DX12 benchmark getting weird results on RX 7900 XT

[sweetpoison]

Hi,
I recently changed my old video card (intel A770) with RX 7900 TX. Out of curiosity I had a go with performance test (version 11.1 build 1002). All nice and smooth except DX12 benchmark with acts weird.
By default, single display at 1440p, I get 47 FPS (with the penalty for resolution lower than 4k). After I enable virtual desktop driver (set clone to physical one) I get 85 FPS. With virtual driver active I activated virtual super resolution from adrenaline software (control panel for AMD video cards) and set resolution to 4k and at the end I got 183 FPS. In games I didn't notice any change in performances regarding the state of virtual display.
Baseline for my pc: https://www.passmark.com/baselines/V...d=249046960287
Video capture of what I've said above (don't do any subscription or "like"; I use youtube for storage/sharing only, not for fame and glory) https://www.youtube.com/watch?v=-aOyZBeMWkA


Reason for this topis is hopping for software improvement by eliminating weird behavior of weird video cards (yeah, I share no love for AMD but there are worse on the market).

Thank you.

[David (PassMark)]

The 47 value seems low. Even the 85 seems a bit low.

It is normal that super resolution increases the frame rate. That is what it is meant to do. But you get some quality loss compared to native 4K.

[sweetpoison]

The 47 value seems low. Even the 85 seems a bit low.

It is normal that super resolution increases the frame rate. That is what it is meant to do. But you get some quality loss compared to native 4K.

The 85 FPS is on AMD as they sabotage own products with the drivers. Windows 10 default drivers for RX900XT provide much better results 129 to 150 FPS in DX12 benchmark / 1440 native display and the 47% penalty. The weird part is the result of benchmark without Virtual Display Driver (link) which highly affect the overall rating.
Driver 23.12.1 (Adrenalin), no virtual display: https://www.passmark.com/baselines/V11/display.php?id=230976809489
Driver 02/12/2022 (Windows) + Virtual Display Driver: https://www.passmark.com/baselines/V11/display.php?id=231568836823

Don't get me wrong, I am grateful for your work and the public database as it helped me many times to choose the components for PC (the most recent: acquisition of this video card). I was just a bit surprised to see that a card scored 28k (26k when I purchased the card) is actually gets a score way above 32k. That raise a question about accuracy if those results. Perhaps a key related to divers version (which can also be used for filters) will increase this aspect.

For testing purpose, is it possible to run DX12 benchmark in window mode 3840x2160 in size (even if it will show partially render on the screens with lower resolution)? It will be like the option found in many games "rendering size 200%".

Thank you.

[David (PassMark)]

We are doing a complete overhaul of the DX12 test at the moment. Almost a total re-write.
Frame rates were too high on the latest cards, but also the CPU load was too high. Among other issues.

There is no "fixing" some of this stuff however, as driver changes and driver settings (of which there are dozens) are always going to impact the results.
Happens with all the vendors. The initial Intel ARC cards got abysmal results, but they are pretty reasonable now, after 2 years of driver updates.

There has also been more recent issues of "real" frames vs "fake" frames. Where the driver adds new frames with a interpolation / AI process. Plus the rise of up-scaling with AI. What is a fair comparison is becoming a grey area with all the permutations and qualifications on results.

[sweetpoison]

We are doing a complete overhaul of the DX12 test at the moment. Almost a total re-write.
Frame rates were too high on the latest cards, but also the CPU load was too high. Among other issues.

There is no "fixing" some of this stuff however, as driver changes and driver settings (of which there are dozens) are always going to impact the results.
Happens with all the vendors. The initial Intel ARC cards got abysmal results, but they are pretty reasonable now, after 2 years of driver updates.

There has also been more recent issues of "real" frames vs "fake" frames. Where the driver adds new frames with a interpolation / AI process. Plus the rise of up-scaling with AI. What is a fair comparison is becoming a grey area with all the permutations and qualifications on results.

Awesome news regarding DX12.

Just a personal note regarding intel A770, the video card is PCIE v4 and on that slot performs pretty much as it is mentioned in database. However, placed in v5 slot increases the performances with 10-15% and that is not only visible in benchmarks but in games as well (personal experience). I'm sure this complicates you job even more but to far you proved capable to sole this aspects so I trust and wait for next updates.

Thank you for support.