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

Given how long it took to go from DX9 to DX11, I'm guessing it's going to be a good long wait. Performance now is decent but Vulkan (mainly due to multi-threading the graphics engine better) should really open up performance in CPU limited situations.

Different situation, DX9 to DX11 was an entirely new shader model and infrastructure, likely a from scratch re-write.

DX11 -> DX12 or DX11 to Vulkan is likely significantly easier. (as far as I've heard from other developers, Vulkan is Easier to code for than DX12, and offers the same features and performance w/o the O/S Limitations)

DX9 to DX11 also took longer because it was announced before it was started, So from (Aug)2011 to (Sept)2015, with several breaks in there to develop other things.

For Example:

http://massivelyop.com/2017/03/20/star-citizen-will-drop-dx11-avoid-dx12-support/

Different situation, DX9 to DX11 was an entirely new shader model and infrastructure, likely a from scratch re-write.

 

DX11 -> DX12 or DX11 to Vulkan is likely significantly easier. (as far as I've heard from other developers, Vulkan is Easier to code for than DX12, and offers the same features and performance w/o the O/S Limitations)

 

DX9 to DX11 also took longer because it was announced before it was started, So from (Aug)2011 to (Sept)2015, with several breaks in there to develop other things.

 

 

For Example:

http://massivelyop.com/2017/03/20/star-citizen-will-drop-dx11-avoid-dx12-support/

Well that is at least encouraging! I really look forward to finally unleashing the full power of a PC on DCS

I rather hope ED takes their time with Vulkan to make it a a scalable and future proof implementation. We have seen how games that got dx12 slabbed onto it often performed barely equal to dx11 at best, often even worse. Afaik, the best Vulkan (and DX12) requires an extensive deep engine 'rewrite' or upgrade. I gather it will take big game engine providers many more months to get this right so those new APIs can deliver what they promise.

Hence; i'd rather see this done properly but coming later to DCS when the API is in a matured state!

And i hope this might get us one step closer for Linux gaming. Concerning that, my main conern lies with all that extra software like mod managers, SRS and what not. Even if the game itself runs fine on linux, the entire ecosystem we rely on needs to be supported as well.

the DX11 -> DX12 "Slap On" Compatibility most developers push out was just that though.

Took a DX11 Engine, and Converted quickly to basic DX12, but still used the same techniques and none of the new features, so while it's DX12, it's essentially still DX11 running on DX12 Libraries.

And the Only reason it happens and works is because DX12 has DX 11_0, 11_1 and 11_2 Compatibility modes build into it.

You cant do that w/ DX11 -> Vulkan, Vulkan does not have a DX 11_x Mode, so using the same code and techniques will not work.

I rather hope ED takes their time with Vulkan to make it a a scalable and future proof implementation. We have seen how games that got dx12 slabbed onto it often performed barely equal to dx11 at best, often even worse. Afaik, the best Vulkan (and DX12) requires an extensive deep engine 'rewrite' or upgrade. I gather it will take big game engine providers many more months to get this right so those new APIs can deliver what they promise.

 

Hence; i'd rather see this done properly but coming later to DCS when the API is in a matured state!

 

And i hope this might get us one step closer for Linux gaming. Concerning that, my main conern lies with all that extra software like mod managers, SRS and what not. Even if the game itself runs fine on linux, the entire ecosystem we rely on needs to be supported as well.

Here is an old post I found that's a bit in your face! WOW!

The question about effective use of multicore is of course a pretty legitimate and a hot one.

C++ code base of DCS is >1.5M lines of non-trivial and sometimes very complex code, which was mosty architectured in the single-CPU era.

The size of the team can be easily estimated using the about box of the game.

 

We are definitely moving to multicore, obviously it's the present and the future (if there's a future for PC's at all).

But DCS is our bread and butter and we just can't throw it away.

 

So the things are definitely moving, but not as fast as everyone, including us, wants.

The Post

Imagine if they they kept us updated! We are now optimizing lines 807,594 to 1,189,698 this will then allow us to move onto...:cry:

That's cRazy, 1.5 Million Lines!!

what is being refactored/rewritten does use multi-threading where possible.

 

atm the major parts which do it are:

resource loading and other I/O (logging, input, ffb ...), integrity checking, sound

 

w/o rewriting the engine, splitting simulation into a separate thread from the graphics will benefit it most - which will be, in fact, the mentioned client-server approach, just inside the app.

This makes the most sense and would stop a lot of asynchronous problems I would think (client-server approach)? This would also help setting up towards the dedicated server I'm guessing too.

There is still going to be some compromises, otherwise someone would have built a flight sim full multicore (How long would that take?) I mean even the large lockheed went down the upgrade path and not rebuild from scratch, that sim is much better, You still need a bit of a power house PC to get a real good experience with addons and eye candy.

I still think this will set DCS up for the future and for future hardware, it will still always be at the upper edge of system requirements going forward like all flight sims always are for good eye candy, especially in the future on the VR side, as Screen res increases. All this should allow ED to spread some of the strain around. I'm just now waiting for my 1440p / 4K VR and 2080 GPU to handle it and I'll be set!

.

1.5 Million seems small to me,

My app was nearly 20K Lines before I re-wrote it and condensed it.

I just wish ED would write detailed posts about that side of DCS...really go into detail (depending on what they want to make public record) how the engine is build, works and what issues they face or how they are trying to improve or change it with Vulkan. Or the eventual goal of a spherical planet earth...what they think might get them there.

Heatblur is the prime example of writing about techincal stuff concernig their projects in depth and exciting but comprehensible for the layman at the same time. Two indepth dev. blogs concerning the engine or the world a year would be totally sufficient.

More along the line; ''don't tell how long you think it might take, but show us what you have achieved already, and what is still missing and we can discern ourselves how long it might take in the end''!

1.5 Million seems small to me,

 

My app was nearly 20K Lines before I re-wrote it and condensed it.

Yeah, Wireshark has over 1.5 million lines. So it seems low for DCS.

Complex topic but…

Vulkan seems to improve the CPU's job part … so no improve to expect with VR in DCS ? (as in VR my CPU is barely used while my GPU is near full usage with only 45fps to render with ASW rendering the rest of the 90fps ?)

Or will Vulkan ease the GPU's job, by a mean or another ? And so be a help for VR/DCS ?

At a time I heard here something like 50 or 60% increase perf (on one of first announcements about vulkan). Using my screen (low res) with max settings I can see CPU being as highly used as my GPU, on or near max, so I can guess Vulkan will improve things … but what about VR ?

(an exemple of high CPU usage in DCS with screen on this afterburner monitoring screen :

https://forums.eagle.ru/showpost.php?p=3413978&postcount=28 )

Complex topic but…

Vulkan seems to improve the CPU's job part … so no improve to expect with VR in DCS ? (as in VR my CPU is barely used while my GPU is near full usage with only 45fps to render with ASW rendering the rest of the 90fps ?)

Or will Vulkan ease the GPU's job, by a mean or another ? And so be a help for VR/DCS ?

 

At a time I heard here something like 50 or 60% increase perf (on one of first announcements about vulkan). Using my screen (low res) with max settings I can see CPU being as highly used as my GPU, on or near max, so I can guess Vulkan will improve things … but what about VR ?

 

(an exemple of high CPU usage in DCS with screen on this afterburner monitoring screen :

https://forums.eagle.ru/showpost.php?p=3413978&postcount=28 )

How exactly are you determining utilization? Because the only thing that really causes high GPU utilization is using AA with DS, which is the result of some inefficiency, or really high pixel density. Without either of those the Caucasus in VR (F-15 Intercept instant action) will get you only 45 FPS on all low settings and pixel density on 1.0 seeing only 50% gpu utilization average. CPU utilization as a whole looks low for multi-core CPUs anyway. You can crank setting up to the point of GPU saturation for "free" because the CPU is the hangup not being able to get render requests to the GPU inside of the 11ms window for 90FPS. A campaign in Nevada (on low) will see 22FPS on the ground for only 25% GPU utilization reaching 45/40 at times while taxiing. Nevada is pretty sparse so it will reach 90 in the air hitting 50% GPU. I mostly play multiplayer on the Caucasus, but my GPU is never maxed while only giving 45FPS. The point at which you can feed the GPU without waiting on the CPU should greatly benefit VR.

Having the simulation on one core and the graphics engine on another @ 5GHz will be huge! Vulkan then can push with just the graphics engine I'm guessing on other cores if available? This of course depends on how this is all coded and I'm sure ED has done their own days of research on the topic before doing anything. VR is not going anywhere and fits well with commercial VR training market moving forward and because I want the best experience doing laps of the boat in my F/A-18 @ 1440P / 4K VR headset :D

Quotes

2016 Developer Nvidia Getting Vulkan Ready For VR

"Naturally, anything related to graphics APIs involves graphics-drivers, and this is where NVIDIA is contributing by collaborating with our partners in the VR ecosystem. We specifically explored the design space of interoperability between Vulkan and other APIs as well as sharing resources between the application process and the process the HMD runtime is running in. We are now shipping this functionality as a set of Vulkan extensions that allow cross-process resource sharing and interoperability between Vulkan and other APIs. "

New Sets of Extensions

"The sharing set provides the ability to share memory and synchronization primitives across process and instance boundaries, which are useful among other things for implementing VR runtimes:"

"The explicit multi-GPU set adds support for treating multiple GPUs as a single logical device, so that applications can easily make use of multiple GPUs in a single application for use cases such as Alternate Frame Rendering, Split Frame Rendering or VR SLI:" :huh:

Link

End Quote

VERY INTERESTING!! Well I wonder if ED is going down this path at all? Should I start saving now for 2X GPU's or More? :D

More cores + Vulkan + Multiple GPU's ??? That would be WOW.... 4K VR here we come! "I hope"

https://www.khronos.org/news/press/khronos-reveals-api-updates-new-workgroups-at-gdc

New Sets of Extensions

Getting Vulkan Ready For VR

Side note: I've also been playing around with this app call Hello V for up coming F/A-18 VR training etc, This beta app creates a web browser console in other VR titles.

It works better when you just pause the sim as you lose your joystick controls when using the oculus hand controllers with it's GUI.

I updated an old post about this HERE in the VR section and included a video.

-

Hi Blkspade - I think I understand well what you say (but not certain…), but I don't agree with you, GPU being the bottleneck in DCS, and the great one in DCS VR, there is no relation between GPU usage (% GPU load in afterburner f.e.) and CPU bottlenecking.

The GPU usage is firstly determined by the "Frames per second" being capped or not :

If FPS are not capped (screen + vsync off f.e.) than GPU usage in afterburner is 100% all the time, whatever settings in DCS, AA, DS, or resolution.

If FPS are capped (screen + vsync on, or Rift 90fps, or 45fps forced) than GPU usage can"t be 100% as long as it can render more FPS than the cap limit, if below the cap limit it will be 100% with bad FPS under the capped value.

The only way to have 22FPS with 25% gpu usage, is to have a CPU core hitting 100% at this moment, or software limitation causing "low gpu usage" (it was the case with 1.5 caucasus) or an external software causing that (like Ge force experience at a time with some games). I don't have 22FPS with GPU or CPU not being at 100%. (only in cases of "low usage" wich is a software "bug")

Of course when the FPS are capped a high GPU usage is mainly a matter of settings : as high as possible to have best looking game without hitting 100% and loosing FPS from capped value. That's it, but no relation with CPU being a bottleneck.

Edit : that said, I've not tried the mission you talk about with 22fps/25% GPU usage but I will, cause I find it weird.

David OC, very nice info, but sadly VR sli or multigpu (vulkan) don't seem to be a "soon" help in DCS VR, as not usable by actual VR techs (HTC, oculus, and others).

So I'm wondering if Vulkan alone will be a real improvement in DCS VR as it will certainly be (+50 / 60% ?) with usage of a screen ? (and as ED only talks about Vulkan integration, and not about VR Sli or multi-gpu)

I know by experience that in a different environnement (a CPU bound flight sim, p3dv4 ), improvements about CPU work/load gives BIG improvements in VR's FPS (allowing an increase of settings while keeping 90fps flat, pixel Density at 1.75) . But what about DCS wich is, from my point of view, GPU bound ?

David OC, very nice info, but sadly VR sli or multigpu (vulkan) don't seem to be a "soon" help in DCS VR, as not usable by actual VR techs (HTC, oculus, and others).

 

So I'm wondering if Vulkan alone will be a real improvement in DCS VR as it will certainly be (+50 / 60% ?) with usage of a screen ? (and as ED only talks about Vulkan integration, and not about VR Sli or multi-gpu)

 

I know by experience that in a different environnement (a CPU bound flight sim, p3dv4 ), improvements about CPU work/load gives BIG improvements in VR's FPS (allowing an increase of settings while keeping 90fps flat, pixel Density at 1.75) . But what about DCS wich is, from my point of view, GPU bound ?

They are making grounds on bring some type of standard together it seems. Steam and oculus are in there...

The Time for Standardization is Now

VR SLI

I think you will see a big difference moving to Vulkan 50%? Maybe?

The way I understand and see it, just because you see how demanding it is now doesn't mean it will be with Vulkan, the overall throughput will be much more efficient and faster, if coded correctly to suit the engine of course. Like I said we know the graphics engine with be separated, this should allow Vulkan to go nuts with your extra cores and better utilization of the GPU with much less latency in the total system.

It will just allow ED to punish your system utilizing the hardware we have now and future hardware to come. You got 10 cores? Well have I got some work for you to do (That's when your GPU poo's itself) this would be handy also if VR sli comes into the mix also. This all should balance the system nicely and have your 1080ti CPU all working much more efficiently and effectively.

Only my 2 cents, lets wait and see, I'm pretty hopeful and the only way to power the next 2nd or 3rd gen VR in DCS. We are going to need 10 cores and VR sli to power those things.

-

Hadn't really considered the opportunities for moving to a multi-GPU setup (probably not appropriate to call it SLI, as it won't be), but yes, that's a good add on for Vulkan and for many games players could be a seriously interesting (if expensive) avenue to go down.

I can guess Vulkan will improve things … but what about VR ?

Based on my *very* limited understanding of how multi-core works, some time spent talking about the subject with a friend of mine who was part of the Khronos-group, and my own observations in DCS after playing around with various settings, I'd say yes: we'll see a substantial performance increase in VR (provided the implementation of Vulkan is done properly - but I don't see why that would not be the case).

I noticed a substantial performance boost after overclocking my (unlocked) CPU from the stock 4 GHz to 4.5 GHz (air cooled).

Given that multi-core really benefits scenes that render a ridiculous object count (draw calls), flight sims are the ideal target for Vulkan. And since VR renders the very high draw call amount twice (one for each eye), I'd expect the performance gain in VR to be larger than the one in non-VR.

But that's just me... I'm a biochemist, not a programmer.

Examples:

Hadn't really considered the opportunities for moving to a multi-GPU setup (probably not appropriate to call it SLI, as it won't be), but yes, that's a good add on for Vulkan and for many games players could be a seriously interesting (if expensive) avenue to go down.

I know I would pay if 4K VR DCS was on the table.;)

-

Got it so it seems that Vulkan will increase the whole system efficiency. Not only focus on CPU's work.

That gives possibility to gain nice perf amount in DCS VR.

I'm confident in what ED will do, multi-gpu or not with vulkan, but Elysian Angel the exemples you give show well how more efficient the whole system is with Vulkan.

Got it so it seems that Vulkan will increase the whole system efficiency. Not only focus on CPU's work.

That gives possibility to gain nice perf amount in DCS VR.

 

I'm confident in what ED will do, multi-gpu or not with vulkan, but Elysian Angel the exemples you give show well how more efficient the whole system is with Vulkan.

Another a way to put it, that does happen where I live and is very frustrating:mad:.

There's to many cars merging onto the highway stuck in only 2nd gear, when the rest of the cars are in 4th or 5th gear and they have to slam on the brakes because of these annoyingly slow cars. Your just using massive brute power (OC + 1080ti, fast ram, m2) with your system to make the cars stuck in 2nd gear go a little quicker that helps them try and match the other cars.

I was really nerding out the other day

The big thing I got from it was machine coding, the low level ability for more control of how the hardware operates. It's why programming languages such as C++ are used in flight control computer systems, there is a latency limit for things like this obviously, we cannot have a delay in the mechanical surfaces because the system ran out of ram.;) C++ can give you much more control over of how the system will handle the tasks.

These control systems are in real time so they need to be super fast and the code to never deteriorate throughout the system. Hopefully Vulkan can help ED get closer to this with their massively complex simulation with all these aircraft systems in it with there own virtual FBW systems.

-

There's a comment in today's update, stating that they plan to have Vulkan available by the end of 2018. Again, doesn't mean that on day 1 it'll be fabulous. However, ED have a strong track record of ensuring that things WILL be improved, so I'm very confident that over time it'll bring good things.

I haven't been on the forum in a while actively, but I have been playing this for decades. You are correct that they do steadily improve and make it better.

There has to be something going on in the latest release, testing or tweaking going on here? I was trying to record a video while flying and was getting crashes to desk top, one even shutdown my pc. I looked a process lasso and my 4 cores (I had DCS lock to the main cores in process lasso) where at 100% when recording!

So a quick test turning all 4 cores and 8 threads back on fixed it...

This is a shot without the recording software running, don't recall DCS using all the threads like this.

After takeoff and flying for a bit.

So you are saying it's time to enable hyperthreading again?

So you are saying it's time to enable hyperthreading again?

not really, actually, we are using ProcessLasso from bitsum.com to avoid hyperthreading but force DCS and all derived processes onto physical cores only. While we do that, we can see how many cores are actually at work, how much load on each, how often they spike, what scene makes them cores spike, what scenes kill fps and how does the GPU CPU look when the fps goes down..etc...

That's what MSI Afterburner is for, knowing what your rig does at any given time to understand why things are as they are and where to change things to make it better.

For example, online, just watch the graphs, they'll tell you when you are ready to get a slot, if you have no graphs or fps counter, you'll be clueless when the rig is done loading MP and you end up clicking and interfering an already busy system.

But who knows what threading possibilities the devs have and use. It could well be that some of the code does already use or -like- hyperthreading, I simply do not know if that is true..or planned..in the pipe..etc.. it for sure was cool if it would work and help distributing load evenly.

Flying a Quick Mission is a totally different thing compared to flying MP at crowded servers with complex missions. Your RAM usage is way higher, the CPU works harder, overall, it's another level of throughput the machine has to sustain. Also LOD makes a difference. LOW vs MAX is like Hell and High Water. There are many knobs and dials that influence the load you force onto it, resulting in different fps.

My hope is that Vulkan delivers better graphics at even less system load and that ED does it's best to code with threading and MultiCore CPU's in mind.

It would mean less expensive systems could run DCS very well, which is sadly not the case atm.

We are going to fly over crowded cities, Dubai, Las vegas and who knows what other cites on other maps. I kept dreaming of this for decades, ever since I was promised this:

Its not a bad thing, you can always turn detail down on your side instead of ED doing it. The era of quad cores is over. We have been asking ED to go multicore for many years, we are getting it (albeit only for rendering purposes for now).

^^ YAY loved Strike Commander, was thinking about the detail in the cities and thought of this game, oh far we have come :thumbup:

P.S: Eagle Dynamics, you have really excelled with this terrain module you should all be immensely proud of yourselves.