VR Performance in DCS World

[SkateZilla]

 

I really wish ED would do something about their thread management, it's crazy that it can't utilize high-end hardware properly. Reminds me of Arma.

 

 

Making DCS Multi Threaded will not resolve the DirectX API getting bogged down by object counts, translating into lower % of draw calls being processed per frame, translating into lower gpu usage while GPU waits for commands, translating into more GPU Cores sitting in standby mode waiting for instructions, translating into lower FPS.

 

 

The only fix for that is to remove the CPU and Software Compatibility Layer from the API and send commands directly to the GPU, ie DX 12_0/1, AMD Mantle, Vulkan.

 

Lower Draw distances will significantly reduce those said commands at any given time.

[DerekSpeare]

Making DCS Multi Threaded will not resolve the DirectX API getting bogged down by object counts, translating into lower % of draw calls being processed per frame, translating into lower gpu usage while GPU waits for commands, translating into more GPU Cores sitting in standby mode waiting for instructions, translating into lower FPS.

 

 

The only fix for that is to remove the CPU and Software Compatibility Layer from the API and send commands directly to the GPU, ie DX 12_0/1, AMD Mantle, Vulkan.

 

Lower Draw distances will significantly reduce those said commands at any given time.

 

What's your best guess for when that could be? Five years? Seven? Since I'm in it for the long haul - and have been - I understand it takes time. I remember the day when VGA graphics were "cutting edge"!

 

Then the next question will be what will exist then that causes bottlenecks? 4k VR HMD technology? :joystick:

[SkateZilla]

Predicting 5 yrs into the future is not a good way to spend time or development,

 

All it takes is ONE innovative piece of software or hardware to change everything.

 

Right NOW, in 5 years, there's 2 Very different paths demonstrated.

 

nVidia is choosing to Stick with LARGE GPU's

AMD is Going for Smaller GPU's Linked on a Interposer.

 

How does this effect anything?

 

Right now, All MAJOR AAA Studios Are Developing for AMD GPUs/CPUs Because that's what All the Current and Next Gen Consoles will be using for the next 4+ Years.

 

That Means Games are coded on AMD Architecture and Ported to nVidia, and they usually result in lower performance on nVidia's GPU's or downright broken.

 

Which means AMD, having invested in Small Size GPUs and APU's, will spend more time and more of their Wafer Allotment on SMALLER GPU's.

 

This effects Gaming Globally because, Instead of MASSIVE GPU's from AMD, you'll have Scalable GPUs,

 

As in every GFX Card will have the SAME PCB, the SAME Memory, the SAME VRM, the SAME interposer on the PCB,

So Say, R9-780X all the Way down through R5-740 will be Identical.

 

The Difference? the amount of smaller GPU's/Memory on the Interposer Layer of the GPU.

 

ie, Same Cards all the way up to the Interposer:

One Unique Smaller GPU being Manufactured on the same Wafers, One PCB Being manufactured to house the GPUs on a Interposer Layer.

Meaning it will have Lower Cost and higher Yields and be used for everything from Entry Level to Enthusiast to Semi Custom Solutions (Consoles, Next Gen are Rumored to be Dual GPU):

 

ie: Hypothetically:

R5-740 will be 1 GPUs on the Interposer

R5-750 will be 2 GPUs on teh Interposer

R7-760 will be 3 GPUs on the Interposer

R9-770 will be 4 GPUs on the Interposer

R9-780 will be 6 GPUs on the Interposer

R9-780X will be 8 GPUs on the Interposer

 

Seeing as DX12+ uses Explicit Multi Adapter, Linked and UnLinked,

As far as the End User is Concerned they have one Powerful GPU.

 

But as far as DirectX12+ is concerned, you have x Number of GPU's each with their own Command processors and caches etc.

 

And it will cost the consumer Considerbly less,

 

as Smaller GPUs can fit more per wafer, and have higher yields as wafer defects etc will not effect as many GPUs.

 

ie (in Basic Geek Speak, actual sizes/prices are no where near this example):

if you have a 36x36 Inch Wafer, and you want to Manufacture a 18x18 GPU,

You'll Get 4 GPU's Per Wafer, Now, if there is ONE Defect in that wafer, One GPU is automatically damaged.

 

So If a Wafer Costs $30K Each, and you only get 3 Functional GPU's out of it, each GPU will Cost $10K to Produce, which effects cost of the GFX Cards.

 

 

Now if you have a Scalable Smaller GPU say, 3x3, on a 36 Inch Wafer, you'll be able to Manufacture 144 GPU's per wafer, With that same 1 Defect, you'll only lose One GPU and Have 143 Working GPU's, At a cost of $210 Each

 

Now you can take 9 of those GPU's and Link them on a GPU Interposer for the same Performance.

 

1 GPU at 18x18 w/ ONE Command Processor and cache bank would Cost: $10K to Produce

9 GPUs at 18x18 w/ 9 Command Processors and cache banks would Cost: $1.8K to Produce.

 

The 9 GPU Version will be able to divide up tasks as well (GPUs 0,1,2,3 Left Eye, GPUs 4,5,6,7 Right Eye, GPU 8 Post Processing etc.)

 

 

 

 

The evidence of this is already present,

nVidia's Huge GPU will cost 5K+ Retail, because of how much a Wafer Costs, and How many of them are thrown out for defects (ie yields).

 

AMD will not be Producing anymore Huge GPU Architectures, 2 to 3 years from now, they will be using Multiple Small GPUs on a large Interposer Layer w/ HBM2/NGM

 

 

The Technology is Already there, the Fury and FuryX Cards use a Interposer Layer to Put the HBM on the Same Substrate Package.

AMD will simply expand on the Tech to facilitate Multiple Small GPU's.

 

The Same Tech Will also make it's way to CPU's, Smaller 4 Core Architecture will produce more Yields at a lower Cost than a Giant 8, 12, 16 Core Architecture.

 

Zen+/Zen 2nd Gen Will be Scalable, meaning they'll take 4 Core Samples and Put them on a Interposer to make a 8, 12, 16 core Chip.

 

As far as the End user is aware they have one CHIP/Substrate Package w/ 4,8,12,16 cores.

As Far as the System is aware, there are multiple CPU's each with their own memory controllers etc.

Edited April 28, 2016 by SkateZilla

[Glowing_Amraam]

That Means Games are coded on AMD Architecture and Ported to nVidia, and they usually result in lower performance on nVidia's GPU's or downright broken.

 

Just curious, why do you state this? In the overall gaming market Nvidia is doing pretty good if you ask me. Unless I am reading this out of context?

[SharpeXB]

Let's not place our bets on AMD. They're on the verge of extinction.

http://www.digitaltrends.com/computing/is-the-eternal-pc-graphics-war-coming-to-an-end/

 

Team Red is losing the market share so badly they may end up out of it for good. Which is actually not good news. Competition is always good for the consumer.

Edited April 28, 2016 by SharpeXB

[Glowing_Amraam]

^^^^^

 

That article is over 1 year old now.

[SkateZilla]

AMD isnt going anywhere,

 

They signed a Significant Deal to Produce CPUs and GPUs for China (to the tune of $293 Million and twice that for the GPUs).

 

nVidia's marketshare is already declining,

As both Current Consoles Run AMD Semi-Custom Solutions, which makes up about 65% of the Gaming market alone, and another 8% for Xbox 360, and When Next Gen Consoles come out to the tune of all 3 Running GCN1.2 GPUs, that 65% of AMD Marketshare will jump to likely 85%+, All because nVidia didnt want to get involved w/ Consoles.

 

All 3 Next-Gen Consoles will Run AMD Semi Custom Solutions w/ Multiple GPUs, which means Most AAA Titles will Run on AMD Multi GPU Solutions Better.

 

**Though this has nothing to do w/ DCS Directly**

 

AMD will be able to produce significant amounts of GPUs per wafer at a lower cost and Link them on a Interposer.

 

While nVidia is still producing single Large GPU's that run at 5K each,

 

Not to mention Console Ports, since they are coded for GCN, run great on GCN, and when You run them on nVidia, Boom, Implosion.

 

That Scenario is only going to get more frequent, along with nVidia GPU Costs and Prices Going Up While AMD's Go Down.

Edited April 28, 2016 by SkateZilla

[SharpeXB]

^^^^^

 

That article is over 1 year old now.

The trend hasn't favored AMD for 10 years

 

Let's not place our hopes for VR performance on any one manufacturer.

Edited April 28, 2016 by SharpeXB

[SkateZilla]

Just so you know:

Discrete Graphics Marketshare is likely less than 20% of Total Graphics Marketshare now.

 

For Total Marketshare:

Current Consoles Own 65+%, Which is all AMD

Previos Gen Consoles own about 10 % (which is 1/3 AMD X360)

Discrete 15%

Mobile / Other 10%

 

AMD Owns All Current Consoles, 33% of Prev. Gen, 18 or so Discrete, and like 5% Mobile (which is ruled by ARM).

 

Then in 2 years, AMD will own All 3 Current Consoles, plus 100 % of the Previous Gen Category. Which is 75+% total Marketshare own it's own,

Then add in Discrete (~20% of 15% = 3%), 78% Total Marketshare

Then add in CPU/Integrated %, Which is likely barely 2% total, as Intel HD has a current hold on that by a large margin,

Then add in any Mobile or Others which is ruled by ARM.

 

Now,

When you look at PC/Discrete, the Marketshare owned by nVidia is mainly Pro Market, which kept nVidia at or above 50% Discrete Marketshare while the 400 Series Bombed.

 

So, with Current Gen and Next Gen Consoles Locked onto AMD GCN 1.2+, Developers will Start Replacing their nVidia Pro Discrete Graphics w/ AMD Options, Makes sense to Develop On AMD for use with AMD,

 

Now, Console -> PC Ports, will favor AMD, they were developed on and for AMD GCN, so when people get tired of the abysmal performance and downright broken PC Ports on NVidia and Game Works, they will dump the nVidia GPU's to Run AMD, since that is what the Software was developed on and For.

 

That market share in a few years will do a Drastic 180.

 

Then Add in the fact that AMD will be Manufacturing their GPU's SMALL and Scalable, While nVidia will be making theirs monstrous and expensive,

Why pay 5K for an nvidia Card when you can get the same performance for 1K or Below on a AMD Solution, Wafer Costs are going up, as are defects due to 14nm and below Fabrication, Bigger GPU Designs will produce lower Yields and higher cost.

 

It will only push people to AMD further.

 

on the CPU Side, Zen Will Close the Gap between AMD and Intel, Intel has slashed their Desktop x86 R&D Budget significantly and decided to go for the mobile market where ARM pretty much owns everything.

Zen 2nd Gen will use the same interposer on the Substrate package as the GPU Division to put multiple smaller CPU's on the interposer to create a Larger CPU, thus, Smaller CPU Designs will produce more per wafer and higher yields and lower cost.

 

Intel is about to License GCN for Integrated Graphics, to Eliminate Cost of Developing their HD Graphics Block, which will also help AMD's Graphics Marketshare.

Edited April 28, 2016 by SkateZilla

[SharpeXB]

Just so you know:

Discrete Graphics Marketshare is likely less than 20% of Total Graphics Marketshare now.

But that's the graphics card segment which is relevant to DCS. And in any case we are all better off having two companies compete for our business.

Edited April 28, 2016 by SharpeXB

[Swoodle]

Here is a pretty good explanation of where AMD is heading;

[ame]

[/ame]

 

[ame]

[/ame]

 

Anyone who wasn't paying attention when they took over the console world missed a pretty important milestone. For me however who is in the middle of a new PC build, I'm waiting to see what sort of output Pascal has. That will let me decide if a 980ti will tire me over while I wait for AMDs next drop. CV1 doesn't arrive till July so can afford to wait.

[SkateZilla]

I posted that already a few pages back.. lol:

http://forums.eagle.ru/showpost.php?p=2754863&postcount=60

[vicx]

Red, Blue and Green Teams aren't going anywhere. Each of them have unique and defensive positions. AMD, Intel and Nvidia are actually making mutually beneficial specialisations to defend their shared turf against the "real threat" which is coming from mobile ecosystem vendors.

 

Five years ago people were predicting the demise of AMD at the hands of Intel. There were big Intel VS AMD threads and a point I always made at the time was that Intel and AMD are actually on the same team (x86) and both of them would be fighting for their survival vs ARM.

 

If you watch THAT video that SZ posted you might think it predicts the death of Nvidia BUT the truth is that what helps AMD ends up helping Nvidia too because they are a duopoly and there is a lot of commonality to their technology. Only companies not already in the market are affected by AMDs move. Bringing the price/perf point down using APIs helps Red and Green.

 

I think Nvidia will follow AMD into smaller dies if/when it makes sense. Right now it is AMD that plays the early mover because they took a leadership role defining the closer-to-the-metal APIs, BUT Nvidia will do just fine being a leader in high performance computing and a follower in visual computing.

 

If you are wondering about Nvidia's pursuit of HPC ... it makes them a much more attractive merger partner. AMD has leadership of visual computing APIs BUT Nvidia is making a much bigger play by focusing on deep learning and HPC.

 

 

VR performance in DCS is a much harder subject to talk about than this other stuff :)

Edited April 29, 2016 by vicx

[Dave1944]

DCS with VIVE

 

Hi, this is my first post to this forum.

I'm a 71 year old who still loves playing games, and although I have had a computer for many years (I had one of the first ZX80s), I still have trouble.

I recently purchased the HTC VIVE and some games. Everything is working as it should be with all the games apart from DCS. I've also bought the Hawk and after first installing it everything was great. To tell you the truth I was blown away by the graphics; that is until it would only run on the desktop and not in VR. I've tried almost everything, including a complete reload but still no joy.

I am now at my wits end and about to jump off the Tamar Bridge so if I don't reply to a suggestion, that's what I've done. All joking aside I would very much appreciate any help.

Many thanks

Dave

[DerekSpeare]

Congrats, Dave! I'm not sure if ticking the box "Autodetect Oculus Rift" will help get the Vive going. That's accessible in your options menu; click the gear symbol at the top center of the screen.

 

I use the Oculus, but this thread here may help:

 

http://forums.eagle.ru/showthread.php?t=139912

 

There are Vive folks who post to that thread regularly. I'd bet they can guide better.

[vicx]

Hi, this is my first post to this forum.

I'm a 71 year old who still loves playing games, and although I have had a computer for many years (I had one of the first ZX80s), I still have trouble.

I recently purchased the HTC VIVE and some games. Everything is working as it should be with all the games apart from DCS.

Dave

 

You have to use the standalone versions of DCS. The "Steam" version of DCS is not new enough. I just mention this because you did not mention which version of DCS you are using.

 

For further assistance you probably need to provide more details.

[j0nx]

You have to use the standalone versions of DCS. The "Steam" version of DCS is not new enough. I just mention this because you did not mention which version of DCS you are using.

 

For further assistance you probably need to provide more details.

 

Steam 1.5 doesn't work with VR yet?? First I've heard of this. Bummer because I run 1.5 Steam myself and my Rift is incoming in 2-3 weeks.

[SkateZilla]

Pretty sure steam's Open Beta Branch is new enough.

[BigDuke6ixx]

Steam 1.5 doesn't work with VR yet?? First I've heard of this. Bummer because I run 1.5 Steam myself and my Rift is incoming in 2-3 weeks.

 

Yeah, so much for Steam's promises on keeping games up-to-date. Bloatware, avoid.

[SkateZilla]

Steam's Version should be new enough, the issue is when you launch a Steam Version you launch w/ the AppLaunch Command for Steam Shortcuts.

 

You Prolly need to make a new shortcut that points to /SteamApps/Common/DCS World/Bin/DCS.exe

[_Acoustic_]

So first thoughts on OR:

 

Wow, wow, wow.

 

I'll write up some more next week but need to figure something out first.

 

I can put every setting at max and get 45 fps average at Groom. I lower everything to medium and get the same, 45. Lower everything to low or off, and I still get 45. Once I get into the air and gain a little alt is the only time things will go up, but really never over 60 on any of the settings

 

Surely my machine can get 90fps over the desert in an empty mission with everything on low and or off?

[DerekSpeare]

I have everything on what's essentially minimum. I ran the same test with the F15 (your plane choice wasn't stated) and encountered fluctuations between 64-75 with the DK2 on the ground at Groom. Dense object areas just wreck the frames, it seems. ATW only goes so far, but this NTTR map is tough for even my machine to handle - on minimal settings, no less.

 

My last PC was a 3820 with 780 and shelved it for this one. It seems to me that DCS and VR need to fine tune their relationship together.

 

NB: I do have shadows on LOW - I like them on and visible from the seat.

Edited May 13, 2016 by DerekSpeare

[KLR Rico]

I can put every setting at max and get 45 fps average at Groom. I lower everything to medium and get the same, 45. Lower everything to low or off, and I still get 45. Once I get into the air and gain a little alt is the only time things will go up, but really never over 60 on any of the settings

 

There's a dozen posts from VR users saying "settings don't make a difference on frame rate", but the general response is "that's the way it is".

 

I've given up on trying to tweak it for consistently good FPS, so I just put it on medium/high and enjoy it the way it is. :joystick:

[Purg]

Surely my machine can get 90fps over the desert in an empty mission with everything on low and or off?

 

It should be able to. I have a GTX780 and according to the ingame fps counter, I get 90fps in 1.5.3 in a KA-50 while doing a startup procedure. Haven't really checked the frames elsewhere since it seems smooth flying around. Waiting for the 1080 to come out before I really start having a go.

[SkateZilla]

FPS on a Single Screen vs 2 3D Stereoscopic Screens will not be the same.

 

You can have 144 FPS on a 1080p Screen, and drop to less than 60 on a Rift CV1