[SOLVED] Is Ryzen 3 3100 4.3GHz 1.25V good?

[Haylaz2961]

On Prime95 Small FTT's it gave no errors for 10 minutes, after 10 min i'm stop test. Cinebench R23: 6406 pts.
Also i played 30 min Apex Legends and it was nice.
Cooler: Snowman T6 with MX-4.

 

[tennis2]

I have a 5600G, so some of my knowledge may not apply for 3000 series settings, but here's what I did for my 5600G tuning:

• Have HWinfo64 running. Look for "Core Effective Clocks" group. While the normal(?) core frequency readout may be 4.3GHz, the "Effective" core frequency is the important one that drives performance.
  • On the group just above that which shows normal(?) core clocks, you may see a couple numbers in parenthesis (perf #2/3 for example). That's the windows and amd performance ratings of that respective core. The greater the number(s), the worse the core. This will help direct your eyes in the steps below.
• Run Prime95 small FFT
• Once Prime95 is running, reset the HWinfo64 readouts (the clock dial symbol at the bottom). You're going to be looking for any threads that drop below the average effective clock. That indicates instability. Oftentimes, these unstable cores will result in a thread crash within Prime95 (one of the boxes, they're in order).
• Based on that knowledge you know if you need to increase voltage to regain stability, or if you can push lower. On the 5000 series PBO Curve Optimizer you have the option to set negative offsets on a per-core or all-core basis. However, if you end up with -20, -30, -30, -20, -30, -30 (for example), but either of those -20 cores gets activated, ALL cores will follow that more stringent voltage offset. The only time you'll get the full -30 is if none of the -20 cores are active.
• If Prime95 can run for at least 30 minutes without crashing any threads or the whole system/program, you're pretty stable.

 

[tennis2]

1.25V for 4.3GHz is slightly better than average, yes.

 

[Haylaz2961]

1.25V for 4.3GHz is slightly better than average, yes.

Just now i tried 4.3 and 1.225. It didn't cause a problem on Cinebench R23 and 30 minutes of Aida64 FPU, but in Prime95 it failed at the 2nd minute.
Should I use with 4.3 and 1.225?

 

[tennis2]

in Prime95 (1.225V) failed at the 2nd minute.
Should I use with 4.3 and 1.225?

Clearly not

 

[tennis2]

I have a 5600G, so some of my knowledge may not apply for 3000 series settings, but here's what I did for my 5600G tuning:

• Have HWinfo64 running. Look for "Core Effective Clocks" group. While the normal(?) core frequency readout may be 4.3GHz, the "Effective" core frequency is the important one that drives performance.
  • On the group just above that which shows normal(?) core clocks, you may see a couple numbers in parenthesis (perf #2/3 for example). That's the windows and amd performance ratings of that respective core. The greater the number(s), the worse the core. This will help direct your eyes in the steps below.
• Run Prime95 small FFT
• Once Prime95 is running, reset the HWinfo64 readouts (the clock dial symbol at the bottom). You're going to be looking for any threads that drop below the average effective clock. That indicates instability. Oftentimes, these unstable cores will result in a thread crash within Prime95 (one of the boxes, they're in order).
• Based on that knowledge you know if you need to increase voltage to regain stability, or if you can push lower. On the 5000 series PBO Curve Optimizer you have the option to set negative offsets on a per-core or all-core basis. However, if you end up with -20, -30, -30, -20, -30, -30 (for example), but either of those -20 cores gets activated, ALL cores will follow that more stringent voltage offset. The only time you'll get the full -30 is if none of the -20 cores are active.
• If Prime95 can run for at least 30 minutes without crashing any threads or the whole system/program, you're pretty stable.

 

[Karadjgne]

1.250 to 1.225 is a decently sized drop in voltage. You know 1.250 is stable, so if looking for lower, drop it to 1.242v first, then 1.234v etc.

Personally, I don't mess with Ryzen vcore, has a nasty byproduct of killing single thread performance, which uses higher voltages to get the full boost speeds on more cores, for longer.

So when running cinebench, check single as well as all core, if single is taking a hit but multi is going up, your fps could suffer since games use a Master thread, all other threads are used in support of the Master.

 

[tennis2]

Personally, I don't mess with Ryzen vcore, has a nasty byproduct of killing single thread performance, which uses higher voltages to get the full boost speeds on more cores, for longer.

So when running cinebench, check single as well as all core, if single is taking a hit but multi is going up, your fps could suffer since games use a Master thread, all other threads are used in support of the Master.

(to expand on this) This is why it's better to use a voltage OFFSET instead of a voltage CAP. Especially if you aren't going fully manual tuning such that you still have different maximum frequencies depending on how many cores are loaded.

For example, my 5600G does 4GHz all-core @ 1.2V, but 4.6GHz single-core @ 1.44V. If I'm testing with Prime95 loading all cores, I'm going to see 4GHz and tune for that. But if I would've set a fixed 1.2V cap, I won't see 4.6GHz because the CPU won't be able to ride up the frequency-voltage curve that far.

Again, if you're doing fully manual tuning some/all of this may be a moot point, but still worth being aware of.

 

[Karadjgne]

I used ClockTuner2 and Dram Calculator. The CTR2 has 2 profiles and a default, so I get stock idle settings upto 18% loads, 4.4GHz on 1-4 cores upto 55% loads and 4.29GHz all core on a 3700x. Single thread went up about 30pts, multi thread went up almost 400 points in cinebench r20. Temps dropped 20°C+ for the profiles since the cpu VID is capped at 1.325v.

That's what the offset does really, lowers the VID, which lowers cpu demand without affecting cpu usage.

 

[Koekieezz]

(to expand on this) This is why it's better to use a voltage OFFSET instead of a voltage CAP. Especially if you aren't going fully manual tuning such that you still have different maximum frequencies depending on how many cores are loaded.

For example, my 5600G does 4GHz all-core @ 1.2V, but 4.6GHz single-core @ 1.44V. If I'm testing with Prime95 loading all cores, I'm going to see 4GHz and tune for that. But if I would've set a fixed 1.2V cap, I won't see 4.6GHz because the CPU won't be able to ride up the frequency-voltage curve that far.

Again, if you're doing fully manual tuning some/all of this may be a moot point, but still worth being aware of.

Could i ask? So by using voltage offset it could be like a same measurement like intel turbo boost? like if im doing 1.35v only if it's all core boos overclock, but when idle or at lower clock, it will drop below that voltage? im new to ryzen cpu overclocking and my aim is to know wether amd could do thing like intel turbo boost and speedstep, lower clock and voltage when nothing heavy or not using much processing power, and crank up voltage and clock automatically to the assigned max? it's just been a displeasure to do 1.35v on all core wether it is idle to light or medium usage. more likely im asking about dynamic voltage for overclock. Thank you

 

[Karadjgne]

It's a Ryzen, not an intel. It doesn't work the same, voltages are applied and used differently. By trying to force a Ryzen to behave like an intel you cripple it, and that generally applies to OC.

Intels are for the most part non self governing. The more you give it, the faster it goes and that works great as long as you can keep the temps under throttle temps. A Ryzen is highly self governing. It'll give you the best it can without allowing you to hurt it. Ryzens are efficiency engines, the more efficient you make it, the more it can give.

Lowering voltages will lower temps, which allows the Ryzen to boost higher, for longer. So whereas an intel will chop the power limits (PL2 down to PL1 at Tau) a Ryzen just keeps chugging away.

An example would be my 3700x, I get higher cinebench and higher fps at 4.29GHz all core and 62°C than I do at 4.4GHz all core and 84°C. And that's without any PBO, because my cpu maxed out is under the power limit, so has no need for PBO to raise them. It literally 'thinks' faster, regardless of clock speeds, because the temp is lower, but there's enough voltage to allow it to behave as it requires.

Chopping voltages to absolute minimum required for stability in order to lower temps, is Intel OC. Using a large enough Efficient cooler to moderate temps, coupled with an offset/vid drop in demanded voltage to minimum working voltage is Ryzen OC.

 

[tennis2]

Could i ask? So by using voltage offset it could be like a same measurement like intel turbo boost? like if im doing 1.35v only if it's all core boos overclock, but when idle or at lower clock, it will drop below that voltage? im new to ryzen cpu overclocking and my aim is to know wether amd could do thing like intel turbo boost and speedstep, lower clock and voltage when nothing heavy or not using much processing power, and crank up voltage and clock automatically to the assigned max? it's just been a displeasure to do 1.35v on all core wether it is idle to light or medium usage. more likely im asking about dynamic voltage for overclock. Thank you

Using an offset will allow the CPU to utilize lower vCore for lower frequencies, yes.

 

[Haylaz2961]

@tennis2 I didn't want to go over 1.3V so I decided to use it at 4.2GHz. I did 4.2GHz and 1.175V. Is it good now? It didn't give any crashes 30 min Prime95 Small FTT's.
Should I decrease Volt more?

 

[tennis2]

4.2GHz @ 1.175V sounds pretty baller. If that's stable for you, I'd call it a resounding success. Save your settings as a profile in your BIOS so it's easy to come back to.

Should you push for lower voltage? That depends on how much more time you want to spend testing.

 

[Koekieezz]

Also try to OC the ram, that would help 3100 delay between CCXs, pretty much improve the latency down.

 

[Haylaz2961]

Also try to OC the ram, that would help 3100 delay between CCXs, pretty much improve the latency down.

I'm overclocked it a long time ago. But my chip is very bad.
3000MHz CL16 --> 3600MHz CL18. That's best result. [XPG Gammix D30 (8*2) 16GB]

 

[Haylaz2961]

@tennis2 Is that voltages of CPU Core normal?
I cant upload photo.
Minimum: 1.155
Maximum: 1.177
Average: 1.167

 

[Karadjgne]

And fps went down...