[MERGED] DirectX 11/12, OpenGL, VulkanAPI Discussion

[chp]

"Vulkan has demonstrated to be very efficient in High Object Count Rendering Scenes (ie Flight Sims w/ Large View Distances), DirectX12, Not So Much"

 

Tiled Recourse of DirectX 11.2

(2013)

Bundles and other features "free GPU" features of DirectX12

(2014)

..and recent DirectStorage https://devblogs.microsoft.com/directx/directstorage-is-coming-to-pc/ (2020)

all of them seem to be in favor of "large distance" sims.

 

"Choosing DX12, would Shut Out Windows 7 and Windows 8 Users, Vulkan would allow those users to continue to use DCS."

 

Vulkan might be a faster and more "rebel" way to go. But choosing "Linux/Tizen" over much needed flow of cash from XBox (potentially)...

 

Wish a great success to Devs in either path they had or will choose.

[BitMaster]

Would Vulkan increase my FPS when I am already at 99% GPU ? NO, it cannot do wonders !

 

Would Vulkan take some load of the "1" core that basically runs DCS and thus make my minimum FPS be higher ? Likely Yes imho.

 

THAT is what Vulkan can deliver to us, free up the main core so it can do MORE DCS and LESS GPU. Let other cores do the work and let the 1 core take care of DCS in a better way.

 

That is what I think Vulkan will be for DCS

[eatthis]

whats the problem when it comes to multicore cpus? why dont games spread workloads around the cores?

[Taz1004]

whats the problem when it comes to multicore cpus? why dont games spread workloads around the cores?

 

Because multithread programming is not easy. You have to be careful on how you distribute the task or you can get into threads trashing each other. Vulkan can distribute graphic side of it but bulk of DCS CPU cycle is AI, weapons, and avionics which can be very challenging to separate. If those three are distributed to 3 threads, AI will have to wait for other threads to complete before moving on to next task for example. Synchronization.

 

This is all complex enough for typical game releases. But for DCS which is constantly changing, complexity of updating it will be multiplied by the number of threads. Higher upkeep. Then the 3rd party developers need to adhere to same.

 

If ED implements Vulkan, it'll probably only help graphic side and maybe see about 20% improvement. But all other CPU task are still likely to be on single thread.

Edited October 11, 2020 by Taz1004

[eatthis]

Because multithread programming is not easy. You have to be careful on how you distribute the task or you can get into threads trashing each other. Vulkan can distribute graphic side of it but bulk of DCS CPU cycle is AI, weapons, and avionics which can be very challenging to separate. If those three are distributed to 3 threads, AI will have to wait for other threads to complete before moving on to next task for example. Synchronization.

 

This is all complex enough for typical game releases. But for DCS which is constantly changing, complexity of updating it will be multiplied by the number of threads. Higher upkeep. Then the 3rd party developers need to adhere to same.

 

If ED implements Vulkan, it'll probably only help graphic side and maybe see about 20% improvement. But all other CPU task are still likely to be on single thread.

 

so they have to manually assign tasks to seperate cores? if somebody could automate that then surely it would be a HUGE breakthrough performance wise??

[Taz1004]

so they have to manually assign tasks to seperate cores? if somebody could automate that then surely it would be a HUGE breakthrough performance wise??

 

If somebody can automate that, they would make a lot of money.

[speed-of-heat]

so they have to manually assign tasks to seperate cores? if somebody could automate that then surely it would be a HUGE breakthrough performance wise??

 

sure, but you still have to have them interact, which means they still have to make sure that they are doing so at the right time... a trivial example if the "sound thread" is out of synch with the game engine .. you might get a super sonic boom after or before you actually go super sonic ... now multiply that by every AI interaction that could take place with the 3D world...

[Harlikwin]

sure, but you still have to have them interact, which means they still have to make sure that they are doing so at the right time... a trivial example if the "sound thread" is out of synch with the game engine .. you might get a super sonic boom after or before you actually go super sonic ... now multiply that by every AI interaction that could take place with the 3D world...

 

Yup, thats the core issue. Synch... And it would likely be even worse in the MP case where you already have synch issues.

 

I frankly don't expect miracles from Vulkan, but I'd be happy with a 25-50% improvment in my CPU frametimes. But then again, most of the "new stuff" will eat that performance right back up IMO.

[eatthis]

sure, but you still have to have them interact, which means they still have to make sure that they are doing so at the right time... a trivial example if the "sound thread" is out of synch with the game engine .. you might get a super sonic boom after or before you actually go super sonic ... now multiply that by every AI interaction that could take place with the 3D world...

 

hows that done now?

synchronization has been a thing in electronics forever

[stormrider]

I frankly don't expect miracles from Vulkan, but I'd be happy with a 25-50% improvment in my CPU frametimes. But then again, most of the "new stuff" will eat that performance right back up IMO.

 

 

Isn't a mere 25% improvement already a miracle for DCS?! 50% is beyond any wild expectations. I'm thinking about ~15% performance increase, using X-Plane 11.50 as a reference and I would already be very happy with that. Means that I could finally have 90fps VR stable in my CV1, with reasonably mid/mid-high setting with any prev (20-series) or next gen (30-series or rx-60xx) GPUs.

 

 

The AI remains the problem tho. They must figure it out sooner than later, both improving it's performance impact and its shoddy behavior.

[speed-of-heat]

hows that done now?

synchronization has been a thing in electronics forever

 

Normally you use a scheduler to do it, this is not simple, and and most manage 2 or three processes at a time...let’s see how many do we have today...

[Polo]

hows that done now?

synchronization has been a thing in electronics forever

 

Synchronization with multicore processors is relatively easy, however it is also very easy to get into a situation where you have every other core waiting on one to finish processing something before they can begin their task(s). End result is you may end up actually slower, as the tasks are being run in sequence plus you have synchronization code adding extra overhead.

 

Loading up multiple cores consistently and efficiently at the same time is extremely complicated, and probably requires you to design that way in the beginning.

 

Here is an old article coding for multicores on the XBox360, but it does cover alot of the issues and what is involved:

 

https://docs.microsoft.com/en-us/windows/win32/dxtecharts/coding-for-multiple-cores

[BitMaster]

I don't expect ED to achieve heavy multicore usage until there is a total rewrite of the code.

After following this topic for almost a decade now I had to learn it's not that easy at all and the bigger, older, more complex the code is that needs to be changed the less chances you have to make it in such a way that it helps more than it brings new issues on the table.

 

My hope is that Vulkan frees enough ressources on the pounded core so that it buys enough time to make this happen way down the road while also increasing processing immedeatly after Vulkan becomes active in DCS. That is my assumption what is gonna happen in the next 1-2 years. I don't expect multi-processing for the core until way after 2025. I hope we are all still alive then :smilewink:

[OnlyforDCS]

I don't expect ED to achieve heavy multicore usage until there is a total rewrite of the code.

 

This pretty much. So if they are doing a total rewrite, they might as well call it DCS World 2 and be done with it.

[SkateZilla]

This pretty much. So if they are doing a total rewrite, they might as well call it DCS World 2 and be done with it.

 

except we are already on DCS World 2.5....

[stormrider]

except we are already on DCS World 2.5....

 

 

Well then, DCS 3.0 and charge an upgrade fee there, I'm in.

[OnlyforDCS]

except we are already on DCS World 2.5....

 

Yeah, well I meant DCS World 3. In any case, the switch to EDGE was 2.5, however most of the promised upgrades that we were supposed to get with 2.X.X other than new maps never really materialized. It's pretty clear that there is a big problem with the complexity of this sim and that the legacy code can't really handle everything thrown at it, hence the huge and ever rising PC requirements. I shudder to think what will happen if Eagle Dynamics ever releases the long overdue weather and ATC upgrades.

[stormrider]

I shudder to think what will happen if Eagle Dynamics ever releases the long overdue weather and ATC upgrades.

And AI and Radar/EW. It's a necessity tho.

[Silver_Dragon]

Yeah, well I meant DCS World 3. In any case, the switch to EDGE was 2.5, however most of the promised upgrades that we were supposed to get with 2.X.X other than new maps never really materialized. It's pretty clear that there is a big problem with the complexity of this sim and that the legacy code can't really handle everything thrown at it, hence the huge and ever rising PC requirements. I shudder to think what will happen if Eagle Dynamics ever releases the long overdue weather and ATC upgrades.

"promises"

 

With 2.5 ED get Dx11 engine, updated Caucasus and NTTR update, Persian Gulf map and updates, new tech used on Channel map, some kind of base updated building (talked by RAZBAM), underwater maps plus 3rd party TDK access, and special effects. Vulkan has in progress as DC, weather engine, more special effect and ED talked about base ATC. DM and AI updates has around the corner and more to coming.

 

Enviado desde mi RNE-L21 mediante Tapatalk

[OnlyforDCS]

If we had 15 Ghz processors, then maybe, just maybe we could get those goodies Silver mentioned. More and more I'm convinced that the problem with the delays to all the features that were supposed to come after EDGE is not the implementation, but how to get them incorporated into DCS without turning out PCs into potatoes.

[Voyager]

An observation I found while I was debugging my laptop is apparently Dx12 and Vulkan can handle something like 3x the draw calls as Dx11.

 

On the radar re-work, at this point I'm thinking it would be best for the devs to implement it via ray-tracing rather than a pure CPU process. We should have a fairly solid ray-tracing support come in over the next year or so, and it should be able to run it much better, and in theory, more simply, than a pure CPU based solution.

 

Also, I wouldn't expect the graphics engine to be ready to take advantage of ray tracing for some time.

[Pilotasso]

They can put the radar code on separate core(s) rather than shove it into the GPU and kill the FPS. Also not everybody has ray tracing yet.

[VirusAM]

They can put the radar code on separate core(s) rather than shove it into the GPU and kill the FPS. Also not everybody has ray tracing yet.

 

They are using ray-casting...it in an alghoritm

[stormrider]

On the radar re-work, at this point I'm thinking it would be best for the devs to implement it via ray-tracing rather than a pure CPU process. We should have a fairly solid ray-tracing support come in over the next year or so, and it should be able to run it much better, and in theory, more simply, than a pure CPU based solution.

 

Also, I wouldn't expect the graphics engine to be ready to take advantage of ray tracing for some time.

 

 

This means that they would have to re-write the game requirements page and make minimum GPU, a RTX. Even a 1080 ti would become unsupported at this point. Honestly, although I'd love to see this kind of technological improvements in this game, I don't see this happening anytime in the next 3 years.

 

 

https://www.digitalcombatsimulator.com/en/shop/modules/

 

[b]Minimum system requirements (LOW graphics settings):[/b] OS 64-bit  Windows 7/8/10; DirectX11; CPU: Intel Core i3 at 2.8 GHz or AMD FX; RAM:  8 GB (16 GB for heavy missions); Free hard disk space: 60 GB; Discrete  video card NVIDIA GeForce GTX 760 / AMD R9 280X or better; requires  internet activation. 

[einarabelc5]

Geez some of the code is from before Flaming Cliffs, that's code that's over 20 years old. Ever heard of Lock On? I might be wrong but based on the behaviors and control mappings of Flaming Cliffs 3.0 in DCS that seems to be the case.