[Help] Power-draw figures [SOLVED]

[Dinnercore]

TLDR: Inflated numbers from watt-meter device, finally found the culprit in an OCP/OVP protection power strip.

EDIT [SOLVED]:
The wattage measurement was off, because of an OCP strip that I used. This one, and only this one and not the other OCP and OVP strips I have mess up with the measurement. It got inflated by over 2x.

I have tested my current daily rig on full GPU load on both GPUs and saw 1210W from that wonky strip.
When I switched over to a different power strip I suddenly see 545W. Which is the correct value (The 2080 Super reports 114W draw with my undervolt and the 3090 reports 275W draw with my undervolt). Since my CPU is currently undervolted for max. efficiency as well this all adds up perfectly.


[Original post below]

Aye ye fine lads and lasses!

I got something started when I joined our WCG-team and decided I finally need to know what my monster machine is pulling from the wall and my wallet each week.
So I got myself a power-meter, this one to be precise, and wanted to see what I get at idle, 100% CPU-load and while benchmarking the Vega 64.
Specs of the meter thingy:
Measured range current: 0.01 - 16 A
Measuring accuracy current: +/-1% or +/-0.01 A
Measuring range power: 0.2 - 3600 W
Measuring accuracy power: +/-1% or +/-0.2 W

The numbers I got are high, but I kinda expected this. Then again I looked at what my parts are supposed to draw and that got me thinking.
Idle 243W // CPU 100% load 643W // Running Heaven benchmark 910W.


I know a Threadripper (1950X) build with Vega is power hungry, but what got me thinking is, at 643W CPU-load my Vega was in idle and that would mean the CPU alone gobbles 400W+ total. This is at 1.25V and 3.675GHz. A slight OC and not 4+GHz suicide runs. The Vega is under water but I run it in my CS:GO profile which is stock power limit! I can´t believe a single stock Vega can make a 1000W PSU sweat. Sure someone like @buildzoid might see that, but not with stock power limit?!
I can rule out that the additional power has to do with the cooling system, because the pumps are a seperate plug that is currently in a completly different phase and the fan speeds are fixed by a fan controller and do not change between idle or load. This system has no HDD, no optical drive. Just an M.2 SSD, the mobo, GPU and CPU.

So I got to think my tool of measurement must read bs. Which is why I plugged it with my 1800x system. It features the same voltage and clock (e.g. it is basically half of the TR), aio watercooling. RM850i PSU from Corsair. And the power meter reads exactly the same as the PSUs build-in stuff does. 105W idle power draw and 200W total system draw on 100% CPU load. That is with multiple HDDs, an optical drive and a fan controller with multiple 140mm fans + the two fans from the rad.
After that I now think that either my little monster has a hunger that starts to frighten me (if one day I stop posting pls send help) or that something messes up the figures. Either some build in protection or filter from the PSU or god knows what.

My question would be, to anyone who knows how power meters work: How do these devices work, and what could possibly upset it?

 

[Vya Domus]

16 cores use a lot of power, even at idle. Especially if overclocked. Vega 64 is also very power hungry.

That being said 900W does seem exaggerated.

 

[king of swag187]

I'd think that 800-850 @ max load would be reasonable, perhaps you have a lot of accessories?

 

[londiste]

1950X is specced to 180W at 3.4GHz. On the good old Summit Ridge this is already in a not very nice part of the efficiency curve.
Stock Vega runs at 300W and is pretty well stuck to that.

When measuring from the wall, you also need to account for the PSU efficiency.
Over 100W is very common but 243W idle sounds like an awful lot. What other devices do you have in that machine?

Btw, what does software monitoring report for power figures?

 

[phill]

@Dinnercore I'd run it stock and see what the difference is, if it's already high then you know it's not a problem. If that's not the case, then you can look into what the overclock is doing when you bump up the vcore, I noticed it with the 5960X (a few less cores but...) add some vcore and the wattage did increase at idle and load (I don't use the C1 states etc) I'll see if I can dig out my results for you just to give an example

 

[Vayra86]

Not sure if these are such incredible numbers. I mean, consider you actually have two 8c/16t Ryzen CPUs in there and the most power hungry consumer GPU on the planet right now, next to the incoming 2080ti that is.

 

[AsRock]

Sorry cannot really help much but by overclocking the CPU maybe disabling the power saving for the CPU\Mobo ?.

Alll so reminds me of a ASUS 775 mobo i had which would run the system at idle 200w were as other mobo's would be down around 90w.

 

[Dinnercore]

I think I need to clear some things up, to point out what I´m wondering about:
Specs you can find under my avatar!
Accessories or other devices, yes there are many fans in my case, controlled by 2 fan controllers. That will explain the high idle, which btw is not what bothers me!

Other then that I have no HDD, only 1 SSD, no USB devices other then a keyboard and mouse. WLAN and any wireless connections are off via bios.

I know my PC can pull those numbers if pushed to the max. but think about that:
Idle power 240W, CPU load 640W! My Vega is in idle at this moment, not consuming any more then before. That is a 400W increase ONLY on the CPU. My PSU is 80+ Titanium with 230V input, so 95-96% efficiency. If I account for that loss, then my CPU would suddenly pull 380W more then it already is at idle. How on earth would this be coolable if true? My temps sit in the 40s!

Software is telling me this about GPU and CPU @ idle:

So if those are somewhat correct, I´m looking after losses at 100W for CPU and GPU, and the other 140W must be my fans and the SSD....
See why I question this?

And for load, I have a screen from a slightly higher Vcore setting:


My meter was showing 700+ W from the wall here.

 

[londiste]

Idle numbers actually check out well enough. 100W from motherboard is not usual but with the HPDT motherboards is not unheard of.
Load does not make sense though. 240W on the CPU at load that the CPU reports itself sounds about right.
Power delivery losses cannot be that high, can they?

 

[kapone32]

That definitely sounds high. Especially under water I also have a Vega 64 and I don't ever see the system pull no more than 650W under load As for the Vega card I see power at 223W when gaming. I do have a R5 2600 instead of 1950 but I also have 2 HDD, 2 SSHDs, 2 SSDs, 4 NVME drives and 10 140MM fans and 3 200MM fans in my system. I don't think there is anyway that it should be showing that much power draw do you have your meter set for 110 or 230?

 

[DeathtoGnomes]

overclocking the 1950 will draw double or even triple the watts, im sure you can search for specifics. the vega cards are easily undervolted while keeping frames with speed increase to offset the undervolt.

 

[Vya Domus]

640W for CPU alone is incredibly high, not even the new 2990X eats up that much power, granted this number is for whole system draw + PSU.

 

[Dinnercore]

@Dinnercore I'd run it stock and see what the difference is, [...]

I´ve done that now (see below), since I agree with you all that OC does bring so many variables into this that it´s hard to tell what´s going on. But my sense for basic physics is telling me that something is very wrong. 900W power draw (aka mostly heat loss) will show in temps, be it on the VRM, PSU or CPU/GPU. I can´t isolate the part and that has me worried. I still highly believe the power meter reads out wrong, but in case it does not (since it works fine on my 1800x system!) there is something in my system that has a high power loss and the only thing I can not monitor temp wise is the PSU... If that thing has a failure I would rather like to know it.

Not sure if these are such incredible numbers. I mean, consider you actually have two 8c/16t Ryzen CPUs in there and the most power hungry consumer GPU on the planet right now, next to the incoming 2080ti that is.

I do partly agree, the numbers are not crazy but my Vega is running in stock power limit and with slightly reduced voltage, the CPU is not on 4GHz all core, but far below with plenty of headroom. I have not heard of a 1950x that blows up 1000W PSUs yet. That sounds more like a 7980XE.
I do have effectivly 2 1800x in there, yes. But my 1800x takes 200W full load for the whole system, if I double that up I´m still at only 400W.

overclocking the 1950 will draw double or even triple the watts, im sure you can search for specifics. the vega cards are easily undervolted while keeping frames with speed increase to offset the undervolt.

I put it to default settings now, see the lower half of this post. And triple the power? Maybe if you let it rip on 1.45V, I´m running 1,21V.

So I went for stock settings on the board and CPU. Just to see what changes:

This is my current GPU profile:

Now I took 3 measurements. One at idle, nothing running, one at CPU 75% threads on 100% and one with CPU load running + Heaven benchmark.

Results:
Idle 202W // CPU load 591W // max out 1187W


This is on stock bios settings, 180W TDP on the CPU. I can not really believe these load figures. And I can not see that power turning into heat anywhere in my system, only the PSU does get hot quickly, but I have no sensor to test it and my hand sadly does not come with a build-in thermometer.

EDIT: @kapone32 the power meter is for 230 input.

 

[kapone32]

I´ve done that now (see below), since I agree with you all that OC does bring so many variables into this that it´s hard to tell what´s going on. But my sense for basic physics is telling me that something is very wrong. 900W power draw (aka mostly heat loss) will show in temps, be it on the VRM, PSU or CPU/GPU. I can´t isolate the part and that has me worried. I still highly believe the power meter reads out wrong, but in case it does not (since it works fine on my 1800x system!) there is something in my system that has a high power loss and the only thing I can not monitor temp wise is the PSU... If that thing has a failure I would rather like to know it.



I do partly agree, the numbers are not crazy but my Vega is running in stock power limit and with slightly reduced voltage, the CPU is not on 4GHz all core, but far below with plenty of headroom. I have not heard of a 1950x that blows up 1000W PSUs yet. That sounds more like a 7980XE.
I do have effectivly 2 1800x in there, yes. But my 1800x takes 200W full load for the whole system, if I double that up I´m still at only 400W.



I put it to default settings now, see the lower half of this post. And triple the power? Maybe if you let it rip on 1.45V, I´m running 1,21V.

So I went for stock settings on the board and CPU. Just to see what changes:

View attachment 107677View attachment 107678 This is my current GPU profile: View attachment 107679 View attachment 107680

Now I took 3 measurements. One at idle, nothing running, one at CPU 75% threads on 100% and one with CPU load running + Heaven benchmark.

Results:
Idle 202W // CPU load 591W // max out 1187W
View attachment 107681 View attachment 107682 View attachment 107683

This is on stock bios settings, 180W TDP on the CPU. I can not really believe these load figures. And I can not see that power turning into heat anywhere in my system, only the PSU does get hot quickly, but I have no sensor to test it and my hand sadly does not come with a build-in thermometer.

EDIT: @kapone32 the power meter is for 230 input.

What is the wattage of your PSU? Unless you have a 1600W PSU your PC would automatically shutdown at the draw of 1187 W

 

[newtekie1]

What is the wattage of your PSU? Unless you have a 1600W PSU your PC would automatically shutdown at the draw of 1187 W

He in fact does have a 1600w PSU.

The fact that the PSU is getting hot quickly kind of leads me to believe these numbers are accurate. A 1600w PSU shouldn't be getting hot unless the load is 800w or more.

 

[Vya Domus]

1187W at stock ? So it went up ? Yeah something is not right, either the meter is wrong or there is something wrong with the PSU.

 

[cadaveca]

Not sure how many 1950X users there are here, but I am one. guess how much mine pulls at idle with a GTX1080? NOT THAT MUCH!!!

At idle, from the wall, I get 81W, with 3 SSDs, and two mechanical, as well as an external mechanical USB drive. I also have a USB-power lamp, my keyboard and mouse, and my headphones and XBONE controller. This is on aircooling, with a Artic Freezer33TR, using two 1.2A fans. Nearly everything is RGB, too. Memory is at 2933 MHz.

running Timespy, max power draw recorded is 351W. This is the highest I can get on stock, since my ASRock X399M Taichi does what it is supposed to do. CPU-only load is actually far less, around 208W.

Since your system specs list a modified ASUS Strix X399 board, I'll say that the board is garbage, and you should get a new one. At stock it should hold that 180W for CPU maximum power draw, and drop clocks as low as needed to maintain that. I can do hard AVX loading, and see my CPU clocks drop to 1600 MHz on all cores as it maintains that 180W figure. I think your board is the culprit here, and somehow it is doubling the power used, which should not be possible at stock settings. It is possible there is something wrong with that PSU as well, but my own 1950X is working exactly as you'd expect. I can OC everything when I put the CPU under water and get higher power use, but nothing near 1000W.

 

[DeathtoGnomes]

I put it to default settings now, see the lower half of this post. And triple the power? Maybe if you let it rip on 1.45V, I´m running 1,21V.

thats still part of overclocking.

 

[Dinnercore]

Since your system specs list a modified ASUS Strix X399 board, I'll say that the board is garbage, and you should get a new one.

I´m thinking about the board being an issue. Still, the VRM sensor never shows more then 35°C (the point I set the fan to turn on). Could be broken, the 'modified' relates to the changes I made on the VRM-Cooling. I replaced the tiny fan with one that can move more air and removed any aesthetic garbage that blocks airflow on the VRM-heatsink and around the DDR modules.
I wrote it like that as a reminder that the standard version of this board is no longer comparable to my temperature figures. I did not yet e-Power it

 

[Vya Domus]

Unlikely that the VRMs really do run at just 35c.

 

[cadaveca]

I´m thinking about the board being an issue. Still, the VRM sensor never shows more then 35°C (the point I set the fan to turn on). Could be broken, the 'modified' relates to the changes I made on the VRM-Cooling. I replaced the tiny fan with one that can move more air and removed any aesthetic garbage that blocks airflow on the VRM-heatsink and around the DDR modules.
I wrote it like that as a reminder that the standard version of this board is no longer comparable to my temperature figures. I did not yet e-Power it

OK, well thanks for that information as that does help suss out what's going on here. Is there any chance you have a clamp-on current meter, or can access one? That would really help sort this, since you can measure actual VRM power use via that. Then you could pinpoint either the board or PSU being the problem.

Ther can be no doubt that these AMD TR chips can pull a tonne of power, and you can get 500W of power draw quite easily, but that is also not going to happen (well, it is not supposed to) over a single 8-pin connector. Do you have a supplementary 4-pin plug connected to the board's VRM? The 8-pin should not allow for more than ~374 W of power draw, by design. It could be that the system is drawing a lot, but the PSU isn't capable of shutting down parts of itself, leading to the increased draw at idle.

I'm on a 120V power plug here in Canada, but what the systems pulls in this situation should be fairly identical for us, and even then, @ 220V, you should see better efficiency than I do.

 

[Caring1]

Have you got the function set to KW/h?
If It is set to Watts, it is cumulative and adds up the longer you run it.

 

[Dinnercore]

[...] Is there any chance you have a clamp-on current meter, or can access one? [...]
Do you have a supplementary 4-pin plug connected to the board's VRM? [...]
I'm on a 120V power plug here in Canada, but what the systems pulls in this situation should be fairly identical for us, and even then, @ 220V, you should see better efficiency than I do.

No clamp yet, and I do not know anyone I could lend one from. In the end I might just buy one, if there is something wrong I want to know it.
And yes I do have that additional 4-Pin plugged in.

Have you got the function set to KW/h?
If It is set to Watts, it is cumulative and adds up the longer you run it.

I get why you think that, but you are wrong in 2 ways. One, this is not the calculated average per hour that builds up! It is the current Watt figure. It drops back to 200 after the load test. And 2nd, if what you think is true, then the accumulation to 1187W after 5 minutes would hint at a power consumption of 14.244 W! My mains fuse would say no to that already around the 3KW mark (each room has a limit of 3KW, only kitchen can pull more).

 

[phill]

I´ve done that now (see below), since I agree with you all that OC does bring so many variables into this that it´s hard to tell what´s going on. But my sense for basic physics is telling me that something is very wrong. 900W power draw (aka mostly heat loss) will show in temps, be it on the VRM, PSU or CPU/GPU. I can´t isolate the part and that has me worried. I still highly believe the power meter reads out wrong, but in case it does not (since it works fine on my 1800x system!) there is something in my system that has a high power loss and the only thing I can not monitor temp wise is the PSU... If that thing has a failure I would rather like to know it.

I do partly agree, the numbers are not crazy but my Vega is running in stock power limit and with slightly reduced voltage, the CPU is not on 4GHz all core, but far below with plenty of headroom. I have not heard of a 1950x that blows up 1000W PSUs yet. That sounds more like a 7980XE.
I do have effectivly 2 1800x in there, yes. But my 1800x takes 200W full load for the whole system, if I double that up I´m still at only 400W.

I put it to default settings now, see the lower half of this post. And triple the power? Maybe if you let it rip on 1.45V, I´m running 1,21V.

So I went for stock settings on the board and CPU. Just to see what changes:

View attachment 107677View attachment 107678 This is my current GPU profile: View attachment 107679 View attachment 107680

Now I took 3 measurements. One at idle, nothing running, one at CPU 75% threads on 100% and one with CPU load running + Heaven benchmark.

Results:
Idle 202W // CPU load 591W // max out 1187W
View attachment 107681 View attachment 107682 View attachment 107683

This is on stock bios settings, 180W TDP on the CPU. I can not really believe these load figures. And I can not see that power turning into heat anywhere in my system, only the PSU does get hot quickly, but I have no sensor to test it and my hand sadly does not come with a build-in thermometer.

EDIT: @kapone32 the power meter is for 230 input.

I'm unsure on the specifics of a 1950X but I do think that the voltage on the VCore is too high.. 1.5?? I would reset everything in the bios and start again.. Backup everything on your drives BEFORE you do this just in case something has a fit, but I'm thinking that's too high for a start.. This is possibly were the CPU is chewing up so many watts Give that a go and let us know your results I'll see if the guy I know can send me a few pics of his 1950X under CPU load and go from there...

 

[Dinnercore]

I'm unsure on the specifics of a 1950X but I do think that the voltage on the VCore is too high.. 1.5?? I would reset everything in the bios and start again.. Backup everything on your drives BEFORE you do this just in case something has a fit, but I'm thinking that's too high for a start.. This is possibly were the CPU is chewing up so many watts Give that a go and let us know your results I'll see if the guy I know can send me a few pics of his 1950X under CPU load and go from there...

Nah those numbers are normal for Ryzen CPUs, it´s the voltage they apply for very short bursts to boost up to 4.2GHz. This is why I don´t like running stock, it does stay within 180W TDP, but jumps up and down with clocks and voltages every splitsecond. While 2 or 4 cores are @ 1.5V, the others are much lower and then they switch, this is why HWMonitor records 1.5V on all cores as this was the max. they reached, but not at the same time.

The high voltages are more often applied in idle, when it does not hit the power limit, under load I see ~1.2V on the non boosting cores.

But I would like to hear from you what your friend/colleague/guy said. Every input from other 1950x owners is helpful.