RTX 2080 TI 'leveling out' at 80% usage... Stumped.

[sxstack]

Hello All! 

 

I have a RTX 2080Ti that I bought a few years ago. I have been getting quite poor performance out of it - 2080 like performance. I am getting well below the average of 14k with a score of 10.4k. I just noticed that its not using 100% power when running 3dmark time spy so I believe this may be the issue.

 

I have uploaded the two pictures for you too take a look at. I am completely stumped. 

 

PC Specs

Corsair HX1000 PSu

ASUS STRIX OC RTX 2080TI 

Gigabyte Z490 master mobo

Samsung 970 EVO SSD

G.skill Trident Z rgb 3200mhz ram

10700k @5ghz all core 

 

 

I tried:

1. Reinstalling drivers completely with DDU
2. reflashing the bios of the card (it has two so it was lower risk)
3. putting the card in my gf's pc, DDUing the drivers, and running it again... and the behavior persisted. 
4. Replacing power supply cables.
5. Switching the 12V rails.

 

I have not tried a new power supply.

 

Thanks for taking the time to help me out, I really appreciate it. I can get the logs of a Timespy run if you feel like that would help.

***EDIT: For clarity, the two humps in the below HWINFO graphs are the timespy run and the sensors were reset at the start.

Edited March 23, 2022 by sxstack
Clarity

[dizmo]

You can't have an X570 motherboard and a 10700k.

Some cards just don't perform as well as others. Someone might have a fix, but that's something to keep in mind.

What model Asus card?

[Jurrunio]

What is the sustained GPU power when it performed like a 2080?

[sxstack]

10 minutes ago, dizmo said:

You can't have an X570 motherboard and a 10700k.

Some cards just don't perform as well as others. Someone might have a fix, but that's something to keep in mind.

What model Asus card?

Thanks for the reply!

My bad - Just built my friends PC with an x570 so it was still on the mind.

 

I edited that in as well as clarifying that it is a Asus Strix OC 2080ti.

[sxstack]

8 minutes ago, Jurrunio said:

What is the sustained GPU power when it performed like a 2080?

Thanks for the reply! Sustained GPU power in furmark and 3dMark is 80% as shown in the photos. It peaks to about the 260W mark (TDP), then gradually declines to 80% while still hitting Performance Limit - Power.

[An0maly_76]

Is VSync or any other stabilization assistance enabled? Lower frame rate could very well do that.

[sxstack]

2 minutes ago, An0maly_76 said:

Is VSync or any other stabilization assistance enabled? Lower frame rate could very well do that.

Vsync and Gsync are turned off. Prefer maximum performance is on.

 

See snip.

[ApolloX75]

I was seeing some similar results to this with my 2080 Ti a while back. Out of curiosity when was the last time you updated your motherboard BIOS? If I'm not mistaken, you're using a Z490 Aorus Master, correct? I was comparing (I think) your benches against mine on 3d Mark. There's a bunch of BIOS updates, particularly F8B and F20 that have GPU performance enhancements listed in the revision notes.

 

Might be worth a shot. I know one of the issues I had with my card was my Z590 A Pro not running the PCIE link at the proper speeds.

 

Have you tried running Afterburner and raising the Power Limit of the card? Yours seems to cap out at 260W and my Zotac will punch above 320W easy in Time Spy.

[sxstack]

1 minute ago, ApolloX75 said:

I was seeing some similar results to this with my 2080 Ti a while back. Out of curiosity when was the last time you updated your motherboard BIOS? If I'm not mistaken, you're using a Z490 Aorus Master, correct? I was comparing (I think) your benches against mine on 3d Mark. There's a bunch of BIOS updates, particularly F8B and F20 that have GPU performance enhancements listed in the revision notes.

 

Might be worth a shot. I know one of the issues I had with my card was my Z590 A Pro not running the PCIE link at the proper speeds.

 

Have you tried running Afterburner and raising the Power Limit of the card? Yours seems to cap out at 260W and my Zotac will punch above 320W easy in Time Spy.

Yes! I am using the z490 Aorus Master. I am currently on the F20d bios. I will try updating to F21. 

 

If I raise the power limit in Afterburner, the initial spike goes up to the 325W (125%), but then still settles at 200W (~80%). 

[An0maly_76]

1 minute ago, sxstack said:

Yes! I am using the z490 Aorus Master. I am currently on the F20d bios. I will try updating to F21. 

 

If I raise the power limit in Afterburner, the initial spike goes up to the 325W (125%), but then still settles at 200W (~80%). 

Weak capacitors bleeding off power and limiting speed? Just a thought.

[ApolloX75]

8 minutes ago, sxstack said:

Yes! I am using the z490 Aorus Master. I am currently on the F20d bios. I will try updating to F21. 

 

If I raise the power limit in Afterburner, the initial spike goes up to the 325W (125%), but then still settles at 200W (~80%). 

Interesting. Does HWInfo show you the Memory Tjunction temps and Hot Spot temps for your card? When I was benching my 3080 after reinstalling the air cooler I was seeing 100C+ temps on the memory and it was causing the card to limit itself for obvious reasons. Had to swap better thermal pads onto it. For comparison my 2080 Ti with it's EK waterblock never exceeds 65C on those modules, but would easily hit 85-90 with the stock air cooler, so if the thermal pads were crap....

 

If you've never pulled the card apart for maintenance and you've been running it a few years, there could be a heat issue that's not showing up in hardware monitoring. It doesn't 100% explain the levelling out, as usually the cards will just clock up and down constantly. Or crash.

 

*EDIT* Jesus I just noticed, you're using an ancient version of HWInfo64. It's into build 7.06 now man, you need to update it, it should able to show you more sensor detail.

[sxstack]

13 minutes ago, ApolloX75 said:

Interesting. Does HWInfo show you the Memory Tjunction temps and Hot Spot temps for your card? When I was benching my 3080 after reinstalling the air cooler I was seeing 100C+ temps on the memory and it was causing the card to limit itself for obvious reasons. Had to swap better thermal pads onto it. For comparison my 2080 Ti with it's EK waterblock never exceeds 65C on those modules, but would easily hit 85-90 with the stock air cooler, so if the thermal pads were crap....

 

If you've never pulled the card apart for maintenance and you've been running it a few years, there could be a heat issue that's not showing up in hardware monitoring. It doesn't 100% explain the levelling out, as usually the cards will just clock up and down constantly. Or crash.

 

*EDIT* Jesus I just noticed, you're using an ancient version of HWInfo64. It's into build 7.06 now man, you need to update it, it should able to show you more sensor detail.

Bios updated to F21.

I just updated HWInfo to 7.2. I procrastinated updating since it was working with my Streamdeck that I use for stats.

 

Not sure where to see the TJuntion temps for memory... but using GPUz I do see the hotspot temperatures (memory specifically) at 70C at idle (oof) and 107C during furmark.. Oh God. That is probably the issue. 

 

I am just surprised this is showing as Performance Limit - Power and not Performance Limit - Thermal in HWInfo?

[sxstack]

17 minutes ago, sxstack said:

Bios updated to F21.

I just updated HWInfo to 7.2. I procrastinated updating since it was working with my Streamdeck that I use for stats.

 

Not sure where to see the TJuntion temps for memory... but using GPUz I do see the hotspot temperatures (memory specifically) at 70C at idle (oof) and 107C during furmark.. Oh God. That is probably the issue. 

 

I am just surprised this is showing as Performance Limit - Power and not Performance Limit - Thermal in HWInfo?

Just purchased some 1.5mm thermal pads as per a few other forums online.. hopefully this resolves my issue. I will update this thread if it does.

[sxstack]

Hello All!

 

I just wanted to update this thread for future people experiencing this problem.

 

So, I replaced all of the thermal pads and thermal paste (1.5mm pads for the 2080ti strix OC), and it reduced the temps significantly. Not only was the memory hot, but the GPU was hot - but it was reporting fine. The 'hot spot' temp was throttling the GPU, not the 'gpu temperature' in HWINFO - the previous version of HWinfo I was using did not have this hotspot temp. The strange thing is that it was reporting the limit as 'Performance Limit - Power' instead of thermal! 

 

My score in 3dmark went from 10,400 to 14,200! Easy 40% performance... I didnt realize it because it happened over a long period of time ~2years.

 

One thing to note - this was an RMA replacement card. It may have been a refurb or something. When I took the cooler off, the thermal paste job on the cooler was HORRIBLE. Not nearly enough thermal paste, which makes sense since the hotspot temperature was the one throttling me.

 

TLDR - Check your thermals - GPU, Hotspot, VRM and memory (in GPUz), even if it says Performance Limit - Power,

 

Thanks everyone!

[sxstack]

One last thing - I used 2x Gelid 80x40x1.5 mm thermal pads which I got on Amazon Canada for about 16.50 ea, and Kryonaut thermal paste.

 

Good luck!