5800x Users - voltage settings.

[ratirt]

Hi everybody.
I been testing my 5800x from time to time. I undervolted it a bit to see how it goes and how low I can go with it.
I been wondering, how low can you go with vcore without losing performance? I been trying to use a set vcore but I have settled for an offset and it would seem it works. I will need to fiddle with it a bit more when I get the chance. Did you change vcore only or SoC as well? How does the boosting look after tweaks?

 

[Hofnaerrchen]

I did not touch vcore at all but I changed PPT, TDC and EDC (125, 80, 100) to lower values. This reduced power consumption and thus heat in a way I wanted it to be without losing performance at all. In Prime95 the CPU still boosts to 4.65 GHz all core which is fine for me.

 

[ratirt]

I did not touch vcore at all but I changed PPT, TDC and EDC (125, 80, 100) to lower values. This reduced power consumption and thus heat in a way I wanted it to be without losing performance at all. In Prime95 the CPU still boosts to 4.65 GHz all core which is fine for me.

Well, I did the -0.125v offset and the vcore settles at 1.25v or so (i need to check that correctly, been changing things a lot recently).
It boosts well over 4.6Mhz but I still think there's something i can improve still. As usual i did not have enough time to explore the voltage more.
Maybe I need to give it a shot with the PPT, TDC and EDC and see where will this take me.

 

[eazen]

I think if you reduced voltage it will pretty much do the trick, no need to fiddle around with those other settings. Doing the other way around, PPT etc. is just another way to limit power output of the CPU, but in my opinion under volting is much better.

Also I would test undervolt by doing a CPU benchmark in between like CB, that way you will know once you reduced voltage too much and lost performance.

 

[ratirt]

I think if you reduced voltage it will pretty much do the trick, no need to fiddle around with those other settings. Doing the other way around, PPT etc. is just another way to limit power output of the CPU, but in my opinion under volting is much better.

Also I would test undervolt by doing a CPU benchmark in between like CB, that way you will know once you reduced voltage too much and lost performance.

I been doing the CB23 mostly to check if there is not regression in performance. It would seem, there is not and in some cases the performance is higher.
I need to find the sweet spot. It would also seem (kinda obvious though) the heat output change makes a difference in the boost and performance as well.
I was just wondering about the settings other users used to attain this level of performance and how does that relate to my finding and settings used.

@EarthDog why a sad face bro?

 

[tabascosauz]

I been doing the CB23 mostly to check if there is not regression in performance. It would seem, there is not and in some cases the performance is higher.
I need to find the sweet spot. It would also seem (kinda obvious though) the heat output change makes a difference in the boost and performance as well.
I was just wondering about the settings other users used to attain this level of performance and how does that relate to my finding and settings used.

@EarthDog why a sad face bro?

I don't think the issue is normally that MT scores in Cinebench are affected, it's ST performance. Back on my crappy 3700X, when we didn't have Curve Optimizer I ran a -0.075V offset with strong LLC and MT scores were comparable, but I lost single-thread score. ST Vcore was about 1.41V but even that was too low to hit the low stock ST clocks of about 4.385GHz (since CO didn't exist).

And Ryzen 5000 MT Vcore (barely 1.2V) is even lower than 3000 MT Vcore, so it stands to reason that there's no impact on MT perf if your power limit is still 142W or less.

The chips generally need most of that 1.5V Vcore ceiling to hit their ST clocks, without the help of CO changing the V-F curve.

Of course, if you don't notice/care for the loss in ST performance, there's nothing really wrong with doing it this way. Just that CO undervolt and power limits are still the "right" way to do it, as they shouldn't affect your 5800X from hitting at least 4850MHz ST.

 

[eazen]

I been doing the CB23 mostly to check if there is not regression in performance. It would seem, there is not and in some cases the performance is higher.

Yea, unless you lower the voltage too much there will not be a regression in performance. Same as with my 5800X3D.

 

[freeagent]

Yea, unless you lower the voltage too much there will not be a regression in performance. Same as with my 5800X3D.

The difference between your X3D and other Zen 3 chips is the others actually boost. So you can lower your voltage quite a bit and get away with it, but the rest of us will see our top boost clocks drop across the cpu.

 

[eazen]

The difference between your X3D and other Zen 3 chips is the others actually boost. So you can lower your voltage quite a bit and get away with it, but the rest of us will see our top boost clocks drop across the cpu.

Every Ryzen since gen 1 has PB or Precision Boost. So I don’t know what you’re on about. I have only slightly lowered the voltage of mine, otherwise I will have a regression. The OP has also stated that he did not lose performance with his UV. People way too often don’t read the OP and just assume things.

 

[Hofnaerrchen]

The point about changing PBO settings is: You don't have to do manual tweaking of voltages - I understand why some people like doing it, but modern cpus/gpus are extremely good at selecting the needed voltages - a reason why manual overclocking apart from LN OC is quite boring these days. By limiting the total power consumption you can get some quite impressive power savings while retraining most if not all of the performance. Ryzen CPUs use quite high vcore voltages in ST scenarios but this is still within their specs. In MT tasks the voltages are much lower by default but - and I might repeat myself here - reducing PBO settings has a much higher impact on total power consumption and generated heat than reducing one or two voltage settings.

 

[freeagent]

J

The point about changing PBO settings is: You don't have to do manual tweaking of voltages

Right, to get the most out of your zen3 is to use curve optimizer in conjunction with ppt tdc and edc limits, based on the available voltage. 58x3d is limited right now in this regard.

but modern cpus/gpus are extremely good at selecting the needed voltages

Not that good, this is why the let us tune to up to +200MHz, if the cpu is willing.

By limiting the total power consumption you can get some quite impressive power savings while retraining most if not all of the performance.

That’s not entirely true. You limit current, and you limit the CPU’s overall output. Be it clocks, or actual horsepower.

Ryzen CPUs use quite high vcore voltages in ST scenarios but this is still within their specs. In MT tasks the voltages are much lower by default but - and I might repeat myself here - reducing PBO settings has a much higher impact on total power consumption and generated heat than reducing one or two voltage settings.

As noted earlier, that high vcore is to drive high single core speeds. If you try to run a static 1.5v your cpu will not be very happy, or cooperative. And as I said earlier, if you reduce PBO settings, you will reduce the cpu output, both performance and thermal.

 

[tabascosauz]

Every Ryzen since gen 1 has PB or Precision Boost. So I don’t know what you’re on about. I have only slightly lowered the voltage of mine, otherwise I will have a regression. The OP has also stated that he did not lose performance with his UV. People way too often don’t read the OP and just assume things.

5800X3D has a hard global limit of 4550MHz, lower than a PBO 5600G. Zen 3 on N7 can literally do 4.5 in its sleep, you barely need any Vcore at all.

Same does not hold true for the other SKUs. 5800X global limit is 4850MHz stock, and max PBO ceiling is 5050MHz. Not hitting either of those on 1.25V.

That said, especially on Windows 11 which does the same stuff with less clocks, being stuck at "only" 4.6 for games isn't a big deal and not even noticeable most of the time.

 

[Valantar]

In my experience with my 5800X, there's no classic UV that doesn't come with a performance penalty, as you're just putting roadblocks in the way of the boost control systems. If you want to save power but maintain performance, CO + lowered PPT/EDC/TDC are the way to go, with the added possibility of a negative PBO offset if it still tries to boost more aggressively than you want it to. Classic OC methodologies either hurt ST performance or just underperformed generally compared to CO and PBO tuning.

 

[Hofnaerrchen]

As noted earlier, that high vcore is to drive high single core speeds. If you try to run a static 1.5v your cpu will not be very happy, or cooperative. And as I said earlier, if you reduce PBO settings, you will reduce the cpu output, both performance and thermal.

For this reason I do not use static voltages on my CPUs at all but the 5800X can run very well with reduced PBO settings without losing (much) performance - same is true for GPUs (reducing the power limit). As the OP mentioned, if I'd wanted to toggle voltages I would also have used offsets. With my settings the CPU runs either 4.85 GHz (ST) or 4.65GHz (all core) in typical usage scenarios and temperatures dropped well below 80° C (ambient approx. 20-22° C) using air cooling.

 

[FreedomOfSpeech]

110/90/100, CO -14, +200MHz, VCore Offset -0,1V.
SC Boost 4.9GHz, MC Boost 4.5GHz (@110W and 70°C). Kraken X63 doesn't ramp up.
Pump 25-40%, Fans 25-40%.

SoC Voltage fixed @1.2V, for IF1900MHz and 3800MHz RAM (1.45V).

 

[AMF]

disable core boost pbo pbo2 enable xfr enable sense mi and leave cpu voltage to auto and then select multiplier close to your boost max pump 100% then tune your fans

 

[Valantar]

disable core boost pbo pbo2 enable xfr enable sense mi and leave cpu voltage to auto and then select multiplier close to your boost max pump 100% then tune your fans

Oh dear, that's rather terrible advice. With that approach you'll be losing ST performance while gaining nothing meaningful in MT, and likely needlessly boosting power draw. A much better approach is to use PBO in conjunction with CO, using a negative CO offset to cool the cores and allow them to boost more/higher within a given power envelope, and using PBO offsets to increase clocks. This will net you the best overall performance, the best dynamic efficiency while maintaining performance, and keep all protections enabled.

For this reason I do not use static voltages on my CPUs at all but the 5800X can run very well with reduced PBO settings without losing (much) performance - same is true for GPUs (reducing the power limit). As the OP mentioned, if I'd wanted to toggle voltages I would also have used offsets. With my settings the CPU runs either 4.85 GHz (ST) or 4.65GHz (all core) in typical usage scenarios and temperatures dropped well below 80° C (ambient approx. 20-22° C) using air cooling.

Offsets can easily lead to crashes at idle though, as there's typically far less UV headroom there than when boosting. That's the magic of curve optimizer - rather than a fixed offset it adjusts the DVFS curve with a higher offset at high clocks. And, of course, it allows for a lot more fine-grained tuning than a voltage offset, with per-core offsets. My 5800X runs better with my CO offsets (-7 on first two cores, - 20 on the rest) than it ever could with a static UV.

 

[AMF]

u set a multplier it boosts single core and multi it boosts beyond specs be sure to enable cool n quiet as well of course if ur rig isnt worthy it wont benefit u does my single core suffer? proof is in the puddin.. granted im not 5800x but it works the same .

 

[Valantar]

u set a multplier it boosts single core and multi it boosts beyond specs be sure to enable cool n quiet as well of course if ur rig isnt worthy it wont benefit u does my single core suffer? proof is in the puddin.. granted im not 5800x but it works the same .

So you're talking about a CPU that doesn't support curve optimizer in the first place. That makes sense - if it isn't an option, you obviously can't use it. For anything newer, CO is the way to go, and PBO makes it far easier to find a high and stable "OC" that covers both ST and MT without limiting either. The lack of CO definitely makes it less useful though, as you're stuck with either fixed voltages or fixed offsets, both of which are far less ideal for fine tuning.

 

[AMF]

So you're talking about a CPU that doesn't support curve optimizer in the first place. That makes sense - if it isn't an option, you obviously can't use it. For anything newer, CO is the way to go, and PBO makes it far easier to find a high and stable "OC" that covers both ST and MT without limiting either. The lack of CO definitely makes it less useful though, as you're stuck with either fixed voltages or fixed offsets, both of which are far less ideal for fine tuning.

my voltages are not fixed .... my voltage is set to auto i have cool and quiet enabled..... if i had better cooler it would boost even higher voltages range from 0.698-1.417 at moment it tunes itself based on thermal headroom xfr is very interesting xfr+neural net aka sense mi it tunes all cores by setting a multiplier u get fantastic single thread speeds as well as boost to multi i run balls out my max boost clock on all cores all the time except when idle i run around 350-500mhz on all cores idle while maintaining thermals say what u want i dont see it working better than that

 

[Valantar]

my voltages are not fixed .... my voltage is set to auto i have cool and quiet enabled..... if i had better cooler it would boost even higher voltages range from 0.698-1.417 at moment it tunes itself based on thermal headroom xfr is very interesting xfr+neural net aka sense mi it tunes all cores by setting a multiplier u get fantastic single thread speeds as well as boost to multi i run balls out my max boost clock on all cores all the time except when idle i run around 350-500mhz on all cores idle while maintaining thermals say what u want i dont see it working better than that

That's indeed about the best you can get on Ryzen 3000. Rathet impressed that you're able to do that on auto voltage, though I guess that means your chip is either relatively well binned or had more reasonable clocks to start with Ryzen 5000 enables you to go a few steps further through tuning per-core voltage curves and allowing for more dynamic boosting. So, more control, more performance, better utilization of the smarts of the CPU's control systems. It's not a huge difference, but given how it's also incredibly easy and allows for impressive efficiency, it's the way to go for 5000-series.

 

[AMF]

That's indeed about the best you can get on Ryzen 3000. Rathet impressed that you're able to do that on auto voltage, though I guess that means your chip is either relatively well binned or had more reasonable clocks to start with Ryzen 5000 enables you to go a few steps further through tuning per-core voltage curves and allowing for more dynamic boosting. So, more control, more performance, better utilization of the smarts of the CPU's control systems. It's not a huge difference, but given how it's also incredibly easy and allows for impressive efficiency, it's the way to go for 5000-series.

right on cool cool i guess i did hit the lottery .... i never ever throttle no matter what im doing just wish was a magic button to enable 8 more cores lol the pro line of processors must be gold tier binning

https://www.cpu-world.com/CPUs/Zen/AMD-Ryzen 7 PRO 3700.html here is the stats