Weird results, RX 6900 XT DIRECTX 12 and GPU Compute.

[MxTestSys]

Hi first of all I want to tank al the performance test team, excellent software, I been using this software for years now, I have to admit, that I like this bench over other alternatives, great tool for IT guys.

System
PROC - Ryzen 9 5950x (Liquid cooled)
GPU – RX 6900 XT (Asrock PG OC) connected by 3x8 different cables directly from PSU.
MB - Asus rog strix b550-f gaming wifi II (latets bios 3611)
PSU - seasonic 850w plus gold
RAM - 64gb RAM ddr4 3600mhz D35G
NVME - HP FX700 2tb
CASE – antec ax61 elite 9x fans (cool cool system)

Long story short
A year ago, build this PC all the same specs than above but initially whit a rtx 3080, all fine, bench a little above standard score, fine, cool system even under heavy loads, voltages fine, amps fine, watts fine, sturdy am4 b550, cheap, great and fast system ($1,000usd mixing new and used parts).


BIOS ACTIVE SETTINGS
D.O.C.P. 3600MHZ
Asus enhanced performance ON
CPU UV offset minus .175v (lost 5% performance and also 10ºc to 15ºc, win win)
SAM Resisable Bar enable

Then 1 week ago, change GPU 6900xt, due to the need for a little bit more VRAM, great, fresh windows install, bios reset to default, insert the new card, turn on the computer, install all updates and new PRO drivers.

First bench and surprise, initial score all default even in bios, 29,816 in 3d graphics 281w, average score of 6900xt is 26,796, 11% above average, I think, something is wrong, too good to be true, bench again, 30,000 ish, again 29,700, again 30,100 and so on.


Then, the same afternoon windows break mi AMD driver by installing an old driver and disabling all adrenalin software, so system properties > hardware > devices install config > NO do not auto install drivers, then DDU windows faulty AMD driver, restart, install new driver normal driver (NOT PRO) and bench again.

And now I’m only getting scores of 19,000 to 20,000 APROX 33% les “performance” even lower than the RTX 3080 10gb (25,500 aprox) voltages, amps and watts are fine hwinfo, watt get to its maximum 281w.

Change al bios to performance, test, the same score, enable SAM the same score, change to pro driver again, same score, disassembly the entire PC to reconnect all and check PSU OK, same score, starting to think on a GPU refund.

After 2 days, I read here that in AMD if you enable anti lag and enhanced sync get better scores, give it a try and now, I’m getting 27,000 28,000 ish, voltages, amps and watts are fine hwinfo, card gets to its bios marked spot of 281w easy.


Looking closely to the score, the lowest value is always DIRECTX12, and GPU compute others numbers are even
SCORE VS DIRECTX12
29816 = 128
27601 = 89
19899 = 35

GPU COMPUTE
My 13,500 aprox vs world average of 15,234

All confused, go whit TymeSpy, and do many test, there get scores above average, 21,217 vs average of 20,751 = 3%
So I think, ok, the GPU is fine.



NOTES
We have build many AMD radeon systems before and never have this problem, mostly 5700xt, 6650xt and 6750xt, passmark software always benching a little above average, this is the first time we have this issue, so as much as we know is not an AMD thing, and we also build many more nvidia systems than AMD ones, and everything always fine.

Is this something DIRECTX 12 or GPU compute only, cuz by the metrics of the card and the system everything seems to be OK.

Any advises, we enjoy a lot learn every day, cheers from Tepoztlan, Morelos, Mexico.

(We are a PC shop who plan, build and maintain Gaming and working hi end PCs since 2003.)

[David (PassMark)]

At the risk of stating the obvious. Different video card driver versions and driver settings can result in very different performance.

The difference might also not be just due to the video card drivers. There was obviously a windows update that took place before the performance went bad. So there might have also been changes to the Windows kernel (e.g. scheduler changes, or power management) and there might have also be CPU microcode changes applied for "security" reasons.

There are many factors that can effect performance. To really get to the bottom of it you would probably need to repeat the whole process, but install the Windows updates one at time and benchmark between each update.

DX12 and Compute are the two tests that depend the least on the CPU (especially if running at 4K monitor resolution).

Sounds like you got nearly all the performance back with the driver settings changes however.

[MxTestSys]

Thanks David, last night I try just as you say, fresh install again, all updates except windows amd driver, run test, same results 27,500 27,800.

And then my head start spinning, in the adrenalin software settings there are more options for screen sync similar to “anti lag” and “enhanced sync”, those two options enable raise my scores from 19000 to 27500 44% uplift, main windows up right corner settings > screen > two more settings related to sync directly from GPU.

-Adaptive Sync Compatible
-GPU scaling


Just enable those 2 bars and again get to 29,500 30,000 scores, again 11% above averge 6900 xt on passmark.


How is this possible, from 19,000 stock to 29,500 trying 2 days and making tweaks here and there its a 55% difference aprox, the average user dont need to know all this tech stuff, they want their performance out of the box, also, I have to admit, this is not normal cuz is the first AMD system that behaves in this way for us, I will try to extract that extra 11% score in all our customers AMD gpu systems from now on.

Anyone else have try enabling this extra settings to se if they get more score (11% aprox) out of a AMD GPU just as this one?

...DX12 and Compute are the two tests that depend the least on the CPU (especially if running at 4K monitor resolution)...

Could you tell me more about DX12 and GPUcompute please.

[David (PassMark)]

How is this possible, from 19,000 stock to 29,500....

I can only imagine that there is some downside (or compatibility issue) that prevents AMD from turning on all these settings by default.
For example I've seen reports that Anti-lag increases stutter slightly (time difference between each frame) and isn't effective if the CPU is weak. Unfortunately it is almost like you need to turn the settings on a per game basis.