• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

RTX 5090 - Incorrect GPU-Z Clocks, when Overclocking with the Curve Tuner

outlw6669

outlw6669

Joined
Feb 29, 2016
Messages
81 (0.02/day)
Location
Germany
System Specs
Processor Ryzen 7 5800X - Curve Optimizer @ -5/-20/-0/-10/-15/-15/-20/-20 - Tested to be Rock Solid Stable
Motherboard ASUS Prime X470-Pro - VRMs @ MAX Switching Frequency
Cooling 2x VPP755 V3 + Eisdecke D5 dual plexi - Corsair XR5 420mm - 2x NexXxoS ST30 V2 560mm - CPU=>RAM=>GPU
Memory 32GB Dual Rank DDR4 @ 3733 MHz 1:1 - Cmd2T 1T - tCL 14, tRCD 8/15, tRP 15, tRAS 31, tRFC 261 (140ns)
Video Card(s) RTX 5090 Suprim SOC (+2000 MT/s; +200 MHz; Curve flattened @ 975mw) + Alphacool Core Waterblock
Storage WD Black SN850 2TB
Display(s) Samsung Odyssey Neo G8 / LG OLED 48CX
Case The Tower 900 - 24x 140mm Fans Installed
Audio Device(s) Onboard to AKG K701 / Atmos surround
Power Supply be quiet! Straight Power 11 1000W Platinum
Mouse Logitech G305
Keyboard Royal Kludge RK61 - Mix of Kailh V2 Super Speed Bronzes, TTC Bluish Whites and Crunchberries
Software Windows 11 Pro
Hey, hope this is the correct place to report a minor bug in GPU-Z...

This particular bug has persisted for at least the past two releases of GPU-Z, as well as nVidia Driver Versions 572.16 and the current 572.47.

Issue:
  • Upon using Afterburner to apply an overclock, GPU-Z reads incorrect GPU Clock and Boost clocks on a RTX 5090.
To Reproduce:
  • Install Afterburner.
  • Open GPU-Z
    Apply Overclock (I am using the Curve Tuner)
  • Clocks in GPU-Z update to show incorrect readings
  • GPU-Z Sensors tab still read the correct clocks
System is shown in my profile; running Windows 11 Pro 23H2, with all latest updates and drivers installed.

The instillation is pretty fresh, with minimal software installed.
Running DDU (in safe mode) and reinstalling the drivers does not resolve the issue.
I have noticed that, after reinstalling the latest drivers, GPU-Z will report clocks matching stock, but it switches back to showing low clocks, as soon as you apply the OC again.

1 - GPU-Z - Restart without Afterburener applying OC.png


2 - GPU-Z - Apply OC with GPU-Z Open.png


3 - GPU-Z - Afterburener OC applied after startup - Stress Test 1.png


4 - GPU-Z - Afterburener OC applied after startup - Stress Test 2.png



EDIT:
Post Title updated, to indicate that the issue occurs when overclocking using the Curve Tuner.
 
Last edited:
I guess that could be due to how you configured the curve. Could you show a screenshot of that and what clock you'd expect to be displayed?
 
Now that you say it, it does make sense, that changing the V/F curve would cause some difficulties :oops:

You can see my curve below (flattened at 975 mV, from a base +200 MHz offset).
I suppose, showing the current Base and Boost clocks, taken at the same voltage points used to determine the Default clocks, would be the expected behavior.

Upon applying a non-curve OC, the correct clocks are shown, as expected.

5 - GPU-Z - RTX 5090 V_F Curve in Afterburner.png


6 - GPU-Z - Correct Clocks Shown with non-curve OC.png
 
outlw6669 said:
taken at the same voltage points used to determine the Default clocks
Click to expand...
good idea. is that how it really works? How is that voltage point determined?

Even if used, I'm sure you can imagine ways to adjust the shape of the curve to "game" the reported number?
 
W1zzard said:
good idea. is that how it really works? How is that voltage point determined?

Even if used, I'm sure you can imagine ways to adjust the shape of the curve to "game" the reported number?
Click to expand...

Sorry i didn"t find how to contact, so i write here, the DLSS DLL 310.2 s infact 310.2.0 since Indiana Jones update 3 has 310.2.1 (newer CL).
EDIT:
it's DLSS G file tho !

I think DLL G & D should have a longer version too !!

www.techpowerup.com

NVIDIA DLSS DLL 310.2 Download

This download provides various versions of NVIDIA's DLSS DLL for download. In this one file, which is bundled with all games that support NVIDIA's
www.techpowerup.com www.techpowerup.com
www.techpowerup.com

NVIDIA DLSS 3 Frame Generation DLL 310.2.1 Download

This download provides various versions of NVIDIA's DLSS 3 DLL for download. In this one file, which is bundled with all games that support DLSS 3
www.techpowerup.com www.techpowerup.com
www.techpowerup.com

NVIDIA DLSS 3.5 Ray Reconstruction DLL 310.2.1 Download

This download provides various versions of NVIDIA's DLSS 3 Frame Generation DLL for download. In this one file, which is bundled with all games tha
www.techpowerup.com www.techpowerup.com
 

Attachments

  • Presse-papier02.png
    Presse-papier02.png
    29.3 KB · Views: 61
Last edited:
Low quality post by Toss
W1zzard said:
good idea. is that how it really works? How is that voltage point determined?

Even if used, I'm sure you can imagine ways to adjust the shape of the curve to "game" the reported number?
Click to expand...
Yeah, I spent some time looking, but could also not determine exactly how the base and boost clocks are being determined.
I am sure it must be documented somewhere, but it is beyond me to locate it.
The closest Info I could Find is that the NVAPI lets you directly read the Base and Boost clocks, but I did not see anything about reading directly from the V/F table.
I assume that the V/F table can be read somewhere, though, as its current state can be seen in Afterburners Curve Tuner...

If GPU-Z is relying on NVAPI to read the reported Base and Boost clocks, then compares them to the values stored in the BIOS, I guess that there is not much that can be done.

If, however, it can grab values from the V/F curve, and compare them to the corresponding values in the BIOS, perhaps some logic like the following could be possible?

If (V/F Curve = Uniform Offset) Then (Report BIOS Clocks +/- Simple Offset)
Else If (V/F Curve = Non-Uniform Offset) Then (Search BIOS V/F Curve for closest match to Base/Boost reported frequency; Sample V/F Point Offset + Sample X-Values Above and + Y-Values Below; Discard any Outlier Offset Values; Average Reported Offsets; Report BIOS Clocks +/- Average Offsets)

Or, with simpler logic: Sample all V/F points, discard outlier offset values, average all offsets together, report BIOS clocks +/- averaged offset.

In any case, I am probably a bit of an outlier here, and I doubt there are many that would ever see this issue, so probably not worth investing any real time into.
 
outlw6669 said:
discard outlier offset values
Click to expand...
That's an interesting idea

outlw6669 said:
In any case, I am probably a bit of an outlier here, and I doubt there are many that would ever see this issue, so probably not worth investing any real time into.
Click to expand...
Yup, but it's definitely something to consider when I touch that code in the future. I have plans for a full rewrite of GPU-Z for version 3, at which point I could consider this. Appreciate the feedback and the detailed investigation
 
I have similar problem relating to this with 5070 TI Windforce. I have had two different 5070 TI Windforce GPU's in the same system. Both with same undervolt profile and mem oc with MSI afterburner. For some reason GPU-Z shows lower clocks on the other one. Actual performance is still within 1% margin with both cards with 3Dmark benchmarks and cyberpunk. One other odd thing is that the GPU that showed higher clocks in GPU-Z had lower idle power usage around 27w with multi monitor setup. Now the one with lower GPU-Z clocks has higher 33w idle power consumption. I have monitored that both cards still handle atleast 1h of normal desktop use without the fans starting to cool the card. The Gigabyte Windforce cooler turns the fans off at 42c and ramps up again at 58c tool cool back to 42c. Both cards idle around 54-57c after 1 hour, in a mesh case with very low rpm case fans and around 23c ambient.
 

Attachments

  • Curve.png
    Curve.png
    33.2 KB · Views: 56
  • 5070 ti gigabyte gpu-z lower clocks.png
    5070 ti gigabyte gpu-z lower clocks.png
    266.3 KB · Views: 42
  • Gigabyte GPU-Z 900 2800 mhz.png
    Gigabyte GPU-Z 900 2800 mhz.png
    51.4 KB · Views: 46
The V-F curve is absolutely identical on both cards?
 
W1zzard said:
The V-F curve is absolutely identical on both cards?
Click to expand...
Should be. They are the exact same model from the same manufacturer. Did not think to take a screenshot of the curve when had the previous GPU. Both are raised with shift from 900mhz to 2800mhz on both and then flatten the rest.
 
oh you don't have the other card anymore. I suspect that the V-F curve is slightly different depending on binning
 
You must log in or register to reply here.
Back
Top