5800x undervolt seems off

[deen]

Hi,

Got my 5800x last week. Spend some time reading about undervolting for more performance with less voltage and temps. Been playing around for a while now but something seems off.

System:

5800x (arctic liquid freezer ii 360)
Gigabyte X570 pro (latest bios)
16gb 3200mhz cl14
Latest win updates + chipset drivers etc.

I managed to get the following stable vcurve offsets:

-8
-30
-30
-30
-30
-5
-30
-5

Did set a +100mhz boost. Can prob. run 150-200mhz.

Measuring my vcore with hw64info, something seems off. At stock my vcore in multithreaded apps is hovering around 1.33v. With vcurve enabled I should see way lower voltages but with the above vcurve set I'm seeing it hovering around 1.33-.135v in multicore benches.

Also, It's running around 4.55 ghz in multicore benches. Although hw64info says it's running @ 4.75. But when looking in Ryzen master and task manager both display ~4.5.

My cinebech mc score = 6050 (same or even less then stock)
My single core score = 640

Temps are no problem at all. Staying below 75 degrees celcius all the time.

I'm wondering why it seems to be stuck at 4.5 ghz all core and running 1.35v while having this much offset already. Even so, stock settings have the same or even less voltages.

Could this be a bios problem? Anyone else with a gigabyte mobo running F33?

Thanks!

 

[Joxeon]

I've found multi core voltages and clocks will only apply from the smallest offset, IE if you have 7 cores set at -30 but 1 core at -5 then multicore workloads clocks and voltages will apply like you have all cores set to -5.

Knock off the 100mhz as its easier to run lower offset and set all cores to -20 etc and you should see all core boosts to around 4.65 and voltages around 1.26, now set 1 of the cores to -5 and you will see all core clocks drop and voltages go up like if you have all cores set to -5

 

[deen]

I've found multi core voltages and clocks will only apply from the smallest offset, IE if you have 7 cores set at -30 but 1 core at -5 then multicore workloads clocks and voltages will apply like you have all cores set to -5.

Knock off the 100mhz as its easier to run lower offset and set all cores to -20 etc and you should see all core boosts to around 4.65 and voltages around 1.26, now set 1 of the cores to -5 and you will see all core clocks drop and voltages go up like if you have all cores set to -5

Hmm, ok but is that how it's supposed to work or is this a bug? Why would you set offsets for all cores individually when the smallest settings on any core will be applied to all cores anyway?

But you might be correct that this is the issue. I've seen higher scores with bigger offsets on all cores before. Will test this!

 

[Joxeon]

This is why I run my 5800X with no extra clock boost but just set a -25 on all cores as this seems offer the best mix of perfomance with lower temps and voltages.

I guess if you get really lucky with the silicon then you can do both.

 

[deen]

So you might as well use the "all core" offset instead of managing individual cores. But it also means if 1 core is stuck on -5 (sometimes it may crash going lower) then all cores will be stuck at -5. Interesting though. I wonder why you can set all cores individually.

 

[alec]

Curve optimizer is not the quite voltage offset. Yes, it would lower voltage for any given speed, but clocl speed is not a fixed value, rather a balancing act between load, temperature, power etc. So by lowering curve offset CPU can boost higher/longer at same voltage. Thats why you don't see voltage drop. Need to lower PBO power limits as well to see the desired effect.

The only suspicious thing seems to be Cinebench score not going up. Multicore should be the most benefitting from CO. Maybe don't go as low as -30, up lower limit to -20, see what happens

 

[Joxeon]

I wonder why you can set all cores individually.

Probably if your more focused on pushing up single core clocks.

Single core works on a per core basis but multicore works on the least undervolted core and sets all cores to this level, that was what I observed from the extensive testing I did when I got my 5800X.

 

[deen]

Curve optimizer is not the quite voltage offset. Yes, it would lower voltage for any given speed, but clocl speed is not a fixed value, rather a balancing act between load, temperature, power etc. So by lowering curve offset CPU can boost higher/longer at same voltage. Thats why you don't see voltage drop. Need to lower PBO power limits as well to see the desired effect.

The only suspicious thing seems to be Cinebench score not going up. Multicore should be the most benefitting from CO. Maybe don't go as low as -30, up lower limit to -20, see what happens

Well, I already tried ajusting PBO power limits but has zero effect. I went all the way down to 125,75,120 where the stock powerlimits are 142,95,140. Still using the same voltage in vcore. But it did ran slower clocks as expected. Score dropping down from 6050 to ~5800. Temps also dropped 5 degrees or so. But I was also expecting a drop in voltage. I did use the PBO power limites and the vcurve at the same time. Looks like the PBO limits have no effect on vcore voltage when vcurve is active.

But Joxeon is probably correct. I did note some other settings I tried. Had all cores at -20, got higher scores and vcore was 1.3v. not sure if it was stable though. Have to test when I get home.

 

[Joxeon]

The way curve optimiser is supposed to work is each -1 you apply equates to reducing voltage by -5mv at light load / SC and -3MV at high load.

The draw back is the -5mv and -3mv levels are not individually adjustable and this IMO is something that means your either limited by single core or limited by multicore unless you have a golden sample which can run +200mhz and negative 30, if you could adjust the amount of offset applied independently at SC / MC workloads then maybe you could get a higher single core boost while also maintaining the higher multicore boost and vice versa.

As it stands it's still a good tool but it's balancing act between higher multicore or higher single core as increasing one will decrease the other.

 

[deen]

The way curve optimiser is supposed to work is each -1 you apply equates to reducing voltage by -5mv at light load / SC and -3MV at high load.

The draw back is the -5mv and -3mv levels are not individually adjustable and this IMO is something that means your either limited by single core or limited by multicore unless you have a golden sample which can run +200mhz and negative 30, if you could adjust the amount of offset applied independently at SC / MC workloads then maybe you could get a higher single core boost while also maintaining the higher multicore boost and vice versa.

As it stands it's still a good tool but it's balancing act between higher multicore or higher single core as increasing one will decrease the other.

Makes sense. Have not seen anyone explaining this anywhere. But i'm curious how this affects PBO limits. In your testing, did you also raise or lower your PBO? I was theory crafting by thinking if a core can only do -15 offset max then raising the PBO limit will give the cores more headroom to work with. I've seen no effect on vcore when lowering PBO limits. Then maybe it also has no effect when raising it.

For example, you can use the "motherboard" PBO setting which increases the stock limits to what the motherboard can deliver. Seen some people in other threads setting PBO limits very high and then -30 all core. But i'm not sure how this will affect load and vcore. Any chance you tested this?

I'm using this rig for gaming only. So the performance tradeoff is that probably for most games a higher single core is faster. Unless you play multiplayers games which are very cpu/core hungry like warzone or battlefield 5. Just a theory though.

 

[Audioboxer]

Also, It's running around 4.55 ghz in multicore benches. Although hw64info says it's running @ 4.75. But when looking in Ryzen master and task manager both display ~4.5.

That might be clock stretching, I guess? Or do you just mean HWINFO says the max it has hit is 4.75? Cause that could be normal. My 5950x hits the max of 5.05ghz on a few cores, but that is obviously light load boosting that these chips can do for a few seconds.

My figures in terms of the curve are quite like yours. My 4 best cores range from -5 up to -20. All the rest are on -30. I tested overnight in CoreCyler to get this stable, as the one thing I noticed is you can do a -25/30 all core undervolt and boot Windows fine and not see it crash/BSOD quickly, but on stability testing a single core or two can fall down quite quickly.

It's one of my 2nd best cores that needs to be down at -5. -10 was unstable on it. Might work with -6~-9, but I haven't tested in that small an increment. -5 is rock stable.

When I do a heavy load/stability testing I tend to get between 4.6 and 4.65 sustained. Often the full 4.65.

My PBO settings are PPT 270, TDC 160 and EDC 190. I've got a +100mhz setting.

 

[Joxeon]

I've always slightly lowered the PPT TDC EDC settings as this reduces heat and improves the boost, the 5800X has the same stock settings as a 5950X yet only half the cores so really doesn't need this much extra power.

 

[deen]

That might be clock stretching, I guess? Or do you just mean HWINFO says the max it has hit is 4.75? Cause that could be normal. My 5950x hits the max of 5.05ghz on a few cores, but that is obviously light load boosting that these chips can do for a few seconds.

My figures in terms of the curve are quite like yours. My 4 best cores range from -5 up to -20. All the rest are on -30. I tested overnight in CoreCyler to get this stable, as the one thing I noticed is you can do a -25/30 all core undervolt and boot Windows fine and not see it crash/BSOD quickly, but on stability testing a single core or two can fall down quite quickly.

It's one of my 2nd best cores that needs to be down at -5. -10 was unstable on it. Might work with -6~-9, but I haven't tested in that small an increment. -5 is rock stable.

When I do a heavy load/stability testing I tend to get between 4.6 and 4.65 sustained. Often the full 4.65.

My PBO settings are PPT 270, TDC 160 and EDC 190. I've got a +100mhz setting.

Nah, hardwareinfo is measuring it wrong i guess. It says 4.75 OC boost all cores when running cinebech MC but ryzen master only says 4.5 ghz all core.

If Joxeon if right , which looks like it, then you will prob have the same problem as me. It does not matter how much boost you can get because that is easy to get. Vcore voltage is the problem here. What vcore voltage do you get when running multicore benches?

Those PPT and other levels are crazy high and will prob never be met. You can check in Ryzen master how much PBO power it draws per setting. But the 5950 has 2 ccd's where the 5800x only has one. So not sure if you can interchange results based on this.

I've always slightly lowered the PPT TDC EDC settings as this reduces heat and improves the boost, the 5800X has the same stock settings as a 5950X yet only half the cores so really doesn't need this much extra power.

Mind sharing your current settings?

 

[Joxeon]

I run PPT 125 TDC 84 EDC 125 with curve optimiser -25 all core with a 0mhz core offset.

 

[deen]

I run PPT 125 TDC 84 EDC 125 with curve optimiser -25 all core with a 0mhz core offset.

Actually tried that but I can't go higher then -15. Windows wont boot otherwise. Strange thing is, im not seeing any lower voltages used vs -5...

Also, my cinebech score is not improved either. Seems like it's using the same voltage no matter what setting.

 

[Audioboxer]

Nah, hardwareinfo is measuring it wrong i guess. It says 4.75 OC boost all cores when running cinebech MC but ryzen master only says 4.5 ghz all core.

If Joxeon if right , which looks like it, then you will prob have the same problem as me. It does not matter how much boost you can get because that is easy to get. Vcore voltage is the problem here. What vcore voltage do you get when running multicore benches?

Those PPT and other levels are crazy high and will prob never be met. You can check in Ryzen master how much PBO power it draws per setting. But the 5950 has 2 ccd's where the 5800x only has one. So not sure if you can interchange results based on this.



Mind sharing your current settings?

I'll double check later on, I'm fine with my chip temperatures (though I have to say they're aboutso other than doing my voltage curve I haven't really paid too much attention to voltages under load.

PPT stays in the green, but EDC maxes out. TDC tends to get into 80 odd % from what I can remember. 160/190 is actually the "safe" hard limit for these chips. Tends to perform pretty decently in most cases. PPT is a bit more up for debate I guess. I've seen people running 250 with good results.

I get a few cores tapping 5.05 with the really light boosts, and with 4.65 all core under load I've been fairly happy with my settings. Never seen anything like 4.75 under all core load



But when I have time later I'll do some more voltage reading from stress testing/cinebench under load.

As above my reading is 1.488v is the highest it's gone today, but I've really mostly been doing work and what not. Nothing too heavy during normal use today.

 

[Yaayuh!]

As above my reading is 1.488v is the highest it's gone today, but I've really mostly been doing work and what not. Nothing too heavy during normal use today.

That's VID, what's the VCore?

 

[Joxeon]

this is what my all core clocks look like running cinebench with the above settings.

R23 multi score

 

[Audioboxer]

That's VID, what's the VCore?

This?



Anyway, I'll get some proper results in an hour or two from Cinebench testing.

 

[harmattan]

Late to the party, but I've just acquired a 5800x. Detail below, but in summary, this chip seems to hate vcore above 1.3v in temps, but undervolts under 1.3v well with a moderate overclock.

At Stock
This chip runs volcano hot at stock clocks and voltage. I see 58c idle with temps butting up to 90c in benches within a minute (and throttling). Vcore is what you'd expect at stock at ~1.1v on all cores and @1.45v on a single core when boosted. It's a veritable furnace in games after a few minutes.

Undervolting
This is where things get interesting... I tried PBO Curve Optimizing (can get to -25 stable everything else stock), but hardly any effect on temps. I then started lowering peak vcore (in Ryzen Master, then BIOS). With vcore at 1.18v, I can overclock to 4.2ghz on all cores stable with temps at 45c idle/67c load; At 1.2375v, I can get to 4400 with 51c idle/75c load; At 1.28, I can do 4.6ghz with 53c idle/79c load; Finally, at 1.3, I can get to 4.7ghz stable, but temps start climbing to the mid 80s load.

Anyone else seeing similar results? My impression -- from this single example, granted -- is this chip's stock vcore is far too aggressive.

I'm using my trust Alpenfohn Matterhorn with AC-4 paste (seated and re-seated to ensure good contact) -- not the beefiest of coolers, but should handle this CPU. Case is a Corsair Obisidan 650D with great flow. Cooling shouldn't be the issue. I do have an Adata XPG Levante 240 AIO on its way, but not expecting miracles.