Deux Ex HR – Director's Cut

[BuckleBean]

Just want to bring to your attention that performance has massively tanked since the latest version of vorpx, especially when using DX11. Something may be amiss. I did a DX11 playthrough several months back using Z3D, running the game at 4k & was generally able to maintain 75fps. I booted it up today and saw that I was getting around 20-25fps, even at 1080p. I tried DX9 and performance was much better, but even at 1080p, it was dropping into the 60s. I know the latest update, bringing direct mode, comes at a cost, but this seems excessive. I haven’t seen anything like this anywhere else and thought it was worth sharing. I’ve already played through the game on the rift and it was incredible, so it’s no biggie for me, but I was wondering if anyone else had noticed this.

Specs:

GTX 980sc (slight further oc)
i5 2500k @ 4.4gHz
16GB ram
Win 10

[Ralf]

Raw game performance is usually lower since 0.9 due to now two render processes working side by side (the game and the direct mode rendering), so your DX9 numbers may be normal. But what you are experiencing in DX11 sounds far too much. In most games the loss is about 20%-25%, which is compensated by the async timewarp.

DXHR (not the Director’s Cut though) is used for testing here from time to time. Usual test resolution is 1600×1200, graphics settings default, FOV maxed out in the game’s options menu. On a GTX 970 it runs with 40-70fps in DX11 using G3D, with some odd slowdowns below 25fps here and there (e.g. Saarif Industries lobby). Z3D performance should be a lot better than that, almost stable 75fps should still be possible on your GTX 980 using Z3D.

In some games forced MSAA through the graphics driver can have a huge negative impact. Not sure whether that is true for DXHR, but that may be worth checking out.

And just in case: until not too long ago nVidia drivers maxed out a full CPU core for doing nothing with the Rift. That could be pretty bad for the async rendering in vorpX 0.9 on CPUs without hyper threading (Core i5 and below). The issue is fixed with the nV drivers 358.91 and (hopefully) above.

[BuckleBean]

Thanks for the reply and for confirming that the performance drop seems off. I have no AA on and most other settings turned down. It’s just unusual behavior, so I figured it’d be worth mentioning. I should note, I’m not seeing this with other games. Perhaps I’ll try reinstalling the game or giving the non-director’s cut version a try. Ultimately, it’s not a big deal for me because I’ve already played through the entire game with vorpx and it was incredible. Still, I like to revisit that world from time to time.

Thanks again!

[BuckleBean]

Quick update: reinstalling the director’s cut didn’t help. The non-director’s cut DX11 version had the same sub-30fps issues in Z3D and even with 3D completely disabled. The DX9 version of the non-director’s cut seems to be the most stable in my limited testing.

But check this out, both version have a weird bug: when selecting Z3D, the 3D separation defaults to 0.00. When I adjust it, it changes the IPD! This does not happen with Z-adaptive or G3D (G3D is totally unplayable, though). Crazy, right?Something is definitely up here. For all I know it could be something on my end, I’m not above user error. But this behavior is just too weird.

[Ralf]

Separation should be 1.0 per default. You can reset the profile to default values in the config app.

Not sure about your D3D11 fps issue, sorry. Sounds definitely odd. Just tried here with the last 0.91 release version to be extra sure in Saarif HQ, Hengsha and the end-boss room. Results are like described above…

D3D11 Z3D is currently broken in DXHR. Will be fixed in the next version. Currently it always has the max. Z-zalue across the whole image. Since Z3D shifts the image based on depth, changing the separation shifts the whole image in this case, which looks identical to changing the IPD. So everything normal in this regard (except the depth values being garbage, of course).

[BuckleBean]

Even after resetting the profile, the separation in Z-normal & Z-adaptive is set to 0.0. Increasing it in Z-adaptive seems to work, but in Z-normal, it’s that IPD-like adjustment bug you mentioned. Makes my eyes go crazy. Weird stuff. But if you can’t reproduce it, and since no one else is reporting anything, I wouldn’t spend too much time on it. Just wanted you to be aware.

[Author]

Posts