Deus Ex: MD has poor low/midrange CPU perf. What Graphics options will ease the CPU bottleneck?

[hsjj3]

Just curious about this.

This is the best case scenario for the game currently in terms of CPU perf (DX 11 + Nvidia GPU)

http://gamegpu.com/images/stories/Test_GPU/Action/Deus_Ex_Mankind_Divided_/test/proz_11_nv.png

Take note, this is at 720p(!) only with maxed settings with a GTX 1080.

So I am asking, for those of us with a mid-range i5 or low-end i3, which graphics settings could likely reduce the CPU load and allow us to play at 1080p60fps, assuming we have a beefy GPU to go along with it?

List of graphics settings: http://www.dsogaming.com/news/deus-ex-mankind-divided-pc-graphics-settings-revealed/

Would appreciate any and all points. I know the game isn't out yet, but I am hoping that members of this forum will regardless be able to offer some help based on past experiences and on an educated guess basis.

 

[lyssword]

cloth physics, motion blur, maybe contact hardening shadows and shadow related things I think..

 

[PhonakV30]

If you want talk about CPU , Please Follow this

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/deus-ex-md-cpu-thread-scaling-and-performance.2483542/

If only graphic then here you go.

 

[NTMBK]

DirectX 12.

 

[DamZe]

DirectX 12.

That might just be the key to help the performance in CPU intensive scenarios of this game.

 

[Erenhardt]

Decrease level of detail to get better performance.

Also, force game to not use thread 0 - the one which runs amd driver. I tested this in couple of games and had very positive results when forcing games to use threads 2-5

 

[MajinCry]

When it comes to the CPU being held back, that's due to draw calls eating up precious CPU time. The answer's the same fer all games; disable settings that add draw calls.

A big offender that results in high draw calls, are reflections; water reflections are essentially another viewport, with objects being rendered multiple times (once for the player's viewport, again for the reflection).

Shadows as well; a draw call or two added for each shadow casting object, with more draw calls being called on the objects with shadows on top of them. Same gig with lights.

So try disabling these:

Contact Hardening Shadows
Parallax Occlusion Reflections,
Screenspace Reflections

Try reducing the field of view; should result in more aggressive frustum culling.

 

[hsjj3]

When it comes to the CPU being held back, that's due to draw calls eating up precious CPU time. The answer's the same fer all games; disable settings that add draw calls.

A big offender that results in high draw calls, are reflections; water reflections are essentially another viewport, with objects being rendered multiple times (once for the player's viewport, again for the reflection).

Shadows as well; a draw call or two added for each shadow casting object, with more draw calls being called on the objects with shadows on top of them. Same gig with lights.

So try disabling these:

Contact Hardening Shadows
Parallax Occlusion Reflections,
Screenspace Reflections

Try reducing the field of view; should result in more aggressive frustum culling.

Thanks for the reply!

When you say lights, would that include Volumetric lighting? What about Subsurface scattering?

Where does "Level of Detail" (assuming it is draw distance) rank in terms of importance?

For shadows, do you mean all shadows in general, or just contact hardening shadows?

Between all the factors listed, would you say that reflections>shadows>lights in terms of import?

 

[Stuka87]

Wait till September to play it, when the DX12 patch comes.

 

[Headfoot]

Thanks for the reply!

When you say lights, would that include Volumetric lighting? What about Subsurface scattering?

Where does "Level of Detail" (assuming it is draw distance) rank in terms of importance?

For shadows, do you mean all shadows in general, or just contact hardening shadows?

Between all the factors listed, would you say that reflections>shadows>lights in terms of import?

Regardless of that hierarchy, DX12 is going to make the single biggest difference in terms of CPU load for both vendors.

 

[hsjj3]

Regardless of that hierarchy, DX12 is going to make the single biggest difference in terms of CPU load for both vendors.

Sadly until today this is a myth for me. The two DX12+DX11 games out there, Hitman and Rise of the Tomb Raider, perform the same pretty much regardless of DX API. I do notice the CPU bottleneck is reduced, but it causes other parts to be taxing and on the overall there is no appreciable performance improvement.

 

[MajinCry]

Thanks for the reply!

When you say lights, would that include Volumetric lighting? What about Subsurface scattering?

Where does "Level of Detail" (assuming it is draw distance) rank in terms of importance?

For shadows, do you mean all shadows in general, or just contact hardening shadows?

Between all the factors listed, would you say that reflections>shadows>lights in terms of import?

Volumetric lighting is a post-processing thing; it relies upon the GPU.

Level of detail...Depends how the models are set up. If the LOD is just a less dense polygon mesh and a lower resolution texture, there's not going to be much difference in CPU performance. If the LOD, however, batches parts of the object together (e.g, instead of having separate hand, feet and torso textures + meshes, the LOD model has a single body mesh with a single texture), you'll get better CPU performance due to fewer draw calls being made to render that specific level of detail.

If you can disable shadows, those will free up CPU performance. But I could only see an on/off thing for contact hardening shadows. If those are another set of shadows cast by objects, that otherwise wouldn't be made if they were disabled, that should also reduce the draw calls being made.

It completely depends upon the scene, as well as the engine. For example, Oblivion will always re-render objects for the reflection shader, even if there is no reflective water in the rendered scene. Disable the reflections, et voila. Suddenly you get loads more fps (+15fps up from 30 with vanilla Oblivion, in my case).

But in Fallout 4, the shadow draw distances are rather far, and eat up the most draw calls; ~7.2k being called in Diamond City with shadow distance @ 8000 units. Reduce shadow distance to 1000 units, and the scene is making ~4k draw calls.

Ya have ta test for your specific game with those specific settings.

 

[Headfoot]

Sadly until today this is a myth for me. The two DX12+DX11 games out there, Hitman and Rise of the Tomb Raider, perform the same pretty much regardless of DX API. I do notice the CPU bottleneck is reduced, but it causes other parts to be taxing and on the overall there is no appreciable performance improvement.

Except you left out the game where it makes a massive difference, Ashes of the Swarm. Which has tons of draw calls and specifically uses DX12 to raise CPU limited FPS due to those draw calls.

Hitman and Rise of the Tomb Raider are probably two of the three worst DX12 implementations so I wouldn't use them as the baseline references for engines like Deus Ex's where they've designed for DX12 earlier in the process, and not as a patch after the fact.

 

[Jman13]

I saw a comparison a bit ago (not sure exactly where) that shows a HUGE performance jump backing away from Ultra preset to High. (from, say 40fps to around 85 fps), with fairly minimal difference in detail and effect in most situations. The Ultra preset starts with 4x AA, while high is 2x...that'll save a bunch, plus moving the shadow detail from very high to high. I think most people with mid-range cards and higher should do just fine with a mix of the high and ultra settings, and it'll look pretty darn good.

 

[TheELF]

First off you have to realize that this is the benchmark,where they make every effect found in the game run in a single scene.
Wait for the game to come out and you will see that it will run just fine on anything.

To get something out of vulkan/dx12 ,on the CPU side, you will have to use a CPU that is forced to run the driver together with an other thread on the same core AND a game where the graphics driver is not very optimized.
DOTA 2 made a lot of sense because it uses ~15% for the driver while top notch AAA titles barely reach 5% . (and you only get this 5% if dx12 is as optimized as dx11)
DOTA 2 vukan/dx11 side by side on a celeron.

 

[escrow4]

Buy an i7 or go hexa core. Games are only going to get heavier as consoles are increasingly multi-threaded. DX 12 is a crutch, sure it will eventually help, but what really helps is six cores hammering the game along. An i3 is pointless for AAA games, an i5 is barely enough, a hyperthreaded is a base and anything higher is cake.

 

[SPBHM]

for the ones with weaker hardware it's a valid alternative to use a 30FPS cap and enjoy a consistent experience on the level of performance the game was designed to run.

 

[TheELF]

Buy an i7 or go hexa core. Games are only going to get heavier as consoles are increasingly multi-threaded.

How are consoles going to get more multi-threaded?
Next gen is late 2017 at the earliest and will most probably feature an 4/8 ZEN,if my math doesn't fail me that's the exact same amount of threads.
That gen of consoles will last 3-4 years at least so if an i3 is good today it will still be good in 4-5 years.

 

[Carfax83]

I kind of agree with escrow4, in the sense that game engines are definitely getting more and more multithreaded due to implementing task parallelism models rather than the standard multithreading model. With task parallelism, the game can spawn as many threads as it needs to complete specific tasks, with several threads running on a single CPU core. In this case, having a CPU with a lot of cores and hyperthreading is definitely a bonus.

But I don't agree that consoles are getting more multithreaded. I think they're going to be stuck with 8 cores/threads or a while..

 

[Bacon1]

How are consoles going to get more multi-threaded?

I think he just meant game engines are, as they are designed console first.

 

[TheELF]

I kind of agree with escrow4, in the sense that game engines are definitely getting more and more multithreaded due to implementing task parallelism models rather than the standard multithreading model. With task parallelism, the game can spawn as many threads as it needs to complete specific tasks, with several threads running on a single CPU core. In this case, having a CPU with a lot of cores and hyperthreading is definitely a bonus.

But I don't agree that consoles are getting more multithreaded. I think they're going to be stuck with 8 cores/threads or a while..

Well if you have an infinite GPU maybe,keep in mind that console cores are jaguar cores that run at less then 1.5Ghz ,probably a lot less when all of them are working,consoles are completely dependant on splitting tasks up into multiple threads since all 6 of their available cores are probably slower than a single i3 core without the HT.
We already have most AAA titles filling up the console cores completely,there is no difference in how they get used the total amount of work stays the same and the i3 has shown that it is fine running such workloads.

 

[Erenhardt]

Well if you have an infinite GPU maybe,keep in mind that console cores are jaguar cores that run at less then 1.5Ghz ,probably a lot less when all of them are working,consoles are completely dependant on splitting tasks up into multiple threads since all 6 of their available cores are probably slower than a single i3 core without the HT.
We already have most AAA titles filling up the console cores completely,there is no difference in how they get used the total amount of work stays the same and the i3 has shown that it is fine running such workloads.

That is false.

In real world i3 struggles to keep even with slower of the consoles.

 

[Carfax83]

Well if you have an infinite GPU maybe,keep in mind that console cores are jaguar cores that run at less then 1.5Ghz ,probably a lot less when all of them are working,consoles are completely dependant on splitting tasks up into multiple threads since all 6 of their available cores are probably slower than a single i3 core without the HT.
We already have most AAA titles filling up the console cores completely,there is no difference in how they get used the total amount of work stays the same and the i3 has shown that it is fine running such workloads.

You're not considering platform overhead. Like Erenhardt posted, some games will run better on the consoles' weak CPUs as the platform overhead is much lower than it is on PC. Doom for instances uses OpenGL, and no matter how good of a job ID did with Doom, OpenGL's CPU optimization is not even close to the level that they can do with actual low level APIs like the PS4's GNM.

So whilst the Core i3's individual CPU core might be much faster than what's in the consoles, it still can't overcome the platform overhead in some games which leads to lesser performance. You see a similar thing in The Division, which chokes on dual cores without SMT even if they're clocked to over 4ghz because the game actually requires a certain amount of threads to run properly due to the task parallelism programming model it uses.

 

[TheELF]

So the celeron can run this game with ~60-70% total CPU usage at well above 30FPS,you can see how much the GPU bottlenecks the celeron in this game,but somehow the i3 is struggling with it?

 

[Kenmitch]

720p moves the bottleneck to the cpu. Why are you running it at 720p if you feel you have a weak cpu?