PerfCap Reason PWR / GPU load 99% (4 x 3090 RTX) - Workstation Rendering issues

[aims_fx]

Hello all! I recently updated my PC setup and been having some performance issues!
First up I'm a VFX / 3D animator working from home so this PC setup isn't designed for gaming, but I assume its the same load / stress and rendering absolutely destroys a computer.

I recently got a custom loop and 4 x 3090s and been seeing some performance issues on Octane Render in Cinema 4D. I opened GPU-Z noticed some things:

GPU load is running at 99%
and the PerfCap Reason Pwr.

I attached my computer specs in this post and they are also present in my profile.

So I installed EVGA Precision, lowered the power target of all 4 GPUs to 60% to see if that had any impact on the PWR consumption and it didn't change one bit. I noticed the noise of the GPUs went down considerably however.
I tried under voltage with MSI afterburner but that didn't seem to do anything.

I've attached a screenshot of GPU-Z mid render, this is just showcasing one of my RTX's, but the others all have similar read-outs.

Please bear in mind, computer hardware specs and testing is relatively new to me, so please have that in mind!

Unfortunately most VFX artists I know don't know a thing about this sort of stuff.

Any help would be greatly appreciated!

- Aims

 

[elghinnarisa]

PerfCap is just a nice little notification telling you that "if I had more X, I could probably increase clock a bit". It is not telling you to is limited below what its supposed to be. It is not the cause of your performance issue if the card is running at 99% load, which it is. It is also running above its base frequency so its boosting as it should.

For the performance problems, look in to the software you are using, because at least in that picture the cards doing just fine.

 

[Chomiq]

Well if you're seeing 99% load with PWR limit trigger then yeah, that's normal for rendering workload or gaming at full blast. If you lower power limit to 60% it will mean that your card will be hitting the power limit earlier.

When it comes to undervolting:

Obvious question - are you hitting the Apply button in Afterburner after doing the undervolt?

@Mussels has undervolted his 3090 he might help guide you.

Run card at default, use "Reset" function in GPU-Z to monitor GPU frequency during render. See your average frequency after render is finished. Use that frequency as the target frequency for undervolt.

TLDR
Hitting power limit at 99% GPU load is normal.

 

[GerKNG]

at around 75% of its actual powerlimit it says limited by power.
could be one of the tons of problems people get with more than 2-way SLI (3 and 4-way SLI is basically not supported by NVidia at all and gets no attention with drivers, fixes and general support)

 

[Chomiq]

at around 75% of its actual powerlimit it says limited by power.
could be one of the tons of problems people get with more than 2-way SLI (3 and 4-way SLI is basically not supported by NVidia at all and gets no attention with drivers, fixes and general support)

There's no SLI bridge on that photo.

 

[Calenhad]

Your GPUs are running at their power limit and maxed out load. I would say that is working as intended. Workstation tasks are different than gaming so you won't see the same boost behaviour.

 

[aims_fx]

Well if you're seeing 99% load with PWR limit trigger then yeah, that's normal for rendering workload or gaming at full blast. If you lower power limit to 60% it will mean that your card will be hitting the power limit earlier.

When it comes to undervolting:

Obvious question - are you hitting the Apply button in Afterburner after doing the undervolt?

@Mussels has undervolted his 3090 he might help guide you.

Run card at default, use "Reset" function in GPU-Z to monitor GPU frequency during render. See your average frequency after render is finished. Use that frequency as the target frequency for undervolt.

TLDR
Hitting power limit at 99% GPU load is normal.

Greetings! I'll try this, would I be trying to aim at the GPU clock MHz or Memory Clock MHz for the targetfrequency?


I only got worried that once one of my renders finished, the computer shut down immediately. I looked into troubleshooting and a few people on the OTOY forums suggested that it was to do with my PSU not having enough power.

There's no SLI bridge on that photo.

Correct. I had the computer put together without the option as as Octane Render recommends disabling the SLI option in your NVIDIA control panel to maximize OctaneRender’s rendering performance.

Also whilst I have you guys here, I'm getting VRAM temps of around 100-106°c on my GPUs whilst rendering. Just wondering if that is normal considering I have a custom loop. A fellow artist said that its not normal but I wanted to get another opinion. Have in mind I don't have any active backplates (do you think its worth me getting some installed?).

 

[Chomiq]

Dunno man, it's 2 kW PSU and you're running 4 GPU's that theoretically can pull up to 450 W each (granted, in Furmark). Get some power meter to measure draw at the wall, see what's the max wattage your system is pulling when rendering.

Pugit Systems ran a 10-core Xeon-W with 4 3090's and they measured a peak of 1717 W power:

Quad GeForce RTX 3090 tested on a single PC running workstation benchmarks

Here at Puget Systems, our customers range from those needing relatively modest workstations, all the way up to those that need the most powerful system possible. These...

www.techspot.com

So, will we be offering quad RTX 3090 workstations? Outside of some very custom configurations, we likely are not going to offer this kind of setup to the general public due to the power draw concerns

 

[OneMoar]

custom loop and the VRAM is hitting 90c?
yikes I know GDDR6X runs toasty but that's up there lol
if anything its probably the GPU drivers 4 way SLI is NOT a supported config even if they really aren't in SLI ( did you install the SLI bridges?)
edit according to the board's page two slots operate at PCIE4 16x and two operate a PCIE4 8x
you may actually be saturating the pcie bus but check and make sure above4g and re-sizable bar is enabled in the bios

 

[aims_fx]

custom loop and the VRAM is hitting 90c?
yikes I know GDDR6X runs toasty but that's up there lol
if anything its probably the GPU drivers 4 way SLI is NOT a supported config even if they really aren't in SLI ( did you install the SLI bridges?)
edit according to the board's page two slots operate at PCIE4 16x and two operate a PCIE4 8x
you may actually be saturating the pcie bus but check and make sure above4g and re-sizable bar is enabled in the bios

Yeah that's what I thought. Before the custom loop I had 2 x 3090s and I was hitting 110 so when I started researching into options to cool it down, I decided to go for a custom loop and whilst I was at it add 2 more 3090s to the build. I just wasn't sure seeing as there is now 4 GPUs that the 100-106 temp was normal for that setup. I guess not. Someone did say bringing in some active backplate cooling will cut the temps down a lot though, so I am considering that.
I'll check out my bios and report back.

 

[looniam]

first impression(s)
firstly sweet loop - esp if your first.

it seems those blocks don't have active back plates for the Vram on the back. some blocks do and there are backplates sold separately (link is example not opinion).

that dual 120mm exhaust looks nice for an addition 280mm rad for that. - that 480 on the front is fine but more gives wiggle room if rendering for hours and hours (and hours).

but yeah acitve cooling on the backplates is often nessary for 3090s. or at least that my understanding.

again sweet rig.

E: and yrah that D5 pump will more than handle another rad - which is (usually - aside from angled fittings) the lowest restriction in any loop.

 

[aims_fx]

first impression(s)
firstly sweet loop - esp if your first.

it seems those blocks don't have active back plates for the Vram on the back. some blocks do and there are backplates sold separately (link is example not opinion).

that dual 120mm exhaust looks nice for an addition 280mm rad for that. - that 480 on the front is fine but more gives wiggle room if rendering for hours and hours (and hours).

but yeah acitve cooling on the backplates is often nessary for 3090s. or at least that my understanding.

again sweet rig.

E: and yrah that D5 pump will more than handle another rad - which is (usually - aside from angled fittings) the lowest restriction in any loop.

Thanks mate! Put my life savings into this one to help my work load. I'll look into active back plate cooling but also noticed just now my GPUs are flashing red lights. Saw on some forums that means they are starved of power and its a bad thing. Is this right? video uploaded to vimeo:

 

[OneMoar]

you can't daisy chain power cables like that on a setup like this
all gpus must have TWO Dedicate lines from the psu

this kind of setup usually requires dual power supplies due to most psus not having sufficient connectors to power 4 way setup

that means a minimum of 8 cables two per card and you can daisy chain the 3d 8 pin

 

[looniam]

Thanks mate! Put my life savings into this one to help my work load. I'll look into active back plate cooling but also noticed just now my GPUs are flashing red lights. Saw on some forums that means they are starved of power and its a bad thing. Is this right? video uploaded to vimeo:

as far as i can tell those are ~same as the gpu-z readout. did a quick search before you edited for the video - i caught that.

ran across a youtube where the guy claims two psus cleared it up - but looking at the comments- mostly *just him* saw a difference. that @K is enough, esp if its not tripping out, but out of the two cables you DO have to each card - its usually better to have one cable for the first connector and the second cable split between 2/3.

refer too the image in your first post :

just mind the power balancing.

 

[aims_fx]

as far as i can tell those are ~same as the gpu-z readout. did a quick search before you edited for the video - i caught that.

ran across a youtube where the guy claims two psus cleared it up - but looking at the comments- mostly *just him* saw a difference. that @K is enough, esp if its not tripping out, but out of the two cables you DO have to each card - its usually better to have one cable for the first connector and the second cable split between 2/3.

refer too the image in your first post :
View attachment 247634
just mind the power balancing.

Yeah I did look at getting 2 x PSUs but most people assured me the 2000w was more than enough for 4 x 3090s. I had this setup put together for me by a local computer shop and I feel they really messed up. I'm not sure if its daisy chained, does this photo show that?

Also how do I balance the power on the pins? (I am total newb, but been trying to learn!). Looked on other forums and the voltage of the pins should all be around 12v - so seeing as there is a mass variation in voltage, maybe thats what is causing the warning lights to go off?

 

[OneMoar]

indeed he's only hitting 75% TDP this could be power or it could be the render load but I would fix the power, either way, overdrawing the connectors can lead to melted connectors and dead gpus

 

[aims_fx]

indeed he's only hitting 75% TDP this could be power or it could be the render load but I would fix the power, either way, overdrawing the connectors can lead to melted connectors and dead gpus

What would your solution be? Getting a second PSU, limiting the power output of the GPUs?

 

[OneMoar]

What would your solution be? Getting a second PSU, limiting the power output of the GPUs?

if that psu in your specs is right you just need to plug more cables in that psu has 9 PCIE connectors make sure each card gets two dedicated cables
and I would get that backplate for the vram as well

 

[looniam]

What would your solution be? Getting a second PSU, limiting the power output of the GPUs?

if you were running 4 dell/OEM 3090s you would not need to worry.

you have the potential with those asus strix to greatly exceed that 2K psu but do you really need it? esp not running compute heavy loads?

just mind what you're doing.-

just swap the two cable runs going to the gpus. - leaving pin 2 alone and swapping 1 for 3. you might not need to tear apart that sweet cable management.

pulll all the #3s out. move 1/2 to 2/3. plug all previous #3s to #1.
E: this is assuming like the gpu-z image that all #1s are the "power hogs"

 

[thesmokingman]

At first glance, the cooling is way too low, not enough pump nor rad and then the psu. You need a ridonkulous amount of radiator if you wanna work at a high level and not drive yourself crazy from the noise. You're water temp must be real high...

 

[1100R]

I would also make sure that the cooling circuit is working properly. This meter is cheap and gives peace of mind.

Bykski Real Time Oled Digital Display For Custom Liquid Water Cooling Flow Meter,flowmeter Alarm Multi Function B-tfc-cs-x - Fans & Cooling - AliExpress

 

[OneMoar]

if you were running 4 dell/OEM 3090s you would not need to worry.
View attachment 247637

you have the potential with those asus strix to greatly exceed that 2K psu but do you really need it? esp not running compute heavy loads?

just mind what you're doing.-

just swap the two cable runs going to the gpus. - leaving pin 2 alone and swapping 1 for 3. you might not need to tear apart that sweet cable management.

pulll all the #3s out. move 1/2 to 2/3. plug all previous #3s to #1.
E: this is assuming like the gpu-z image that all #1s are the "power hogs"

I don't think he would exceed the psu rating at least not at stock but hes close at 1900w theoretical but either way hes in the relm of needing duel psus

this build strikes me as a zero research fuck it maximum everything speaking of that

 

[thesmokingman]

I don't think he would exceed the psu rating at least not at stock but hes close at 1900w theoretical but either way hes in the relm of needing duel psus

this build strikes me as a zero research fuck it maximum everything

Yea this.

The rig below is only dual 2080ti and a 3970x and it can sustain 1300w with water temps around mid 30s. OP's rig is so under supported it must spend its life throttled hard. These machines have to be able to run flat out for 5hr-8hr stretches. You need stupid cooling to maintain keep it from heatsoaking. Also, 128mb is not enough with quads.

Build: 3970x, dual 2080ti, 8TB m.2 RAID = Render Monster

I'm building a render machine for work on a variety of applications. It will be an investment in reducing completion times in rendering and content creation workloads. This is a build for an old college buddy who works in the industry. On to the component list. Also, going to start with two...

hardforum.com

 

[eidairaman1]

This is just a base line, this is not accounting for boost speeds of the cpu or gpu.

AMD Ryzen Threadripper 3960X NVIDIA GeForce RTX 3090 - PSU Calculator - Build 0WpkH1

AMD Ryzen Threadripper 3960X NVIDIA GeForce RTX 3090 - Power Supply Calculator Build 0WpkH1

outervision.com

Get a gpu-z screenshot of all 4 GPUs

 

[OneMoar]

Yea this.

The rig below is only dual 2080ti and a 3970x and it can sustain 1300w with water temps around mid 30s. OP's rig is so under supported it must spend its life throttled hard. These machines have to be able to run flat out for 5hr-8hr stretches. You need stupid cooling to maintain keep it from heatsoaking. Also, 128mb is not enough with quads.

Build: 3970x, dual 2080ti, 8TB m.2 RAID = Render Monster

I'm building a render machine for work on a variety of applications. It will be an investment in reducing completion times in rendering and content creation workloads. This is a build for an old college buddy who works in the industry. On to the component list. Also, going to start with two...

hardforum.com

theoretically if it renders fast enough it won't spend enough time at load to heat soak lol