ComputerBase & GameGPURise of the Tomb Raider: DX11 vs DX12 + VXAO Tested

[Sweepr]

An upcoming patch for Gears of War Ultimate Edition PC will be improving the game’s performance, and will allow for ambient occlusion tweaks. Developer The Coalition announced the PC update through their official forums recently. According the game’s technical director, the update is scheduled for a release in the upcoming week. These are the key features for the upcoming patch:

Performance optimizations
Refactor of texture settings to better reflect VRAM requirements and to better manage the texture levels for each setting.
Ambient Occlusion tweaks.
Tuned mouse coming out of roadie run.
Disabled mouse smoothing.

https://gearsofwar.com/en-us/forums...11/ce4e8579-c73b-49df-8a7d-cc76de6d83c4/posts

 

[Qwertilot]

Oh, look. I didn't say that this a shining example of what you can do with this stuff in an ideal world, but I do think it might be quite representative of actual results.

You don't get the extra performance by magic! No doubt some frills, but basically:
a) helping CPUs out and
b) letting people do low level GPU optimisations.

This implementation seems to do the CPU bit just fine. That alone makes it useful I think. Gaming laptops if nothing else.

The low level GPU work? Great for consoles of course. For a PC game releasing now?

You've got to optimise for 3 AMD architectures, a few NV ones and several from Intel (which, like it or not, are market relevant). And you then have to redo it all over again when the next gen cards launch in 3-6 months

Do you really think that the people responsible for the recent PC ports are going to do all of that? Or anyone sane actually.

Ashes seems to be trying but they're getting huge publicity from it all, so it makes sense for them.

 

[Sweepr]

This implementation seems to do the CPU bit just fine. That alone makes it useful I think. Gaming laptops if nothing else.

Good point. Someone should try to run these new titles at a locked framerate and compare power consumption (DX11 vs DX12).

 

[3DVagabond]

It's not just performance enhancements.


http://www.geforce.com/whats-new/articles/geforce-gtx-is-game-ready-for-windows-10-and-directx-12

Be a bit more specific. I'm not reading an entire article and then trying to figure out your point.

 

[omek]

It looks good but a 20%-30% performance hit from HBAO+ to VXAO? I think a voxel based solution is pretty hardcore for an effect like AO.
I'd rather have proper GI with this technique.

Yep. This is taking ambient occlusion to a stupid level, it's like using crayons to draw a 5 when asked what 2+2 equals. It's somewhat pretty but you still fail.

 

[showb1z]

Damn, looks like the NDF was so frustrated over recent benchmark results they blew their load all over this sorry excuse for a DX12 patch.
This obviously needs some more work. Nixxes is usually pretty competent but it's their first go at DX12 so I guess it's to be expected.

 

[flynnsk]

Be a bit more specific. I'm not reading an entire article and then trying to figure out your point.

LOL best part of that page, nV's "key features" are primarily optional features to get to Feature Level 1 AND they include Async Compute (btw, how is that coming along ?! )

 

[Bacon1]

Be a bit more specific. I'm not reading an entire article and then trying to figure out your point.

I think he meant Async Compute, since its stated right there as an D3D 12 API feature that they support D:

 

[Silverforce11]

VXAO vs HBAO+

See if you guys can tell the difference, and whether it's worth the 20-30% performance loss.

Which one is HBAO+ and which one is VXAO, #1-4?

http://www.computerbase.de/2016-03/rise-of-the-tomb-raider-directx-12-benchmark/3/

1.

2.

3.

4.

 

[2is]

Based on the other photos, my guess is:

VXAO = 1,3
HBAO = 2,4

 

[boozzer]

from the picture sets, the top ones are a tiny, tiny bit darker in the shadows. is that all the patch does for dx12?

 

[Silverforce11]

Based on the other photos, my guess is:

VXAO = 1,3
HBAO = 2,4

Nope!

Hard to tell but you will definitely notice a 20-30% performance loss.

 

[Silverforce11]

from the picture sets, the top ones are a tiny, tiny bit darker in the shadows. is that all the patch does for dx12?

No, it's DX11 VXAO vs HBAO+, NV hasn't implemented it for DX12, yet.

And ps. SSAO and HBAO+ applies to vegetation just fine, the GIF is a fake AO OFF vs VXAO mislabeled to exaggerate VXAO.

 

[Head1985]

Hey i have few more benchmarks in CPU bottleneck scenario.I underclock 6700K to 2.5Ghz and run some benchmarks in most CPU intensive area-geothermal valley village.

I dont know if its 6700k just that super fast, but even at 2.5Ghz a have 60fps in worst CPU bottleneck scene..i read village is nightmare for cpu but not much to 6700k.Still i needed lower resolution to 1280x720 to be CPU bottleneck even at 2.5Ghz with GTX970 1500/8000
Test system
6700k 2.5Ghz/2.5ghz cache
3000Mhz DDr4
GTX970 1500/8000

Anyway results
1920x1080/1280x720 all max + low textures and HBAO+
First fraps 60s run
1920x1080
DX11 Avg: 59.103 - Min: 49 - Max: 80
DX12 Avg: 59.641 - Min: 49 - Max: 78

1280x720
DX11 Avg: 76.913 - Min: 53 - Max: 125
DX12 Avg: 84.331 - Min: 70 - Max: 126

CPU bottleneck scene screenshot at same place
DX11 58.6Fps

DX12 77fps

Under dx12 there is 31% FPS increase on test screenshot.But if you have good CPU DX12 is pointless.

 

[Sweepr]

Hey i have few more benchmarks in CPU bottleneck scenario.I underclock 6700K to 2.5Ghz and run some benchmarks in most CPU intensive area-geothermal valley village.

I dont know if its 6700k just that super fast, but even at 2.5Ghz a have 60fps in worst CPU bottleneck scene..i read village is nightmare for cpu but not much to 6700k.Still i needed lower resolution to 1280x720 to be CPU bottleneck even at 2.5Ghz with GTX970 1500/8000

CPU bottleneck scene screenshot at same place
DX11 58.6Fps
DX12 77fps

Under dx12 there is 31% FPS increase on test screenshot.But if you have good CPU DX12 is pointless.

Thanks for testing. If you don't mind I will add these results to the OP.
That's good news for Core 2/Nehalem/SB users if you're on Intel.

 

[Head1985]

I will ADD no HT results later

 

[Sweepr]

2.5GHz i7-6700K HT on/off with fast DDR4 is probably close to i7-2600K/i5-2500K stock performance, and considering the amount of gamers still using these chips your results are relevant.

 

[Head1985]

With HT Off i see +1-2Fps increase under dx11 and same FPS under DX12
DX11-61
DX12-61

Btw this is first game i ever seen(With World of tanks) that didnt increse performance with HT under dx11.
Skylake have improved HT that increse performance in most games around 30-40% under dx11.Here pretty much not.
Only under dx12.
Here is my HT test when i tested 6700k.
http://www.overclock.net/t/1578480/i5-2500k-4-5ghz-vs-6700k-4-5ghz-in-games/10#post_24549601

 

[prtskg]

The game is an OK implementation of dx12. It helps weak cpus but that's it. It hasn't been optimized for performance improvement.

Some people are writing that in dx12 architecture specific optimization is required to improve performance. While low level apis gives capability to do so, it's not the only way to improve performance. Generic optimization and use of features that improve performance is better way. Oxide's Ashes is the perfect example of this where all gpus gain performance.

 

[JDG1980]

You've got to optimise for 3 AMD architectures, a few NV ones and several from Intel (which, like it or not, are market relevant).

I'm still not convinced that the issues with Tonga and Fiji are related to GCN 1.2, rather than these chips simply being crappy, unbalanced stopgaps.

I'd be interested to see tests in these poorly-optimized DX12 games for Kaveri (GCN 1.1) versus Bristol Ridge (GCN 1.2) when the latter is released in a couple of months. Both have the same number of shaders (512) and the same memory bandwidth bottleneck (assuming you match up RAM speeds on the boards), so it should be a pure architecture comparison.

 

[sontin]

Some people are writing that in dx12 architecture specific optimization is required to improve performance. While low level apis gives capability to do so, it's not the only way to improve performance. Generic optimization and use of features that improve performance is better way. Oxide's Ashes is the perfect example of this where all gpus gain performance.

No, this is just wrong. If you dont use architecture specific optimizations then you will never tap the whole potential of the hardware.

The driver is doing less so most of the work must be done by the application.

 

[Krteq]

What do you mean by "architecture specific optimizations"?

 

[sontin]

Here: http://www.dualshockers.com/2016/03...-nvidia-and-amd-cards-lots-of-details-shared/

For example:

Place constants, CBVs, SRVs and UAVs directly into the root signature if possible on NVIDIA Hardware

• Start with the entries for the pixel stage
  • Constants that sit directly in root can speed up pixel shaders significantly on NVidia hardware – specifically consider shader constants that toggle parts of uber-shaders
  • CBVs that sit in the root signature can also speed up pixel shaders significantly on NVidia hardware

https://developer.nvidia.com/dx12-dos-and-donts

 

[Krteq]

Oh, you mean resource binding limitations on nvidia hardware. Yes, that's unpleasant

 

[Erenhardt]

Oh, you mean resource binding limitations on nvidia hardware. Yes, that's unpleasant

Ouch!:thumbsup: