AMD Software: Adrenalin Edition 23.9.1 - Driver Download and Discussion

Back here to quote my previous post. High Idle Power issue is now fixed on Windows 11 using the Lies of P driver, with the 7900XTX. I report 13 watts using 2 144Hz monitors at their full capacity. Congrats on AMD for fixing this. I was happy with the 7900XTX before, but now all problems are fixed I simply love it.

https://www.amd.com/en/support/kb/release-notes/rn-rad-win-23-20-11-01-liesofp

 

Don't get too excited at least wait a few days to confirm if it's actually working. Anyway check again after some hours of using the PC and see if the power draw is the same or it goes up because this "fix" was deployed in 2022 for RDNA2 and it's still buggy and only works for a limited time until the vram clock gets stuck in some random power state and the power draw jumps from 5W to ~20W.

 

Yep, apparently the "VRR trick" doesn't work 100% reliably and sometimes gets stuck in a middle VRAM clock state. At least on RDNA2.
It also causes frequent short refresh rate switches when VRAM clocks change, doesn't seem to be the case with the 4070 here.

 

I will report back after a few hours then. As of now, it's still reporting normal power usage.

 

It appears y'all are correct. It now sits at 30watts-28watts, but that's already way better than the 100watts I had before at idle. LOL

 

With the Lies of P Radeon software\drivers the TBP is down from circa 90 watts at idle to 55-56 watts. GPU is a Sapphire 7900 XTX Nitro+ Vapor-x with all settings at default.

 

It's funny how no one wants to make a new separate thread for the new Adrenalin for Lies of P.

 

It uses FreeSync to to trick. Disable FreeSync and power consumption will skyrocket.
VRAM frequency lock looks like it getting stuck at 0 mHz until you hit GPU with actual 3D load or reset driver (Win+Ctrl+Shift+B)

And don't be so happy. This fix is effective to reduce power consumption, but it can trigger additional issues instead.
For example it exaggerates issues that are linked to image corruption in browser with DX11/12 game running in background with flip-buffer presentation mode.
In addition when you actually push load that move VRAM freq state to proper value, image will freeze for 2-3 seconds EVERY TIME. And, in some games if you do recording at this moment (even though VRAM usually doesn't lock during recording, it can do so before you start one), and then launch game/alt-tab into it, you can cause game or driver crash as well.

Only real fix is enforcing of CVT-RB(or CVT-RB2) spec of v.blank byx creating custom resolutions... But for this you also may want to check if monitor signal integrity is still on same level as with innate resolution.
For me i also needed to technically tweak lower bar of FreeSync range (not VRR range, btw, just FreeSync extension specifically) from 48 to 49+ but as any refresh rate below 54 was technically LFC (low framerate compensation), basically nothing had changed for me, even though monitor became stable. And definitely not set it above 55 iirc, as it actually changes FreeSync behaviour so LFC stops working and minimum refresh rate bar is being set to 60 Hz

-------

P.S. looks like since i doubled up my custom resolution through Adrenaline, i stopped having issue of driver randomly going off tracks with it and breaking into VRAM frequency bug again. Looks like i had my issues fixed, even if wasn't a trivial experience to figure out.

 

What is "driver reset" and what is it supposed to do?
I tried this key combo, but other than beep and very short blank screen, nothing happens.
At least the GPU was definitely not reset.

 

Yeah it doesn't restart or reset the driver. I think it's just reinitializing the DWM buffers.

 

It is not DEVICE reset, but driver reset instead. It can solve some issues, but more often than not it doesn't do crap.
It does return VRAM state to normal though

Maybe, probably googling what exactly this keybind does is best choise

----------

 

Interesting to see on my Linux drive. This seems to be only the case on Windows, on Linux VRR works properly and my idle wattage stays 13 watts all the time. I even played Starfield and it goes back to 13 watts afterwards.

In Windows it's different, as after a few minutes the idle wattage is 28 watts. Seems like Windows is heavier reliant on the gpu.

VRAM frequency bug does not appear on Linux it seems on zen kernel 6.5.3 (Garuda Linux).

 

2 more builds of 23.9.2 are detected as in process.

Build number is quite a bit higher than first one that was detected on last week (395627)

 

Drivers 23.9.2 are released, but it's branched.

There's a separate branch for Polaris to Radeon 7, with changelog available.

RDNA+ changelog for 23.9.2 isn't up.

 

Important note from that changelog btw

• Users pairing an RDNA based graphics product with either Polaris or Vega based graphics product are recommended to use the AMD Auto-Detect and Install Tool.

Hmm... Unsupported HW? Looking at my 6750XT
And updating from Adrenaline worked... Hmm

Also there are 2 display drivers there!
Display = u0395628
Display2 = u0395974

In driver downloaded from driver page it is only Display2 driver that is provided.

Looks like u0395628 is from 23.20 branch, while u0395974 isn't

 

link do not wrong just for polaris/vega

whole 23.9.2 package anybody can download from adrenalin 23.9.1 system panel. it store after download in C:\AMD\AMD_Software_Installer_23.9.2

 

AMD Radeon™ RX 7900 XTX Drivers & Support | AMD

AMD official website download page is wrong. The download link is for polaris/vega driver (is NOT for RX 7000/6000 Series).

I mean the offline driver package download link is wrong.

 

I Know 23.9.2 package can download from adrenalin 23.9.1 system panel.

 

Used the Auto-Update Tool, and got:

6800M: 31.0.22011.1008
Vega iGPU: 31.0.21902.5

Let's see if this blows up my system again :erf:

At least Radeon is detecting both devices this time, and SmartShift is enabled. When I go to manually download, and using the Laptop Package option, I get the 23.9.2_combined package.