AMD Software: Adrenalin Edition 23.5.2 - Driver Download and Discussion

1440 - 144Hz
1080 - 60Hz

That behaves for me. The 1080 panel is capable of higher but the memory clocks max out if I go above 60. For my situation it works as I game on the 1440 screen and watch videos on the 1080 screen. I don't have superwide screens but might be worth looking at similar in your case?

I also have a 4K 120Hz TV connected to the card but keep that turned off unless I want to watch a film or something. No matter what I do, as soon as a 3rd screen is introduced the clocks max out.

 

Try setting the 1440 to 120Hz, and let the 1080 go higher in multiples of 30.

 

I was talking about 2 screens.
With 3 screens I don't think you'll have idle memory clocks whatever you do.

 

These damned rare system lockdowns... There is proper sound in background (video is played normally for example), but image freezes and system stops reacting on anything. Forced to do hard reset.
And they happen quite rarely, once a week or so... No idea what causes them still, i literally tested whatever i could. And i doubt it is PSU, as they happen on low GPU load with steady CPU one.

 

When does this happen? All of a sudden in windowed browser etc?

 

All of a sudden in games. Usually at quite low load, where i cannot even edit voltage table for GPU. Often on alt-tabs, but last time wasn't on alt-tab iirc.

 

Do you have fast timing enabled on the VRAM tuning? That caused similar issues for me in the past. Since then I don't even touch the fast timing option.

 

I do, but i pretty much tested stability of VRAM as well. FT2 does cause issues, but not FT.
Also, too random and rare, don't you think? Once of week of consistent usage? (And i almost never turn PC off)

 

Idk but it's worth a try for a week or two to see if that is the cause. For me it was causing random freezes only in some games like elite dangerous and warframe. Same symptoms like system freeze with audio playing in the background or audio looping in the background. This was 3-4 years ago when I had a 5700XT. Since then I only enable fast timing when I run 3dmark or some other benchmarks. Never had these random system lockdowns with default timings.

 

Not quite related to the topic but...

It is.

No.

Weirdest explanation I ever seen, but ok.

 

Thanks for the suggestion but doesn't make a difference for me mate. 1440 at 120 and 1080 at 90 or 120 and the clocks max out.
I'm happy with how 've got it set up anyway. Don't need any more than 60 on the 1080 panel and 144 on the 1440 is needed for rocket league, the only game I play

 

Surprise surprise, but it is not. There is such thing as voltage drop across plane. Meaning if VRM reports it sent one value, CPU reports it requested another one, voltage it gets will be some third value. And that's why even probes on back of the socked aren't showing you actual voltage CPU gets. It gets lower when it reaches CPU, and even lower when it reaches cores.

Depends on mobo. Normal ones have chip that copies readings from CPU internal sensor to Super I/O, or other workarounds, which are seemingly decently widespread now. Or AMD made it easier to read such voltage, who knows? But it was a thing, as Buildzoid is quite knowledgeable and i trust him in that.

 

Didn't mean to quote you sorry. Intended to reply to the guy asking the original question, hence my comment about superwide.

 

Warframe hadn't frozen on me once yet iirc. But i will do gpu memtest again. This one for 2 hours
I also did FAH + Prime95 Large test for FCLK/SOC stability. No errors. No sound stutters. No mouse movement issues.

It may just be something in drivers and freesync / flip-buffer model / MPO / DWM

This issue can also happen once a month, rather than once a week. So much for VRAM inconsistency... VRAM issues usually appear decently fast. Few hours at worst.

 

With what are you testing vram stability ? I used to run the VRAM test in OCCT for 2-3 hrs and it was 100% stable....and had the same issues until I set the default timings lol.

 

Memtest vulkan.

https://github.com/GpuZelenograd/memtest_vulkan

 

"AMD forgot to launch HYPR-RX"

I wonder if their lack of HYPR-RX completion is holding up the next version of Adrenalin.

The month of May was quite a while ago...

 

Let them cook good stuff. Sometimes giving people time to work without tight deadlines is actually quite productive.

 

That's what i get if i set FT2 min clock (2248 for me)

Spoiler

And that is what it should look like [used FT 2312 (my max VRAM clock available)]

Spoiler

 

Yes it is, if we're talking about the voltage supplied to the cores and L3 logic plane, i.e. RVDD, and reported as Vdd core SVI2 by hwinfo. What CPU request is reported as Core VID.

How AMD read voltages?
Vdroop on each rail is measured by the PWM controller using each power stage's IOUT-sense (intersil) or IMON (mps) to provide SVI2 compliant telemetry data.
Zen has no internal sensors for the voltage planes and relies upon SVI2 TFN. It's own power measurement afaik based on data from locally distributed PSMs or Cac interface (legacy)

PS: by the way, buildzoid himself figured out how LLC works not so long ago, what he honestly admitted in one of his videos )

PPS: Llc is just adjusting the load-line slope defined by setting Rcs when pwm controller boots, and could be done runtime, for example via SVI2 VOTF request.