core, for longer periods of time, and so the CB scores went up.

With a more effective cooling solution, and making no other changes, your CB scores will increase, even if temps remain the same, because instead of occasionally hitting 4.0GHz, you'll be constantly 4.0GHz and more than occasionally hitting 4.1GHz.

I see - i found that out the hard way (mentioned at end of post) - now im running at 1.0 V instead of default 1.1V at 67-69 degree in games where it hits 4025MHz occasionally - should i try little more ... in terms of tuning voltage down

 

[Karadjgne]

With cpus there's a line drawn in the sand as far as voltages go. They'll handle anything above that line (within reason) just fine, no difference to performance if the temp is regulated. But, and that's a Capital B, cross that line and you're done.

So no, lower to your hearts content, but take notes and move voltages in tiny steps because you'll hit that invisible line sooner or later and require you back up a step. That's an essential part of OverClocking, lowering voltages as far as they'll go, until you crash, then bumping back up one step or two to maintain stability.

1.2 to 1.1 is a very big step. When dealing with voltages, like the vcore on my old Intel, I was moving from 1.132v to 1.124v to 1.118v etc, drops of 0.008v, not a full 0.1v.

Your CB score is not that important as a number to attain, it should be more important as a number to judge by. And that number isn't static per test as it'll start/end blocks in difference. So multiple passes may see 6000 to 6050, with zero changes to anything. What you'd be looking for is trends, changing a voltage and seeing if that number is always closer to 6000 or closer to 6050, change voltage again and does that change etc.

 

[drea.drechsler]

I see - i found that out the hard way (mentioned at end of post) - now im running at 1.0 V instead of default 1.1V at 67-69 degree in games where it hits 4025MHz occasionally - should i try little more ... in terms of tuning voltage down

Please note, ever since gen 2 Ryzen CPU's do let you lower voltage quite a lot and it will stay stable. You might think you're helping things, and even think it's holding higher clocks for longer if just looking at that. But if you run actual performance tests you're very likely to find it performs more poorly, often by a lot. It's been called clock stretching in the past, although I don't know if that's technically accurate. But it seems to be a response to maintain stability when voltage is artificially too low for it.

So, when you've lowered voltage and feel it's stable run some real-world benches that stress the processor all-core: Cinebench is very good for this. Be sure to have some baseline scores to compare to (scores where settings are stock).

 

[HOLDMYPC]

Please note, ever since gen 2 Ryzen CPU's do let you lower voltage quite a lot and it will stay stable. You might think you're helping things, and even think it's holding higher clocks for longer if just looking at that. But if you run actual performance tests you're very likely to find it performs more poorly, often by a lot. It's been called clock stretching in the past, although I don't know if that's technically accurate. But it seems to be a response to maintain stability when voltage is artificially too low for it.

So, when you've lowered voltage and feel it's stable run some real-world benches that stress the processor all-core: Cinebench is very good for this. Be sure to have some baseline scores to compare to (scores where settings are stock).

Yea i noticed while going lower than 1.0V - the performance did drop but from 1.1V to 1.02V the cinebench score increased and well it did hit occasional 4.1GHz compared to 1.1V baseline, scores also increased with 1.3V but temps increased by 5 degrees but at 1.02V my temps r better like down 1 degree from 1.1V which may be margin of error but it's fines ince i get more performance

 

[HOLDMYPC]

With cpus there's a line drawn in the sand as far as voltages go. They'll handle anything above that line (within reason) just fine, no difference to performance if the temp is regulated. But, and that's a Capital B, cross that line and you're done.

So no, lower to your hearts content, but take notes and move voltages in tiny steps because you'll hit that invisible line sooner or later and require you back up a step. That's an essential part of OverClocking, lowering voltages as far as they'll go, until you crash, then bumping back up one step or two to maintain stability.

1.2 to 1.1 is a very big step. When dealing with voltages, like the vcore on my old Intel, I was moving from 1.132v to 1.124v to 1.118v etc, drops of 0.008v, not a full 0.1v.

Your CB score is not that important as a number to attain, it should be more important as a number to judge by. And that number isn't static per test as it'll start/end blocks in difference. So multiple passes may see 6000 to 6050, with zero changes to anything. What you'd be looking for is trends, changing a voltage and seeing if that number is always closer to 6000 or closer to 6050, change voltage again and does that change etc.

I did that - and the difference is larger like 300 points between voltages (in post)