FFXIV poor performance outlier on new Radeon 9070 XT cards!?

[s32ialx]

I suspected that. Any driver since the 23.11 update broke something for ffxiv in this case.

Going to try this once the servers come back up cause ermergerd the stuttering was just sooo bad it made the jump puzzles even more annoying than normal xD

[Miohazuki]

Heya everyone ^_^/

So, the news is out on AMDs Radeon 9070 XT that many may be looking to upgrade to on PC, but unfortunately it seems that whilst it's providing great value for nearly every other game, for FFXIV in particular, it severely under performs



Is there any technical reason why FF might perform so poorly compared to other games or do you all think we can expect this to be some kind of limitation that will be patched out via drivers or game update?

Source: GamersNexus 9070 XT Review

Ffxiv has always historically struggled on amd cards and the msrp on both new Nvidia and amd cards is bad .so might as well go with Nvidia card

[Zantetsukenn]

There is like 250€ difference in my country wdym?

[Velcio]

Since I didn't see any actual A/B testing wrt DX11 vs DXVK I asked a friend to run the Dawntrail benchmark Max settings, 4k. He's running on a Ryzen 5 7600 CPU and the 9070 XT with default clocks on the card.

DX11

Score: 10808
Average Frame Rate: 75,61709
Minimum Frame Rate: 47
Performance: High

DXVK

Score: 21593
Average Frame Rate: 163,9759
Minimum Frame Rate: 49
Performance: Extremely High

That is an absolute stupid amount of performance uplift by changing to a 3rd party renderer and tells you something is absolutely ridiculously amiss here.

Despite the performance uplift I am reticent to buy the 9070 XT because of reports of stuttery performance regardless of the renderer in crowded areas as reported by this thread. Any update on this by anyone? Supposedly FreeSync causes stutters in windowed mode too so that's another cherry on top.

Anyone running the 9070 XT and Windows 11/DXVK seeing any improvement if they go to
System > Display > Graphics and set Optimisation for windowed games to "On" (should be by default I think?) and add/change FFXIV settings to set the optimisations "On" and set GPU preference to "High Performance"? Like so:


At this point it just seems to me I should shell the 100€ or a little more I see the price difference being between the cheapest 5070 Ti and 9070 XT models locally despite all the negative things happening around Nvidia to ensure my own sanity, as much as I hate to do it.

[Larirawiel]

Supposedly FreeSync causes stutters in windowed mode too so that's another cherry on top.

Freesync/Gsync are broken in windowed mode. It's a Windows issue and not a driver/AMD/Nvidia issue. It may work in some games but do not expect, that it will always work. Nvidia has an option to enable it in windowed mode. But is basically an OS hack.



Cheers

[aiqa]

That is an absolute stupid amount of performance uplift by changing to a 3rd party renderer and tells you something is absolutely ridiculously amiss here.

It's not a 3rd party rendered. That is translating DirectX API calls to Vulkan. The FF14 renderer isn't changed at all.

That just shows AMD drivers are very inefficient for certain DX11 calls. Only AMD can fix it, but they probably wont.

[selmon]

Since I didn't see any actual A/B testing wrt DX11 vs DXVK I asked a friend to run the Dawntrail benchmark Max settings, 4k. He's running on a Ryzen 5 7600 CPU and the 9070 XT with default clocks on the card.

DX11

Score: 10808
Average Frame Rate: 75,61709
Minimum Frame Rate: 47
Performance: High

DXVK

Score: 21593
Average Frame Rate: 163,9759
Minimum Frame Rate: 49
Performance: Extremely High

That is an absolute stupid amount of performance uplift by changing to a 3rd party renderer and tells you something is absolutely ridiculously amiss here.

I had a friend who owns a 9070 try this in-game today and he saw an increase from ~51 FPS to ~65 FPS in Limsa (1440p) - that is massive but no-where near the numbers claimed here. I also tried it with my trusty old 980Ti and for me it was slightly worse. In both cases a logfile was written confirming that Vulkan was used.

I then tried to run the same benchmark as described in the quoted post - stock and with DXVK - and noticed that with DXVK the benchmark tool would only render in a small window on my screen. Regardless of whatever window setting you set... resolution, windowed, fullscreen, it did not matter, it only renders into a small window. So that's likely where this > 2x uplift likely comes from(?)

I will also note that my friend very sadly is one of those dreaded plugin-users and that he saw no performance-lift whatsoever when he had the plugin-system enabled (both ~50fps).

[s32ialx]

I had a friend who owns a 9070 try this in-game today and he saw an increase from ~51 FPS to ~65 FPS in Limsa (1440p) - that is massive but no-where near the numbers claimed here. I also tried it with my trusty old 980Ti and for me it was slightly worse. In both cases a logfile was written confirming that Vulkan was used.

I then tried to run the same benchmark as described in the quoted post - stock and with DXVK - and noticed that with DXVK the benchmark tool would only render in a small window on my screen. Regardless of whatever window setting you set... resolution, windowed, fullscreen, it did not matter, it only renders into a small window. So that's likely where this > 2x uplift likely comes from(?)

I will also note that my friend very sadly is one of those dreaded plugin-users and that he saw no performance-lift whatsoever when he had the plugin-system enabled (both ~50fps).

So just to explain that yes it is in a small window but it's still rendering at the scale you set, as well because it's a window and not full screen not only is it rendering at the resolution you set but it's also having to do extra workload of drawing the desktop, taskbar and everything else open in the background... AND IT STILL PREFORMS BETTER. D: please square fix this.

And yes those magic users unless the people who made those magics update them to support the 9070 they will not see any stable uplift in performance

[selmon]

So just to explain that yes it is in a small window but it's still rendering at the scale you set, as well because it's a window and not full screen not only is it rendering at the resolution you set but it's also having to do extra workload of drawing the desktop, taskbar and everything else open in the background... AND IT STILL PREFORMS BETTER. D: please square fix this.

That is not true, you can set arbitrary resolutions, and the window it renders in will always be the same size and the benchmark result the same. The benchmark is just bugged with DXVK. It's also not the behaviour you get with regular DX11 on the benchmark.

[s32ialx]

I wrote another email to AMD asking them why they think they can claim that it's Squares problem to FIX DX11 in the email they sent me about the performance but they then a month later go and fix a performance issue in WoW which also uses DX 11.... and not to mention MAJORITY OF THE GAMES OUT RIGHT NOW USE DX 11 Fixed Issues and Improvements: Lower than expected performance may be observed while playing World of Warcraft (western plaguelands) with Radeon™ RX 9000 series graphics products.