GPU miner pulls 40% more power after 20 minutes

[fk1]

Hi guys, I just got a call by my mining partner. We are running 6 of the following sertup in a warehouse atm:

- asus z270 prime
- celeron g3900
- 4gb ddr4 ram
- 8x sapphire nitros rx580
- hp dps1200 server psu
- 160w pico psu

he just installed new wires for power to place the miners at a different spot in his warehouse. now he is telling me that on the new place when the miners are just started up, the power consumption is about 3.5A at 230V wich should be around 800W. After 20 minutes, this climbs to 4,8A or 1100W. I told him this is very unlikely but he claims that he has professional measurement equipment which should be ok.

I only can check the gpu power consumption per software by remote which is around 85W per card and doesnt change. all fans are running at 100% so when miner gets warm that shouldnt make such a big difference. The only thing I could imagine is that these server power supplies with breakout board are pulling close to their 1200w max after some time regardless if the system needs it or not.

what is happening here please?

[1713483990]

1713483990

[1713483990]

1713483990

[leonix007]

Hi guys, I just got a call by my mining partner. We are running 6 of the following sertup in a warehouse atm:

- asus z270 prime
- celeron g3900
- 4gb ddr4 ram
- 8x sapphire nitros rx580
- hp dps1200 server psu
- 160w pico psu

he just installed new wires for power to place the miners at a different spot in his warehouse. now he is telling me that on the new place when the miners are just started up, the power consumption is about 3.5A at 230V wich should be around 800W. After 20 minutes, this climbs to 4,8A or 1100W. I told him this is very unlikely but he claims that he has professional measurement equipment which should be ok.

I only can check the gpu power consumption per software by remote which is around 85W per card and doesnt change. all fans are running at 100% so when miner gets warm that shouldnt make such a big difference. The only thing I could imagine is that these server power supplies with breakout board are pulling close to their 1200w max after some time regardless if the system needs it or not.

what is happening here please?

It seems its a normal consumption depending on the algo you are mining

for eth, default RX580 would go around 200watts per card

not unless you do modding and underclocking

which I bet you already did

I only can check the gpu power consumption per software by remote which is around 85W per card and doesnt change

Power consumption per software is different compared to power draw at the wall

https://forum.ethereum.org/discussion/13772/rx-580-power-consumption

[philipma1957]

Do you use windows 10?

Sometimes  it drops  all power settings and just draws top power.

If you use  smos  ie simplemining  you can clock  from any pc in the world and the setting tends to be stable power wise.

[fk1]

I use rxOC(linux)

[philipma1957]

I use rxOC(linux)

I have used nvOC

not sure what  rxOC  is stabilitywise.

but  the only way to test is with another  os.

your voltage/power creep  seems wrong. 

I did have it happen often with win 7 and msi afterburner  or win 10 and msi afterburner.

the msi afterburner settings would drop and boom  full power used on the rigs.

[fk1]

but yours sounds like all of a sudden while my example is growing slowly...
nvmd, i think we will test with simplemining

[brunocrypto]

Maybe your overclock setting are too high, and the card reset the parameters, putting the "underclock and underpower" off on your nvidia rigs ?

That's will lead to a upper power consommation..

You need to check your hashrate at lunch, and after the power augment change, to see if there is a change.

[fk1]

yeah that might be the case since i did an individual bios mod. I will investigate the power consumpton behaviour onone rig. ifconsumption jumps high all of a sudden that might be a good indicator.

ill also have a look with ohgodatool if consumption stats might differ from rocm-smi stats. after all i guess i will need to focus on off-the-wall no matter what...

[fk1]

So I had some more investigation on this: first of all, my rigs are in a closed 19" server case 4U. I have installed 3 of these babies in front for pulling air straight into the front where the 8 gpus are installed:

http://www.scythe-eu.com/produkte/luefter/ultra-kaze-120.html

according to my measurement equipment, 3 of these pull 60 watts otw in total.

when I boot up win10 for bios editing, in idle the system takes 80w with one card installed.

when booting up rxOC(ubuntu), system uses about 115w when idle. now the strange part is no matter what settings or BiOS mods I do, I am not able to get the card running with less then 120w. I can sometimes see in rocm-smi that the cards pulls less but off the wall thats simply not true.

I always observe the same when fireing up claymore mining tool: wattage climbs from 115 to lets say 170-180w. Then comes a second where measuring show 0w, then the next second I always have >200w off the wall. somehow I really think this cards is resetting every time. and I even observe this with stock BIOS.

I have read so much about modding this card and people are reporting 75-85w consumption per card. the only thing i am not sure about now is if there was ever somebody who measured/calculated with off the wall results, because right now I believe one cant take gpu-z or other software emasurements serious...

[WhyMe]

On RX, if we are talking about ETH, 75-85w per gpu is a software measure. In real, ETH takes 120-140w per gpu ...

[fk1]

so hunting for this is a waste of time? weird thing. I just had a following test:

mine with one card, take measuremtn which was 215w
now add a second card -> measures 335w
difference is the second card as is 120w. rocm-smi software measurement: 95w

I have a 8x1060 rig where the nvidia-smi is telling that one gpu consumes 75w which is accurate off the wall 680w. I regret them atis so much

[fk1]

Guys

I accepted that software measurement doesnt equal off the wall measurement when it comes to amd gpus although with my nvidia rig this seems to be the case.

Since I reach 1100w off the wall and my power supply is HP DPS1200 which should be 80Plus Platinum I wondered if the PSU isnt a little bit too less dimensioned and if I should get other PSUs. Maybe power consumption will also be lower since PSUs should tend to be more efficient when not working close to max?! lmkwyt

[melpheos]

Guys

I accepted that software measurement doesnt equal off the wall measurement when it comes to amd gpus although with my nvidia rig this seems to be the case.

Since I reach 1100w off the wall and my power supply is HP DPS1200 which should be 80Plus Platinum I wondered if the PSU isnt a little bit too less dimensioned and if I should get other PSUs. Maybe power consumption will also be lower since PSUs should tend to be more efficient when not working close to max?! lmkwyt

For max efficiency you shoud add a second PSU but at 90% usage you are still at around 92% efficiency vs 94% at 50% usage. Is the 2% gain on electricty usage really worth the added PSU ? Only your electricity cost and calculation over a year will tell you @ .15c/kw you will save only 27$ so it's not worth it.
However, if your electricity cost is higher and have several rigs it would be worth the additional cost

[blacktux88]

sound like the fan speed spinns up and because of that the power goes up too.
we had also this problem

 

[fk1]

fan speeds are at 100% straight from boot

[lordvek]

Is the power supply (from the electro company I mean) stable in the area?