Question Asus Strix 3090 Severely Underperforming in 3DMark with high GPU load.

[GSF2021]

CPU: Ryzen 5900x, H170i AIO Cooler
Motherboard: ASUS TUF x570+
Ram: 3600mhz 16x2 GB Corsair Vengeance
SSD/HDD: 970 Evo
GPU: Asus ROG Strix 3090 OC
PSU: EVGA SuperNOVA 1300w G2 80+ Gold
Chassis: Corsair 7000x
OS: Windows 11
Monitor: Asus VG279QM

Hello everyone,

I recently began working on a 3090 for a friend that had some thermal issues - the hotspot was reaching 105C and throttling, so I repasted it and eventually got it down to 95C maximum when running heavy benchmarks, in this case 3DMark's Timespy, which isn't great but would be enough to stop it from shutting itself down from heat which was my main goal. I figured I'd repaste it again to get it even further down, but I noticed something about the actual results themselves: they sucked.

3DMark Link: https://www.3dmark.com/spy/36473789


Bottom 10% in general in 3DMark, and bottom 8% of graphics score. Average clock frequency clock of 1838.

My CPU score is about average, GPU memory frequency was stable at 1219mhz, but my GPU score was down by 1,550. It's performing only sightly better than a 3080.

My initial thought was that the GPU wasn't actually under load and something was bottlenecking it, but looking at GPU-Z while running showed it at 97% during most of the testing. I just can't figure it out.

Would a hotspot temp of 95C really cause it to thermal throttle that much? I don't think so but maybe I'm missing something.

I'm going to try reinstalling my GPU drivers just incase, but any other advice would be appreciated. Can send screenshots if needed.

Thank you!

 

[Lutfij]

For the sake of relevance, can you include the BIOS version for your motherboard at this moment of time? Age of the PSU in your friend's build?

 

[GSF2021]

For the sake of relevance, can you include the BIOS version for your motherboard at this moment of time? Age of the PSU in your friend's build?

I can check the BIOS tomorrow morning, but it was lasted updated a few months ago IIRC.
The PSU is a year and a half old, ordered in July of 2021. I've used it with a 4090 with no issues if that helps.

Thanks for the response, apologies for not including the driver or bios versions before I left for work.

 

[Lutfij]

The RTX4090 is actually more power efficient compared to the RTX3090, the latter also has some bad transient load spikes...as showcased in this video.

 

[Roland Of Gilead]

I recently began working on a 3090 for a friend that had some thermal issues - the hotspot was reaching 105C and throttling, so I repasted it and eventually got it down to 95C

Hey there,

So, this is not exactly normal. Even at full load, and with the right cooling like the Asus 3090 has. Typically you can expect temps to be at about 65-75c at full load/gaming.

The case and fan orientation are decent for that case. So airflow shouldn't be an issue.

What thermal paste did you use?

Are all system drivers, chipset etc up to date? Did you clear CMOS after the last bios update? This would be important.

 

[GSF2021]

The RTX4090 is actually more power efficient compared to the RTX3090, the latter also has some bad transient load spikes...as showcased in this video.

Hm. I see. I might be able to record a GPU-z log while running 3DMark and look at the power usage, but I'm not sure what I'd be looking for exactly or how it would cause my GPU to be handicapped that badly. I'm not too knowledgeable about power supplies.

Hey there,

So, this is not exactly normal. Even at full load, and with the right cooling like the Asus 3090 has. Typically you can expect temps to be at about 65-75c at full load/gaming.

The case and fan orientation are decent for that case. So airflow shouldn't be an issue.

What thermal paste did you use?

Are all system drivers, chipset etc up to date? Did you clear CMOS after the last bios update? This would be important.

Ah, not good then.
Artic MX-4 thermal paste was being used. I repasted the GPU die as well as the pads to the right like demonstrated (poorly) below:

When I use GPU-Z or HWMonitor, I see that my GPU temps and GPU memory temps are around 60-75C, so I believe then that the hotspot is still the GPU die, right?

I didn't actually clear CMOS - thank you for the idea, I'll try that!
I believe all my drivers are up to date, but I can reinstall them just to be sure.

I'm really new to repasting GPUs - if adding new thermal paste to the die didn't drop temperatures below 95C, I'm wondering if I did it wrong.

Thank you again, I'll get back to you soon!

 

[Roland Of Gilead]

Yeah, MX-4 is pretty decent. Should work just fine. Being honest, the difference between it and say Noctua NT H1/2 which would arguably be better, are just a couple of degrees C in difference.

Yeah, it kinda does point to the GPU core, but if the repasting wasn't well done (and I don't mean to be rude - It took me many attempts to redo my first laptop cpu repaste.) then it could cause significantly higher temps.

Yes, do let us know how you get on.

It's possible the CMOS clear could fix it, but lets rule one thing out at a time and get it sorted.

 

[GSF2021]

Yeah, MX-4 is pretty decent. Should work just fine. Being honest, the difference between it and say Noctua NT H1/2 which would arguably be better, are just a couple of degrees C in difference.

Yeah, it kinda does point to the GPU core, but if the repasting wasn't well done (and I don't mean to be rude - It took me many attempts to redo my first laptop cpu repaste.) then it could cause significantly higher temps.

Yes, do let us know how you get on.

It's possible the CMOS clear could fix it, but lets rule one thing out at a time and get it sorted.

For the sake of relevance, can you include the BIOS version for your motherboard at this moment of time? Age of the PSU in your friend's build?

No offense taken - it was quite literally my first attempt, I'm sure it sucks.

Taking a look, my BIOS is actually out of date by roughly 3 months, so I'll be updating that tomorrow morning, I just got done with 16 hours of school and work and need some rest.
I have the latest 531.18 driver for Nvidia, so I don't think's impacting it.

After updating and clearing CMOS, I'll try repasting the die and posting some more informative results hopefully by the end of the day or soon after.
Thank you for your time, there's not many people where I live who are able to help with this kind of thing, and its much appreciated

 

[Roland Of Gilead]

No offense taken - it was quite literally my first attempt, I'm sure it sucks.

Taking a look, my BIOS is actually out of date by roughly 3 months, so I'll be updating that tomorrow morning, I just got done with 16 hours of school and work and need some rest.
I have the latest 531.18 driver for Nvidia, so I don't think's impacting it.

After updating and clearing CMOS, I'll try repasting the die and posting some more informative results hopefully by the end of the day or soon after.
Thank you for your time, there's not many people where I live who are able to help with this kind of thing, and its much appreciated

No probs, you're very welcome.

Yes, it can be that a bios update could fix this. Defo update, and let us know how it goes.

 

[GSF2021]

No probs, you're very welcome.

Yes, it can be that a bios update could fix this. Defo update, and let us know how it goes.

Sorry it took a bit longer than expected, but I reset CMOS and my BIOS is up to date. Turns out the "most" up to date version they released is a beta so I decided to not update to it.

I took apart my 3090 again and took some photos, I noticed a few potential issues.

The thermal paste is a bit messy on the die but it does cover everything. I'm not sure if the consistency would have a big impact on it though. The thermal pad on the left of the die is torn with a small bit sticking to the other side. I imagine I could take it off and put it back.
There's thermal paste on top of some areas because I guess the pads don't come with paste on them, which is I guess what that crust is, which I was told he had to add in a spot because it wasn't connecting fully.
The previous pads used by my friend were these Kritical Pads: https://kriticalpads.com/evga-3090-ftw3.

I honestly don't know much of what I'm looking at, but I'll note that however weird it looks, memory temperature and GPU temperature always seemed okay in testing. In 3dMark, memory temperature maxed out at 82C while GPU temp maxed out at 78C. From what I can tell, those are both normal, if a little on the high side, so I don't think any issues are coming from them.

However, in the bottom image, there seems to be a strip completely without a pad except for the ends. Do you need a full strip? Could that be the issue?
Thank you!

Edit: I found this site that had some of the pad sizes and stuff (https://thermalpad.eu/thermal-pad-sizes/asus-rog-strix-nvidia-geforce-rtx-3090/) and it seems that the whole area does need 0.5mm thermal pads. I have some spare GELID thermal pads in that size, which aren't great I'm sure would be better than nothing. I can add those if it might be a good idea.