GPU power dipping when stressing CPU

[khesax]

Hello, I noticed that when stress testing my GPU and CPU, the power on my GPU varies a lot compared to the CPU, but only happens when the CPU is also being stressed, if not, the power seems to be pretty stable, is this normal? I don't have crashes or anything I just noticed this when trying my new case fans and was wondering if I should be concerned or not.

I used Cinebench to stress test the CPU and Furmark to stress test the GPU, my CPU is a i5 11400F with power limits unlocked and my GPU is a RTX 3070, my PSU is a Bronze rated EVGA 700W BR.

 

[OneMoar]

brauh thats called being cpu limited

 

[khesax]

brauh thats called being cpu limited

How? Does Furmark also use CPU when stress testing the GPU?

 

[OneMoar]

the CPU is also being stressed, = not enough cpu available to render stuff rendering anything requires cpu doi

 

[dark2099]

You also have to realize with the design of the boost being both power, and temp limited, what you're seeing is likely a result of it jumping up and down the boost table as power or temps trigger their respective limiters. Then when not under load, since the power and temps should stabilize, which you are seeing. I am wondering too if you're asking just a bit much out of your PSU too. As the recommended is 650w, and you're only 50w over that, means during stress testing, which tends to use the most power, you're seeing the PSU struggling a bit.

 

[looniam]

You also have to realize with the design of the boost being both power, and temp limited, what you're seeing is likely a result of it jumping up and down the boost table as power or temps trigger their respective limiters. Then when not under load, since the power and temps should stabilize, which you are seeing. I am wondering too if you're asking just a bit much out of your PSU too. As the recommended is 650w, and you're only 50w over that, means during stress testing, which tends to use the most power, you're seeing the PSU struggling a bit.

you are on, what i think, is the right trail but . . recommended 650 watts for what? and by who?

you don't need to answer that, its the skeptical thoughts in my head.

but i believe the output wattage is fine but the transients on that budget HEC (oem) unit is lacking and can cause a hiccup like the OP described.

i really think the corsair CX series at the same or slightly less wattage would have been a better choice. that locked intel part (maxPL is 125 not 250) and 3070 will going to max at ~475 watts.

 

[OneMoar]

simply put he is trying to run a CPU and GPU stress test at the same time

of course thats going to result in weird behavior the power reading has nothing todo with it neither does his power supply thats not what those sensors are for those are reporting what the gpu is drawing

the gpu does not know or care what the psu is rated at and the psu does not know or care what the system load it will run right until the OCP kicks in and turns the system off



700W is plenty the maximum drawn from a 3070 is 220 - 250w(+15% if you have messed with the power limit slider)

11400F even with the PL limits completely removed is going to have a tough time getting to 150W at stock even under avx load

Rest the system does not draw 280W no way no how

he is seeing expected behavior stressing the GPU requires CPU power to feed the GPU all that data its crunching this is computers 101

 

[looniam]

he is seeing expected behavior stressing the GPU requires CPU power to feed the GPU all that data its crunching this is computers 101

and an old group regulated psu will exaggerate the behavior.

 

[OneMoar]

and an old group regulated psu will exaggerate the behavior.

explain that
the power limits are strictly on the component side not on the psu there is no interaction between the two the psu will run right until it it trips OCP/ Or fails
if the psu can not deliver the required power they system doesn't throttle or anything of the sort. it just turns off POOF thats it good night sweet prince

power supply's are NOT smart they do not have any idea what is going on downstream of the Lowside of the rectifiers outside of a OCP/failure condition

except in the case of particularly awful units vdroop doesn't have an appreciable effect on power consumption and there is no case where vdroop could lead to a run away condition, even if it did OCP is going to kick on and turn the power supply OFF (again note nothing todo with the motherboard cpu or gpu OCP in this context is strictly on the PSU)

this thread is done op had a non problem problem no point in continuing this further

 

[looniam]

explain that
the power limits are strictly on the component side not on the psu there is no interaction between the two the psu will run right until it it trips OCP/ Or fails
if the psu can not deliver the required power they system doesn't throttle or anything of the sort. it just turns off POOF thats it good night sweet prince

power supply's are NOT smart they do not have any idea what is going on downstream of the Lowside of the rectifiers outside of a OCP/failure condition

except in the case of particularly awful units vdroop doesn't have an appreciable effect on power consumption and there is no case where vdroop could lead to a run away condition, even if it did OCP is going to kick on and turn the power supply OFF (again note nothing todo with the motherboard cpu or gpu OCP in this context is strictly on the PSU)

this thread is done op had a non problem problem no point in continuing this further

what are you talking about? dear god man periods (.) would help.

psus now a days are much smarter than you give then credit for. most any high end psu will have monitopring ICs that monitors OCP, OVP and temps. so a psu will shut down from more than over currant.

older group regulated psus are notorious for not keeping up when an instantaneous high load is reacquired.

source: hundreds of psu reviews and tutorials.

 

[W1zzard]

First of all, this is completely normal and not unexpected.

Adding to what others said, this could also be an artifact of the measurement process.

Power = Current x Voltage. Both have to be measured and multiplied in real-time.

If the CPU draws a lot of current, the voltage on +12V from the PSU will drop, introducing noise for the measurement. If that happens quickly, the (GPU) power math will be slightly off.

Is your FPS constant at that time or does it fluctuate the same way as power?

 

[dark2099]

you are on, what i think, is the right trail but . . recommended 650 watts for what? and by who?

you don't need to answer that, its the skeptical thoughts in my head.

but i believe the output wattage is fine but the transients on that budget HEC (oem) unit is lacking and can cause a hiccup like the OP described.

i really think the corsair CX series at the same or slightly less wattage would have been a better choice. that locked intel part (maxPL is 125 not 250) and 3070 will going to max at ~475 watts.

650w I think I pulled for Nvidia's site, or maybe eVGA.

 

[looniam]

650w I think I pulled for Nvidia's site, or maybe eVGA.

fine print:

4 - Requirement is made based on PC configured with an Intel Core i9-10900K processor. A lower power rating may work depending on system configuration.

even unlimited/unlocked i5 (non K) will draw half that amount; 125 watts for the non K 6 and 4 core cpus for intel.
source: