Low TDP and vRel on a 3080 only while 3D rendering (48.5% TDP) - Issue

[Fuchs]

Been noticing that my GPU has started to have really poor performance when rendering (I use Redshift 3D). A scene that usually takes 1 min to render takes now up to 4 minutes.
 

After digging around, noted that my 12v rails going to the gpu are at 12.3 some, and even 12.4v in some cases and the card hitting vRel at half the TDP. Resetting usually fixes the issue, but it is still something that didn't happen before.
 

When rendering at full performance, it goes to 95% tdp and renders at the usual speed.
 

This is the card when hits vRel. Sometimes it also hits VOp but that is not what concerns me. This state lowers the rendering performance the most.

 

And here, it is hitting all vRel, VOp but also Pwr, meaning the card is also reaching it's actual power limit and going as full performance as it can. (Even tho it could be better without vrel or vop)

 

I've compared these metrics with a friends 3080 in his system, and his is as close at 12v as it can, and his card power consumption is also lower. I've tested my card in his system and performs normally while consuming the normal amount of 12v via the PCIe psu cables.

AFAIK vRel tells if voltage is reliable or not to boost, but it doesn't seem right to be happening at less than 50% TDP. I haven't been able to recreate the issue in any other software so far and have been exploring a lot of threads everywhere with not a single match.

Sometimes it also hits vOp and vRel at the same time, again, at less than 50% TDP.

 

[voskobiinyk]

I have same issue( on my Lenovo laptop. I describe this problem here https://redshift.maxon.net/topic/42682/tdp-problem?_=1654290829424 but doesn't have any answer( Pls add some comments to that topic to up this problem. 

 

Do you have any solution?

 

[voskobiinyk]

i am just reistall nvidia driver to 497.29-notebook-win10-win11-64bit-international-dch-whql and works fine now

[Aragorn-]

vrel and vop arent errors. Its just telling you whats limiting the GPU core clocks. The way the boost algorithm works, the card will clock its self up until it reaches some limit. More clock speed needs more voltage, so sometimes (quite often infact) it simply reaches the maximum allowed voltage and thus cant clock any faster. In other cases, it will reach a power limit first and thus power is the limiting factor, not voltage. It could also be limited by thermals. The exact workload the card is doing will depend on which limit kicks in first. Some workloads are power hungry, and so you'll be pwr limited. Other workloads are not power hungry, and you end up voltage limited instead.

 

In your first example, your nowhere near the power limit, as your only running 53% TDP. But the frequency is tapped out because the GPU has reached the max voltage limit. 

 

None of this explains your problem.

[Fuchs]

On 6/7/2022 at 8:09 AM, Aragorn- said:

None of this explains your problem.

Card is running at 150w when it should be running at well over 300w.

 

Anyway it seems to be a driver issue. Had to roll back to 471.

[Fuchs]

On 6/7/2022 at 4:20 AM, voskobiinyk said:

i am just reistall nvidia driver to 497.29-notebook-win10-win11-64bit-international-dch-whql and works fine now

I had to roll back to 471 and haven't had the issue since. Came here because nobody at RS had any clue besides "Oh it's just the drivers", but I had similar problems back in 2020, and again, is something that only and just only redshift triggers.

[Aragorn-]

5 hours ago, Fuchs said:

Card is running at 150w when it should be running at well over 300w.

 

Anyway it seems to be a driver issue. Had to roll back to 471.

well exactly, my point was you were fixating on power limits, and it had nothing to do with those, instead some software issue was causing the card not to be fully utilised.

 

I thought nvidia cards never have driver issues, thats an AMD thing.... right? 

[voskobiinyk]

https://redshift.maxon.net/topic/42190/super-slow-redshift-ipr-and-rendering/59 here is discussing problem with Maxon Devs 

 

"I've just repro this for an Nvidia meeting, the latest drivers slow down renders considerably especially if you're rendering on the display card
We'll see what they can come up with"

 

To fix a problem at the latest drivers, you need to decrease a memory limitation at Redshift settings to 60% and waiting to a better solution 

 

[Fuchs]

On 6/9/2022 at 5:41 AM, voskobiinyk said:

https://redshift.maxon.net/topic/42190/super-slow-redshift-ipr-and-rendering/59 here is discussing problem with Maxon Devs 

 

"I've just repro this for an Nvidia meeting, the latest drivers slow down renders considerably especially if you're rendering on the display card
We'll see what they can come up with"

 

To fix a problem at the latest drivers, you need to decrease a memory limitation at Redshift settings to 60% and waiting to a better solution 

 

Repro with driver 472.

Also the thread you linked talks about abnormal Vram usage.

 

 

On 6/9/2022 at 5:16 AM, Aragorn- said:

well exactly, my point was you were fixating on power limits, and it had nothing to do with those, instead some software issue was causing the card not to be fully utilised.

 

I thought nvidia cards never have driver issues, thats an AMD thing.... right? 

Not fixating in power limits, not even close. Power limits are just a way to proof the card is throttling itself for some reason. Saying "It is just slower" isn't enough. 

[Fuchs]

If anyone has come across this lately. Yep, it is a vram reporting error by the nvidia driver and got it on a screenshot:


Geometry using over 140gb (It shouldn't take more than a gb)

Interesting how Nvidia releases info about games updates in their studio drivers.