AMD Software: Adrenalin Edition 23.8.1 - Driver Download and Discussion

1. That's not how it works. Game doesn't request voltage. Neither GPU requests voltage directly. Moreover game i test voltage curve is ALWAYS same and in same place.
2. Voltage that was shown is HIGHER than instability threshold of 925, and by quite significant margin, so why would i want to increase voltage even further. If anything i have now more headroom to tighten down UV
3. To be completely honest, that't not a thing that could or should've happened at all. Driver is NOT responsible for voltage curve. VBIOS is. And i didn't change VBIOS, so this is quite sus.

 

Where do you think firmware blob's updates on Windows system come from?

 

From manufacturer site.
Aka, for me it will be there: https://www.aorus.com/graphics-cards/GV-R675XTAORUS-E-12GD/Support
And since i own card, i only had V1.2 VBIOS version, which is installed.
VBIOS is device specific, so AMD cannot provide update for every AIB out there from driver.
Not to say that VBIOS flash requires device reset and specific binary, which is non-existent in driver package.

 

PowerColor HellHound 7900XT Worked fine for a little bit after install - then while playing WoW I got a black screen crash to BSOD. Upon reboot the video card was present in the system but in a disabled state. Rolled back to 23.7.2, and eventually back to 23.5.2 (temps were better with this version than the 23.7.x's using the balanced mode preset). 23.7's never crashed or anything just like the temps while gaming in balanced more better.

 

AMD starts to use separate binaries for each IP since Vega iirc, and most of these IPs SKU specific, each having its own version id (maj.min.rev)
Just take a look into linux/firmware/amdgpu folder.

Not sure how relevant is VBIOS flashing here. PSP has to preload firmware blob to VRAM in any case, and can patch it on-the-fly before uploading. Besides PSP can load a binary from the system memory through the gart.

 

Let me re-phrase what I wrote: It might be this game that pushes GPU harder than most and thus, the UV you have now is more than the GPU needs to be stable when pushed this hard. That is why when I tested 3D Mark Firestrike using one of my UV profiles it was stable but once I tested Port Royal I got error in the first few seconds and needed to up the voltage to allow this test to finish without errors.

 

I think there's a good reason to assume UV issue, although there's always a chance to get an error even in the same 3D app just with a different driver version.

As an example, I've tested Time Spy with every driver since 22.5.1 for the minimum stable Voffset and found it to be 1075mv until May ’23 (there were a couple of releases with min 1082 though). Then up until current version, 1063-1067 mv.

The reason is very simple - from a low level perspective, a driver may change the required number and/or type of circuits that are switching during a particular task, which in turn causes the SMU or RLC to adjust voltage (or even engage some PG/CG functionality) based on measured Cac of said circuits' replica paths or corresponding PSM counts to maintain the target frequency.
However, it should be borne in mind that temperature variance make things more complicated and less predictable. Thats why the additional Vgfx guardbands exists.

 

Yeah, yeah, except i have HIGHER voltages than before on most unstable fragment of a curve despite using same game and same scene. So you suggest that higher voltage is more unstable because of UV? Mind you it is medium part of a curve, i don't hit any thermal limits, not even close.

Sure, but not with this as condition


111.8 hours of gameplay in this specific game and only one system freeze during all this time. Not to say more hours in other games.
UV instability is definitely much more definited, as steps are quite heavily granular, meaning it would've happened more than one time.

 

22.8.2 was sent to compile, expect it within next few days.

 

That's why when I find a stable UV I always add 15-20mV. Some people like living on the edge and that's ok if they are willing to accept that it's not a driver issue, it's their UV.
I mean if your UV crashes if you lower voltage by 5mV expect to have crashes with different drivers. Always have a ~20mV buffer if you don't want to deal with this.

 

That is a smart way to do it. People do go agressive on the uv then blaming the drivers for crashing.

 

I would've agreed if you had full control over curve, or if my tested vulnerable spot was at max voltage (not that you can go over 1200 mV btw).
But you cannot just get +20 mV if you offset curve by 40 mV and tested spot requires only 935 mV from 1200 max. Only thing in my power is frequency to voltage offset. And at this low point of GPU frequency even 6 mV is enough to turn unstable into stable.
It is almost same as wanting to add +20 mV to idle voltage despite no direct way to control this voltage at all.

UPD... Hmm, did they somehow changed how voltage slider in Wattman works? Because it doesn't seem like like almost no difference at some parts of a curve as it did few driver versions before. Now it looks like much more significantly affecting lower frequencies.
Still, at 2440 freq my voltage now is 950, and it should be stable down to 931-937. Don't you think it is enough overhead here?

 

Anyone with Polaris card? how does the driver works for you compared to 23.4.3 or 23.5.1 which where great drivers for us.

 

Idk what are your UV settings. For a 6750XT the voltage should be higher unless you also lowered the max frequency. If you mess around with the frequency it will also affect the curve and idk how that plays out.
Assuming you didn't modify the min/max frequencies there should be a minimum voltage that is stable and that allows the GPU to reach it's advertised GPU game clock frequency in most games...2514MHz for my xfx 6700XT and I assume it's 30-50MHz higher on the 6750XT so why are you talking about 2440 MHz ? It should reach ~2550 MHz.
So you either lowered the max frequency or your voltage is too low for the GPU to reach the advertised frequency...and in that case it could also be unstable at lower voltage/frequencies.
Either way I didn't modify the min/max frequencies and I can't comment on that as I don't know what happens.
I'm testing stability with fire strike extreme, time spy, port royal, superposition, heaven, furmark, metro Exodus, AC Valhalla and FC6 )
When I find a voltage that is stable in all those benchmarks and games I add 20mV to that. That's all.

 

It is for specific setup of 2440 max frequency, which was my voltage gap before. This is SPECIFICALLY to test voltage stability at weakest point of curve for my GPU. And i did quite extensive research at some time. Voltage of 925 and lower causes GPU to crash at 2440 set / 2390 get, while 931 is stable (at this point of curve even 6 mV is enough to guarantee stability as scalability is really good)
At 2825-2850 i am using daily voltage will ofc be 1200 which is default max.

3DMark tests are too easy to pass for stability to me. Superposition is just bit harder. But games like Warframe for me are toughest. Singlehandedly dropped max OC from 2915 to 2850 at least.

 

Hey guys I realize I am probably super late to this table but I never noticed Enhanced Sync before and was wondering if I should have that enabled? I have a 240hz Samsung G7 with Freesync 48-240hz and FRTC set to 238. I am going to set it on for now as it seems like I should have it enabled. Do you guys use it? Why or why not.

Thanks

 

It allowed for FPS unlocking in some games with fullscreeen, like Nvidia VSync = force off.
Technically it still does that, but Enchanced Sync really doesn't like Win11 flip-buffer model atm, so i currently hold it on "off"

 

Thanks for the input! I am skipping Windows 11 so that doesn't affect me I guess.

 

I was living on the edge with my 1025v UV on this 6800 XT and graphically intensive games were rock solid, for instance I could run Time Spy and Port Royal stability tests with no issue, but when I started to play ADACA I had crashes because the lower voltage steps that I hadn't triggered playing more graphically intensive games were unstable. I popped it up to 1050v and all that went away - I am just leaving it at 1050v now. Wise words of wisdom to find your "stable" UV and add a little bit on so you're rock solid stable.

 

Depends on game. Some games have flip-buffer presentation model by default or as option in settings.