Ryzen voltage very low under load

[Gothmoth88]

so my brother has an asus crosshair hero 6 board with a ryzen 1700.
he overclocked it a bit (all cores @3775 mhz).

now we both wonder that after a bios update (6101 to 6903 if it matters) the core voltage took a huge drop.

before he had around 1.32v when running cinebench (as shown by hwinfo, cpu-z).

now he has flashed the latest BIOS and put in the exact same settings (he took BIOS screenshots to make sure of that).

but the voltage is now at 1.12v -1.18 when under load.


the other thing is... the temps are now 4-5 degree celsius higher.
more power draw with lower voltage.

then we upped the LLC to a higher level (well with the old BIOS it was at auto so we can only guess that level 3-4 are higher as before).

but the core voltage as shown by the latest hwinfo or cpu-z stays way below 1.3v.
at LLC level 3 values as seen in the screenshots.


the CPU should not be stable at just 1.12 volt under full load for hours... right?
but we could not get it to crash with cinebench or aida64 torture test.


whats going on here?

 

[LTC8K6]

Sounds kinda' good to me.

Maybe they just fixed an over voltage problem?

 

[elpokor]

Chances are CPU-Z was previously reading SVID=Vcore, and something in the new BIOS is telling CPUz to stop it and read the proper Vcore, which has now confused hwinfo64's SVID who is reading actual Vcore.

As far as I'm concerned, whatever you read in hwinfo64's MoBo sensors as "Vcore" is the closest you'll get to measuring with a multimeter from the back of the socket. Open hwinfo64 in "Sensors Only" mode, click the arrows on the bottom left or scroll until the Motherboard sensors (in between the CPU and GPU ones, sometimes named Nuvotron with some serial numbers), and locate Vcore. It's usually there with VRM PowerIN/PowerOUT , Current In/Current Out or other VRM measurements.

Do not trust readings of SVID or VID as that voltage never gets to the CPU: it's the voltage the CPU is requesting from the motherboard power delivery, not a measurement of the voltage actually reaching the CPU after going through the caps and the socket's own resistance/capacitance.

 

[Markfw]

I trust Ryzen master the most for things like voltage and temperature.

 

[Gothmoth88]

As far as I'm concerned, whatever you read in hwinfo64's MoBo sensors as "Vcore" is the closest you'll get to measuring with a multimeter from the back of the socket. Open hwinfo64 in "Sensors Only" mode, click the arrows on the bottom left or scroll until the Motherboard sensors (in between the CPU and GPU ones, sometimes named Nuvotron with some serial numbers), and locate Vcore. It's usually there with VRM PowerIN/PowerOUT , Current In/Current Out or other VRM measurements.

Do not trust readings of SVID or VID as that voltage never gets to the CPU: it's the voltage the CPU is requesting from the motherboard power delivery, not a measurement of the voltage actually reaching the CPU after going through the caps and the socket's own resistance/capacitance.

the thing is.... whenever you read about overclocking hwinfo or cpu-z is mentioned to check voltages.
i have a multimeter but i am usually to lazy to open the case and measure it myself. my brother won´t do it at all (and he lives a few hundred km away so i can´t test).

im not sure what value these readings have when they are so dependend on whatever the mobo is deciding to report to you.

we are just doing very tame overclocking but when the numbers really can be all over the place as you say.... most overclockers could be in trouble i think.
the board could still deliver 1.3v as before theBIOS update but shows us 1.12v?
so we give it more voltage via offset and in reality it gets 1.45v then?

or the board never delivered 1.3v before but the sensors where reporting 1.3v?

i attached a hwinfo screenshot that shows the numbers under full load.

i have never seen vid voltages that low when someone posted screenshots of an overclocked ryzen.

the CPU CORE VOLTAGE SVI2 TFN sensor is close to the real voltage the CPU get from the mobo?

 

[LTC8K6]

Maybe you just have a very good chip?

 

[Gothmoth88]

Maybe you just have a very good chip?

but why the different numbers with the exact same BIOS settings?

i am no overclocking expert.

i normaly just set the multiplier and push the voltage offset until the system is prime stable.
but i never noticed such a difference in voltages when doing a BIOS update.

that´s why i am asking if this is some wrong voltage reporting or why this happens.

 

[Markfw]

but why the different numbers with the exact same BIOS settings?

i am no overclocking expert.

i normaly just set the multiplier and push the voltage offset until the system is prime stable.
but i never noticed such a difference in voltages when doing a BIOS update.

that´s why i am asking if this is some wrong voltage reporting or why this happens.

OK< I said it once, I will say it again.... USE RYZEN MASTER and see what the voltages are.

 

[LTC8K6]

Version 6201 2018/06/22 7.85 MBytes
CROSSHAIR VI HERO BIOS 6201
Update Agesa Code to 1.0.0.2c
Update compatibility protocol for 3rd party hardware monitoring software
Fixed miscellaneous issues with fan calibration/options
Improve memory compatibility

Maybe that is the BIOS change that affected the readings?

 

[Gothmoth88]

OK< I said it once, I will say it again.... USE RYZEN MASTER and see what the voltages are.

more or less the same voltage. see screenshots.

Maybe that is the BIOS change that affected the readings?

yes of course that could be possible but still isn´t the voltage to low for ryzen 1700?

i have two ryzen systems and have seen a lot of systems from friends. none had such low voltgages.

 

[DrMrLordX]

Asus high-end boards have lots of VRM options. Updating the UEFI might have changed overcurrent/current limits and other things meant for bleeding edge overclocks. You should check some of those settings and see what's going on in there. Also, don't assume that power draw has gone up unless you can confirm that with an external power meter (Kill-a-Watt or similar). Higher temps are suspicious, though.

 

[Gothmoth88]

Asus high-end boards have lots of VRM options. Updating the UEFI might have changed overcurrent/current limits and other things meant for bleeding edge overclocks. You should check some of those settings and see what's going on in there.

as i said we are both no overclocking experts. the values are at default with the old BIOS and the new one.
so "checking" them would be difficult.

the only thing my brother and me usually touch when overclocking is the multiplier, offset and when needed LLC.

when i assume that the voltage reported by ryzen master etc. is correct (around 1.12-1.18V and all cores at 3775 MHz) there should be plenty of room for overclocking right?
we did run prime95 over night and the system showed no issues with the voltages seen in the screenshots.
temps stay put at 64° celsius and the fans could go faster when we change the fan profiles.

 

[DrMrLordX]

It seems like your system is working well enough. If you're happy with the operation of the CPU, then leave it where it is. I was only suggesting that if the voltage dropped but temps went up, and if those readings were accurate, then the logical conclusion is that current limits might have changed between UEFI revisions.

Someday you may wish to peruse some of the vendor-specific parts of the UEFI to see what's in there. Most of what's provided by the vendor is documented, and you can learn more on the Asus forums if you wish. OCN may also have a thread for your board with a lot of board-specific information.

 

[Abwx]

First 2017 batch of 1700X with 8C active :

https://www.hardware.fr/medias/photos_news/00/53/IMG0053411.png

https://www.hardware.fr/articles/939-1/amd-ryzen-7-1700x-ryzen-7-1700-test.html


Overclocking test with default voltage measured in the mobo, stock voltage while running Prime 95 is at 1.155V@3.5GHz :

https://www.hardware.fr/medias/photos_news/00/53/IMG0053444.png

https://www.hardware.fr/articles/939-3/overclocking-pratique.html

We can see that measured voltages (tension lue) are lower than what is set as VID (tension réglée).

 

[Markfw]

more or less the same voltage. see screenshots.



yes of course that could be possible but still isn´t the voltage to low for ryzen 1700?

i have two ryzen systems and have seen a lot of systems from friends. none had such low voltgages.

I see NO Ryzen master screen shots

 

[Gothmoth88]

I see NO Ryzen master screen shots

you dont see idle.jpg and load. jpg attachement?
i guess then something must be wrong with the forum software.
because i see them.

Overclocking test with default voltage measured in the mobo, stock voltage while running Prime 95 is at 1.155V@3.5GHz :

https://www.hardware.fr/medias/photos_news/00/53/IMG0053444.png

https://www.hardware.fr/articles/939-3/overclocking-pratique.html

We can see that measured voltages (tension lue) are lower than what is set as VID (tension réglée).

thx that´s some useful info.

 

[Markfw]

you dont see idle.jpg and load. jpg attachement?
i guess then something must be wrong with the forum software.
because i see them.




thx that´s some useful info.

I see them, but thats not Ryzen master. At least not recognizably so.

 

[Gothmoth88]

I see them, but thats not Ryzen master.

believe me it is. i just did a crop for the important part.

satisfied with this screenshot?

it´s german but it´s still ryzen master.

 

[Markfw]

believe me it is. i just did a crop for the important part.

satisfied with this screenshot?

it´s german but it´s still ryzen master.

Yes ! And it shows you over clocking and the temp, all 3 very important. Looks like you have a great chip for that speed at that vcore.

 

[elpokor]

believe me it is. i just did a crop for the important part.

satisfied with this screenshot?

it´s german but it´s still ryzen master.

that's certainly not the voltage you're running at. I've highlighted the areas of your previous hwinfo64 screenshot to give you an idea of what you're looking at. Whether you trust the on-die sensors or the MoBo VRM sensors is up to you, although from 1,269V to 1,264V there's little to worry about the discrepancy . That's also the reason why on-die measurement of power is slightly higher (110W to 108W). Here's two wiki entries in case you wan't to dig a bit deeper
https://en.wikipedia.org/wiki/Voltage_regulator_module#Voltage_identification
https://en.wikipedia.org/wiki/Low-dropout_regulator

 

[elpokor]

forgot to add the reason why I'm certain you're not running at any voltage lower than 1,264V. Check SiliconLottery's OC stats:
unfortunately, they didn't test AVX2 (Prime95 v29 and up) when Summit Ridge was mainstream but luckily we have those stats for Threadripper's iteration of Summit Ridge. Those are binned CPUs but it serves the purpose of checking how a golden sample of Summit would perform. All of the 1950X tested did 3.8Ghz at 1.250V, which would suggest that your OC of 3.775Ghz at 1.264-1.269V is perfectly reasonable and within what would be expectable. Temps being 64ºC would also concur that voltage is lower than 1.3 (70's ºC) but higher than 1,15V (high 50's ºC). Your CPU power is 110W, and given that the 1700X with it's base clock of 3,4Ghz draws 95W seems reasonable the 25W increase (+15%) for the 375Mhz (+11%) gained.

On the flipside, you could be the first person on earth to have a R7 1700 running at near 3,8Ghz under 1,2V if you trust Ryzen Master; but I'd be worried about temps and power consumption as they should be in the high 50's and low 90's respectively. Choice is only yours, I guess, but I know where I'd put my money on.

Moving on, if you trust my considerations and want to push the OC further, chances are you have enough room to get Vcore and frequency a little bit higher. First try to stabilize 3,8Ghz under 1,300V (as per hwinfo64 readings). Then increase frequency to 3,9Ghz and increase your voltage in 25mV steps (+0,025V) until you reach stability. Hopefully you'll get it stable under 1,4V. I'm inclined to think so given the relatively low voltage you need for (almost) 3,8Ghz but the last steps of the voltage-frequency curve get pretty rough. Here's The Stilt's volt-freq chart: last 100Mhz could take 0,125V easily, while the previous 100Mhz step only takes 0,050V. And by the way, the voltage it shows for 3,8Ghz is awfully close to your 1,269V: another hint in the same direction.

Good luck!

 

[Abwx]

forgot to add the reason why I'm certain you're not running at any voltage lower than 1,264V. Check SiliconLottery's OC stats:View attachment 5689
unfortunately, they didn't test AVX2 (Prime95 v29 and up) when Summit Ridge was mainstream but luckily we have those stats for Threadripper's iteration of Summit Ridge. Those are binned CPUs but it serves the purpose of checking how a golden sample of Summit would perform. All of the 1950X tested did 3.8Ghz at 1.250V, which would suggest that your OC of 3.775Ghz at 1.264-1.269V is perfectly reasonable and within what would be expectable. Temps being 64ºC would also concur that voltage is lower than 1.3 (70's ºC) but higher than 1,15V (high 50's ºC). Your CPU power is 110W, and given that the 1700X with it's base clock of 3,4Ghz draws 95W seems reasonable the 25W increase (+15%) for the 375Mhz (+11%) gained.

On the flipside, you could be the first person on earth to have a R7 1700 running at near 3,8Ghz under 1,2V if you trust Ryzen Master; but I'd be worried about temps and power consumption as they should be in the high 50's and low 90's respectively. Choice is only yours, I guess, but I know where I'd put my money on.

Moving on, if you trust my considerations and want to push the OC further, chances are you have enough room to get Vcore and frequency a little bit higher. First try to stabilize 3,8Ghz under 1,300V (as per hwinfo64 readings). Then increase frequency to 3,9Ghz and increase your voltage in 25mV steps (+0,025V) until you reach stability. Hopefully you'll get it stable under 1,4V. I'm inclined to think so given the relatively low voltage you need for (almost) 3,8Ghz but the last steps of the voltage-frequency curve get pretty rough. Here's The Stilt's volt-freq chart: last 100Mhz could take 0,125V easily, while the previous 100Mhz step only takes 0,050V. And by the way, the voltage it shows for 3,8Ghz is awfully close to your 1,269V: another hint in the same direction.

Good luck!

Methi nk that you missed the part where measurements made with a multimeter on the mobo are compared to the values displayed on CPU Z among others.

https://www.hardware.fr/medias/photos_news/00/53/IMG0053444.png

Besides the curve you posted say 1.2V@3.7GHz and that s only for a single sample, the one owned by The Stilt...

 

[Gothmoth88]

good infos guys.

now i am not to concerned anymore that something is wrong with my brothers mainboard (my main concern).

again, the thing that puzzled us was that before we needed 1.3x volt (as shown by CPU-Z, hwinfo) to get the 1700 stable at around 3775 mhz.
now with the same voltage offset but the new bios, the voltage displayed in all tools (cpu-z, ryzen master, hwinfo) dropped but the CPU still runs prime stable.

unfortunately we have not written down the hwinfo voltages measured from the VRM or on-die sensors.
how i understand it now is, it could be that just the VID values changed with the new bios but the actual voltage going into the CPU was around 1.26xV before too.

thx for the replys!

 

[elpokor]

Methi nk that you missed the part where measurements made with a multimeter on the mobo are compared to the values displayed on CPU Z among others.

https://www.hardware.fr/medias/photos_news/00/53/IMG0053444.png

Besides the curve you posted say 1.2V@3.7GHz and that s only for a single sample, the one owned by The Stilt...

Me think you missed the part where I actually said "whether you trust the VRM sensors or the on-die dLDO is up to you". Use common sense: two voltages reported by different sensors in different parts of the system are reporting the same value within 5mV. Do you really think a correct measure with a voltmeter (soldering wires from the back of the socket) will be far off from those voltages? If you stick one probe into a capacitor leg and the other probe in the nearest stand-off screwhole chances are you'll get a more inaccurate reading than via software. Sure, you can't trust every sensor on every budget motherboard but when two of them "agree" (one from mobo, one from CPU) it's pretty obvious why. Hey, they might actually be reading Vcore!

Besides, there's a reason you have "The Stilt" memory settings available on ASUS boards. He was part of the mobo engineering team with Elmor. I think they both left recently, anyway... My point is that curve is not "The Stilt's Curve", it's the curve he plotted for ASUS during pre-production. They had to have at least dozens of samples for board qualification, and that's why it's incredibly accurate to most of the first gen dies and why Silicon Lottery (which for a fact they said pretested hundreds of CPUs) have binning results that pretty much paint the same last steps of the volt-freq curve. There's also a Pinnacle Ridge edition of "The Stilt's Volt-freq Curve" which has also been proven accurate over time.

I'm going to miss his review so much when Zen3000 launches... His username seems deleted from this forum .

 

[DrMrLordX]

His username seems deleted from this forum .

He left not so long ago. I think he's still active over on OCN.