Rise of the Tomb Raider patch #7 adds DX12 mGPU, Async Compute support

[VulgarDisplay]

Do people think async compute is going to somehow save AMD or something?

It would be easier and more to the point just to say you don't believe the 1080's are benefiting from async.

What is the purpose of these posts other than to start an issue in a thread that hasn't devolved yet. A poster said, "hey, my 1080's saw a big speed up with this patch." Now people are wondering if it's because of reduced CPU overhead shifting the bottleneck in his setup, or if async compute is actually doing anything and you try to turn it into another nvidia vs. amd issue.

No one even mentioned amd at all, they just wondered if async actually sped up pascal.

 

[gamervivek]

Main point is that Pascal supports Async Compute and it can actually deliver a meaningful speedup with it. This is contrary to what some people claim.

It seems to do pretty poorly here.

https://www.youtube.com/watch?v=fecd4LV0TSU

 

[crisium]

I'm not really seeing any difference in the worst FPS areas. 1-2 fps better. Also the game still crashes often in DX12, so that didn't change. I would definitely not use DX12 here just because of the crashes.

 

[OVerLoRDI]

Installed the latest drivers and gave it a go.

1440p, high setting (except with motion blur off). Rig is in the sig. 290X crossfire.

Benchmark results:

DX12:
Min: 30
Score: 103

DX11:
Min: 3-10 fps (ran it a few times and the mins jumped around)
score: 85

I also did some gameplay around the Soviet Installation, felt smoother IMO, but then again, I haven't touched the game in a few months.

 

[Silverforce11]

What a contrast with id Software's Vulkan patch (first iteration) and this 2nd attempt for RotTR, still broken and slower. They should just give it a rest and move onto the next project. Next time, ask AMD to send some guys over to help them do DX12 properly. :/

 

[Cookie Monster]

It seems to do pretty poorly here.

https://www.youtube.com/watch?v=fecd4LV0TSU

114.72FPS (DX11) vs 150.13FPS (DX12) at 3:25min. Roughly ~30% gains.

How is it doing poorly? the overlay is confirmed to be inaccurate btw.

 

[Bacon1]

the overlay is confirmed to be inaccurate btw.

Source? How is the overlay inaccurate? I've tested multiple overlays and they all get the same results.

 

[Cookie Monster]

Source? How is the overlay inaccurate? I've tested multiple overlays and they all get the same results.

Read the comments?

Watch the vid? look at the actual fluidity of the game seen in the game?

 

[Termie]

I've put together some 980 Ti and 1070 benchmarks for ROTR using both DX11 and DX12 that you all might be interested in.

 

[Silverforce11]

I've put together some 980 Ti and 1070 benchmarks for ROTR using both DX11 and DX12 that you all might be interested in.

At this point I don't think it's valuable to judge DX12 with RotTR at all when it's consistently slower. I mean, how do you get slower MIN FPS in DX12 vs DX11? It takes special sauces to get the one result that should be much better on a lower overhead, higher draw call API as slower.

 

[Bacon1]

Read the comments?

Watch the vid? look at the actual fluidity of the game seen in the game?

I've been using DXtory as an overlay for months on DX12 games and it is correctly displaying

I've said multiple times that the Rise of the Tomb Raider benchmark is bad because it seems to take loading into account (minimums are totally wrong when comparing to overlays)

So either...

Msi Afterburner, Dxtory, PresentMon, Steam Overlay and others are all incorrect, or the benchmark is wrong.

I'm going with the benchmark is wrong, which is why I don't recommend using it for testing. Not to mention it doesn't have any gameplay related stuff happening, its just a fly by.

So, if you have any proof that DX12 overlays (Dxtory and others) are incorrect, please provide it.

 

[Cookie Monster]

I've been using DXtory as an overlay for months on DX12 games and it is correctly displaying

I've said multiple times that the Rise of the Tomb Raider benchmark is bad because it seems to take loading into account (minimums are totally wrong when comparing to overlays)

So either...

Msi Afterburner, Dxtory, PresentMon, Steam Overlay and others are all incorrect, or the benchmark is wrong.

I'm going with the benchmark is wrong, which is why I don't recommend using it for testing. Not to mention it doesn't have any gameplay related stuff happening, its just a fly by.

So, if you have any proof that DX12 overlays (Dxtory and others) are incorrect, please provide it.

Well it works both ways. If you have proof that its the ingame benchmark that is wrong, please feel free to provide the proof because i've yet to see any other than assumptions based on anecdotal evidence.

All i've said so far is that the overlays dont match with the benchmark numbers for the GTX1080 tested in that video. Could be a benchmark/software/driver problem. Im also assuming it works fine with other cards seeing as the same YT reviewer had no problems with other cards.

 

[dacostafilipe]

Using the canned benchmark, all settings maxed out (HBAO+ ambient occlusion in both DX11 and DX12) I get 128.58 fps in DX12 mode, 103.5 FPS in DX11 mode. Nice speedup. Using 1080 SLI.

Thanks, but I was talking about Async Compute, not DX11 vs DX12.

Could you maybe retest it?

 

[3DVagabond]

Thanks, but I was talking about Async Compute, not DX11 vs DX12.

Could you maybe retest it?

I don't think Async Compute is activated in nVidia's drivers yet, is it?

 

[dacostafilipe]

I don't think Async Compute is activated in nVidia's drivers yet, is it?

That's why I asked for the test in the first place

I just find it strange that a lot of people did Async on/off tests with AMD cards, but nothing on Nvidia.

This is an nVidia game, so it should show us how their Async version would improve performance.

 

[PowerK]

I don't see any reason to use DX12 rather than DX11 in this game. VXAO is such a nice improvement over HBAO+. For the best visual fidelity, DX11 is superior to DX12 in this game.

 

[Bacon1]

I don't see any reason to use DX12 rather than DX11 in this game. VXAO is such a nice improvement over HBAO+. For the best visual fidelity, DX11 is superior to DX12 in this game.

VXAO adds very little IQ for a huge performance hit. There is also no reason it wouldn't work in DX12 if Nvidia added support for it. Though you have 2x Titan XP so not surprised you would run VXAO.

 

[dogen1]

VXAO adds very little IQ for a huge performance hit. There is also no reason it wouldn't work in DX12 if Nvidia added support for it. Though you have 2x Titan XP so not surprised you would run VXAO.

Well, that's only because just about every other AO tech used is a complete hack, while VXAO isn't. And as a result VXAO avoids things like AO pop-in and has better temporal stability.

 

[ThatBuzzkiller]

VXAO adds very little IQ for a huge performance hit. There is also no reason it wouldn't work in DX12 if Nvidia added support for it. Though you have 2x Titan XP so not surprised you would run VXAO.

I'd argue otherwise since VXAO is one of the few scene geometry independent methods for approximating occlusion. VXAO is highly useful in cases for minimizing errors from image space methods like any other forms of "screen space" ambient occlusion (HBAO, HDAO, SSAO, etc) would exhibit. Not only is it noticeably more temporally stable but it also has more information to work with to increase accuracy ...

VXAO could work on any DX11 hardware but Nvidia's implementation is using driver extensions so it could be that they don't have a similar driver extension ready for D3D12 ...

That being said, I wish every developer would start adopting occlusion systems based on primitives much like how Naughty Dog approximates ambient shadowing in their latest games by approximating each mesh with sphere and ellipsoids representations ...

 

[antihelten]

That being said, I wish every developer would start adopting occlusion systems based on primitives much like how Naughty Dog approximates ambient shadowing in their latest games by approximating each mesh with sphere and ellipsoids representations ...

Something similar was also used in Sniper Elite 3 I believe, with their Obscurance Fields. At least they also used sphere approximations, and theirs wasn't a screen space solution either.

 

[Jackie60]

I've gone from 83.59 average fps with 39.77, 40.38 and 42 minimums to 111.18 average fps 42,17,48 minimums everything maxed but using HBAO on 4K with dual Titan XPs and 5960x at 4.6. A pretty impressive performance lift imho.

 

[TheRyuu]

Sorry but you give them too much credit.

They release yet another broken DX12 patch. From benches I've seen, minimum FPS tanks hard with Async Compute enabled. Looks like these guys just can't do DX12 right. Leave it at that. The more they try, the worse it gets.

Min FPS is the most useless stat ever. You may very well be right in that the performance tanks but it's actually useful to show that with frame time analysis not min fps.

 

[dogen1]

Something similar was also used in Sniper Elite 3 I believe, with their Obscurance Fields. At least they also used sphere approximations, and theirs wasn't a screen space solution either.

Yep, and it looked really good.

 

[PowerK]

FYI. PlayStation 4 Pro tech demo of Rise of the Tomb Raider.
https://www.youtube.com/watch?v=zIm27wrT4jo

 

[PowerK]

The game's pure eye candy. I love VXAO. I wish more games support this.
Also, TXAA would have been amazing with this game.

DX12 + VXAO + TXAA could be amazing.