ryzen Ryzen voltage questions

[Distelzombie]

Hi

 

Everyone knows Ryzens power saving and turbo features clock the core up and down all the time. But it also changes voltage on-the-fly.

I have a Ryzen 5 2600 and a Gigabyte B450M DS4H mainboard.

 

For me it changes like this:

 

Ryzen Master: 0.89375 idle, to 1.2375 V under load. (under "Voltage Control" the CPU voltage) I can see spikes to 1.55V when idling. I guess that happens when Windows does something and the core clocks suddenly higher. Seems very high.

 

HWiNFO64 v6: The minimum and maximum of "core #x VID" are 0.863 and 1.300 respectively. Of "CPU Core Voltage (SVI2 TFN)" they are 0.875 tp 1.306V. Of "Vcore" on the mainboard they are more or less identical.

 

CPU-Z: Core Voltage 0.864 to 1.296V

 

 

When I overclock my CPU to 4GHz I only need to change the "Dynamic Vcore" setting in the BIOS to an offset of +0.138V, which results in an unchanging voltage of 1.1375V in Ryzen Master (unchanging as in equal to non-OC voltage, when the turbo and power saving features are off),

a "Core #x VID" of 0.4 to 1.138V and a "CPU Core Voltage (SVI2 TFN)" of 1.1 to 1.137V in HWiNFO,

and a "Core Voltage" of max. 1.272V in CPU-Z

(I am a little uncertain as to what voltage I am actually changing with this offset, since a change to it isn't visible in Ryzen Master, but changing it has an effect)

As you see, the peak voltages with my 4GHz overclock are lower than the default settings of Ryzen. Also the weird 1.55V peak in Ryzen Master is gone, since that value doesn't change anymore.

 

Having the voltages somewhat fixed (some are still changing fsr) will definitely prevent the CPU overvolting itself due to it's automatisms.

 

 

 

TL:DR with 4GHz overclock my peak voltages are lower.

 

Is this better for the CPU than leaving it default? Or: What is more damaging: Hertz or Voltage?

 

 

This is a very technical question, but I hope someone can answer it with educated certainty. XD

Thank you!

[BRONXTECH]

I wouldn't worry to much, I am a Ryzen GURU and finally got my Ryzen 1600 overclocked stable with 10 Cinebench 15 Runs, as well as 10 hours of stability AIDA64 Testing @ 1.425 Volts. It sometimes spikes to 1.5 but thats common and not to worry to much, If it starts hitting over 1.6 then that will not kill your cpu as it will just decrease the average life expectancy, "SUPPOSEDLY".

[Distelzombie]

But it spikes to 1.55V (Ryzen Master only, not visible in HWiNFO) with default settings. The OC is way less demanding and practically does not spike)

My question was what is less damaging.

 

Thank you

[Jurrunio]

Voltage hurts more than hertz (because hertz doesnt hurt at all), but the milliseconds of 1.5v+ shouldnt hurt anything. It is something for AMD to fix though.

[Distelzombie]

So... in this case overclocking the CPU will improve it's lifespan/-time? (As little as it is )

Its why I asked, but it sounds so counterintuitive.

[Distelzombie]

Well...

I am doing power consumption and heat measurements. (Switched to CPUID Hardware Monitor) I'm using prime95 Small FFTs.

looks like my overclock causes the CPU to draw roughly 40-50 Amps MORE than stock, but watts are only 10 more for Core and 28 for Package. Crazy! Why suddenly so much current?

From stock to Overclock:

CPU Amps: 91-98A to 140A (very fluctuating in default settings)

Core Watts: 41.5W to 55.8W

Package Watts: 80W to 114W

Temp CPU (delta over 15°C): 41°C to 57°C

Temp VRM (delta over 15°C): 41°C to 62°C

 

So... there has to be something that limits the CPU power draw with default settings. But I can't set this when it's overclocked. It's an invisible thing.

Ryzen Master says there are three limits when BIOS settings are default:

"Sustained Current Limit CPU" (TDC) of 60A or 114A, I don't understand why there are two,

"Peak Current Limit CPU" (EDC) of 90A or 168A,

and "Total Socket Power CPU" (PPT) of 87W or 1000W. (Wth?)

PPT is reaching 91% of 87W under load (no OC), TDC reaches 100% of 60A and EDC reaches 81% of 90A

There is only one Current sensor and it says 91-98A at the same time.

I can't see the limits in the top row when overclocked, but they are visible under Control Mode.

Why does everything have to be so uncertain? How does it even work?

 

I can set cTDP in my BIOS. By default it is set to Auto. Changing that to 50 does not have any effect on my CPU overclock. Nor does 25. What is that?

 

I tried setting the clock 3.4GHz, effectively just removing the "auto" that's there by default. That changed nothing of the behavior.

If Set above 3.4GHz the turbo is off and the voltages are stabilized. Power draw is 100A with 3.425GHz. (Mostly no fluctuations.) If set to 3.7GHz with same voltage I can see a current draw of 107A. With 3.8GHz it's 115A and with 3.9GHz the limit of short-time stability is reached and it crashes immediately. (All with stock voltage of 1.128)

That shows that for some reason just increasing the clock increases the power draw substantially.

 

Why does it suddenly draw up to 50A more??

 

I don't think, in light of this, that it is actually any good to run it at 4GHz. I only have a Gigabyte B450M DS3H - though it does have "4+3 Hybrid Digital PWM Design", so it's not the worst.

What do you think?

Edited January 29, 2019 by Distelzombie

[nick name]

1.55V is par for the course.  That's the CPU running single core max frequency briefly.  Sustained single core loads at max speed don't use that much voltage.  

 

If you want to reign that in then load line calibration with a negative offset is usually most effective.  I recommend the second highest level of LLC for CPU Core and on a Ryzen 2700X a negative offset of around -.075V for less demanding workloads and -.04V for all workloads.  This will definitely vary for you, but hopefully it will serve as a start.  

 

PPT, TDC, and EDC are what you would increase to push Precision Boost Overdrive further.  

 

The default multiplier for a 2700X is 37 and setting any multiplier equal to or less than that won't actually do anything to the multiplier.  The CPU will behave as it normally would.  As soon as you set the multiplier higher than 37 the CPU will behave as if it's overclocked.  I imagine the 2600 has a default multiplier and will behave the same if you set the multiplier at or below its default value.  

 

If you do set an overclock multiplier or set a static voltage then the CPU behaves accordingly (as an overclock) and that means it won't go through all its p-states and won't run through its full range of speeds.  So you won't see the same spikes in voltage or speed.  

 

Also, Gamers Nexus has a good video exploring Ryzen and its power needs.  They've found the curve doesn't grow linearly, but exponentially.  The larger jump in power as speed increases is just how the CPU behaves.  So power isn't just single steps up for each increment of speed added, but multiple steps of power for each step of speed.  Temps will follow accordingly.

https://www.gamersnexus.net/guides/3290-exponential-ryzen-voltage-frequency-curve

[Distelzombie]

Ok. I can't set Precision Boost Overdrive on my 2600. The default multiplier is 34, single-core turbo goes to 4GHz. All-core turbo seems to be inexistant.

 

I expected the power draw only to rise when I change the Voltage. That's why this is such a big thing for me.

I also wonder about the extreme current jump of +50A, while Watt don't change as much. This must be a result of the in average lower voltage. Package Watt only increased by 34. In my opinion Watt should increase more than ampere.

 

Gamers Nexus didn't write anything about current or Watt. So this is interesting, but irrelevant right now. (Even more so since I have a 2600 that behaves way different to their 2700X. EDIT: In that it needs less voltage to reach 4GHz)

 

 

I guess the only questions left are:

 

-Why an increase in current when only the clock is changed?

 

-Does 140A of current to the socket kill my mainboard?

(With Cinebench, which is more of a real-world load, it only reaches about 120A)

 

-Why does the current lower itself to 130A after 5 minutes of Small FFT prime95?

(I even deactivated Cool&Quiet in BIOS. VRM and CPU temps are 75°C. It reaches these temps 3 minutes before it throttles the current, so they are not the culprit. Clock stays at 4GHz all the time, only current throttles.)

 

 

Thank you very much guys

[nick name]

Different workloads require different power loads.  I believe it's similar to horsepower and torque.  Honestly, I would do a quick Google on amperage and wattage versus voltage.  It will explain it way better than I ever could.  

 

A quick explainer from me would be:  Those high spikes in voltage from idle indicate the CPU doing extremely light workloads.  So the CPU needs the higher voltage to reach speed, but the workload isn't enough to require higher amperage.  

 

And as far as the CPU reducing power is that it's how it's designed to operate.  The longer it works the hotter it gets so it scales back performance.  A true overclock would make it hold its speed, but a stock CPU is gonna run as fast as it can as long as it's within FIT parameters and then begin to scale itself back.  

[Distelzombie]

Sorry if I was unclear.

I do electronic stuff for a hobby. I know how voltage and current relate. But I was dumbfounded when just increasing clock increases current by such a huge amount.

 

II meatnt those last three question in relation to the 4GHz overclock and not a stock CPU. So it does throttle current when overclocked and when all the power saving functiins are deactivated. It should not throttle. Not because of temperature.  Because, as I said, it stayed at 75°C for a while before it throttled. And the throttle doesn't even lower temperatures by more than 1 degree.

I have no idea why it would do that. The safe temperature is set to 85 degrees.

 

[Distelzombie]

Here, a graph of the throttling issue. (Set to one second interval)

 

The Chassis Fan blows directly at the VRM. In this case the VRM didn't even reach 75°C, but maxed at 73°C, CPU at about 70°. (Max. from statistics tab)

Clearly not thermal throttling. It can't be power throttling either because the package was running at 140A/114W for a while.

Something either got hot and there is no connected sensor for it, or something else decided to get lazy. Please mind that this only happens with Prime95, as far I know.

 

My 4Ghz overclock is stable at 1.260 Vcore, BTW. (I lowered it from the 1.272 V I wrote in the original post)

Anyway, I would let it stay at 4Ghz, but I'm unsure about the peak power draw of ~114W. (1.26Vcore x ~140A)

 

Can someone please give me advice? I don't know if that is  toomuch or if it is within spec. (Gigabyte B450M DS3H) I have no references.

Thank you again

 

 

EDIT: Oh I forgot: I don't have a all-core turbo with stock settings. Any idea why? Seems like a power limit is reached (Ryzen Master says so), but I can't change that, and it is supposed to have a limited-by-power/thermals all-core turbo, afaik. But the limit shouldn't be this low, should it?

[Distelzombie]

On 1/28/2019 at 7:00 PM, Jurrunio said:

Voltage hurts more than hertz (because hertz doesnt hurt at all), but the milliseconds of 1.5v+ shouldnt hurt anything. It is something for AMD to fix though.

I wonder how that fits in with my observations. Just raising the clock speed to 3.8GHz increased ampere by 15 to 115A. (Compared to 3.425GHz (100A) with static default voltage.)

Because 3.7GHz with the same voltage only draws 107A, I would argue that if we follow the curve and if it would be stable at these voltages, with 3.9GHz current would be about ~125A and at 4GHz with the same voltage ~138A.

So basically the only reason why the power draw is higher is clock speed. I'm unsure why

 

Years ago when I last overclocked something the only thing raising the temperature was voltage.

[Jurrunio]

2 hours ago, Distelzombie said:

I wonder how that fits in with my observations. Just raising the clock speed to 3.8GHz increased ampere by 15 to 115A. (Compared to 3.425GHz (100A) with static default voltage.)

Because 3.7GHz with the same voltage only draws 107A, I would argue that if we follow the curve and if it would be stable at these voltages, with 3.9GHz current would be about ~125A and at 4GHz with the same voltage ~138A.

So basically the only reason why the power draw is higher is clock speed. I'm unsure why

 

Years ago when I last overclocked something the only thing raising the temperature was voltage.

because voltage readings from software is wrong and you're actually using more voltage? Unless you use one of those flagship boards, treat the software voltage reading as merely a suggestion of voltage used.

[Distelzombie]

On 2/1/2019 at 3:25 PM, Jurrunio said:

because voltage readings from software is wrong and you're actually using more voltage? Unless you use one of those flagship boards, treat the software voltage reading as merely a suggestion of voltage used.

Yes, ok. But since the voltage reading in HardwareMonitor didn't change at all between the changes of 3.4GHz and 3.8GHz, the real voltage hasn't changed either (That's how things work) - might it be higher or lower in reality than what it displays is irrelevant in this case: because we're only looking for a change. And there was none. The software would definitely show a change if there was one.

If it wouldn't be a huge hassle to remove the board from the case again and run it outside with access to the back of the socket, I would make accurate voltage readings.

 

So my observations stands: Increasing clock speed alone increases power consumption significantly.

[Jurrunio]

1 hour ago, Distelzombie said:

Yes, ok. But since the voltage reading in HardwareMonitor didn't change at all between the changes of 3.4GHz and 3.8GHz, the real voltage hasn't changed either (That's how things work) - might it be higher or lower in reality than what it displays is irrelevant in this case: because we're only looking for a change. And there was none. The software would definitely show a change if there was one.

If it wouldn't be a huge hassle to remove the board from the case again and run it outside with access to the back of the socket, I would make accurate voltage readings.

 

So my observations stands: Increasing clock speed alone increases power consumption significantly.

if you actually watched the video you wont be saying this

[Distelzombie]

1 hour ago, Jurrunio said:

if you actually watched the video you wont be saying this

Sorry I don't think you understood me correctly, please elaborate when you see the real voltage change in the video without also seeing the voltage the sensors read change.

 

The sensors do read the real voltage. They are essentially 12 to 14bit analog-to-digital converter (That's why there is a certain granularity to the measurements) that sit on the vcore rail. What the actual reading is is irrelevant when you only care for a change.

They always show the change, by nature, but the reading might be of by +-0.2.

Of course, at the two extremes the A-to-D converter reaches it's limit and doesn't show a change anymore. For these applications it might be 0V to 2V, divided by 512 or 2048 steps. (12 or 14 bit)

 

Anyway. That's not in the video, right?