Rx6700xt Baseline Variations

[docarter]

"[T]he default device driver settings are rubbish" . . . Now we're getting somewhere.

Sample size isn't the correct variable, there are clearly enough results to be statistically significant in the context of GPU performance. The 3060 is an example with a good curve and only 237 results. The 6900 XT has 438 samples but the standard deviation is still much higher.

Windows has corrected the stutter issue with a patch.

I'll just say that if you are attempting to make a GPU benchmark, but the benchmark has issues like driver incompatibility and CPU bottlenecks on what is a pretty high-end CPU. . . Then you need to go back to the drawing board.

[David (PassMark)]

We don't control default video card device driver settings. We don't control when Microsoft issues game breaking OS patches (and rolls them back). Microsoft did a similar thing with security patches that messed up the 2D benchmark not so long ago. We don't control power settings, cooling, overclocking, monitor resolution, PCIe slots used, CPU used and a dozen other factors.

It is totally unsurprising that some cards have a larger standard deviation when so many factors influence the results for real world data. Laptops are often even worst by the way, with thermal and power constraints.

All 3D games depend on both the CPU and GPU performance. You can't render 3D without using the CPU.

There is no driver 'incompatibility'. Maybe the default setting from AMD isn't optimal, but video card drivers have been like this for 25+ years. It isn't anything new.

Here is the driver tweaker screen from 1999 for a Voodoo 3 video card.

[DavyDee]

Thank you guys for discussing this issue in depth.

I have now a lead to giving my newly acquired 6900 XT another chance (was about to return it), in turning on Radeon Anti-Lag and Radeon Enhanced Sync and hoping for better scores in PT10.1.

One question I'd still have is how come the DX9 score in PT10.1 still plays such a big role - is it still that common in the industry?

Thx
Davy

[David (PassMark)]

DX9 doesn't play such a big role in the overall result (no more than DX10, 11 & 12).
But yes DX9 was the most popular version of DX in use by far a couple of years back. This is staring to change now however.
Developers largely skipped 10 & 11 (and the associated Win8 release).
DX12 was Win10 only and only on some video cards. So wanting to keep compatibility with the maximum number of video cards and Win7, lots of developers continued to target DX9. They moving on to DX12 now however, but there is a huge installed base of DX9 games out there.

[DavyDee]

Makes sense, David. Thanks for explaining.

I can also happily confirm that switching on Radeon Anti-Lag and Radeon Enhanced Sync (Adrenaline default is both off) does boost my DX9 and thus G3D score by 69%!

This is on Win10 x64 (20H2), Adrenaline 21.4 but been having this issue also with previous Adrenaline versions and cards.

[docarter]

Makes sense, David. Thanks for explaining.

I can also happily confirm that switching on Radeon Anti-Lag and Radeon Enhanced Sync (Adrenaline default is both off) does boost my DX9 and thus G3D score by 69%!

This is on Win10 x64 (20H2), Adrenaline 21.4 but been having this issue also with previous Adrenaline versions and cards.

Maybe report the issue to Radeon via the driver bug submission?

[DavyDee]

Great idea - thanks and done!

[David (PassMark)]

It does seem strange there is such a massive performance difference. Generally driver settings effects aren't so pronounced (i.e. the settings aren't generally too far from optimal for the general public).

We don't have an example of this video card to test with, but we've got some contacts in AMD we can report this to.

[DavyDee]

cool!