Why is VR FPS so poor in comparison to monitor?

For the same scenario in DCS, on my 4K monitor I would get around 60/80 FPS that on my Vive Pro I am getting 30-45 fps. And that is on lower settings.

Now given that the Vive Pro is 2880x1600 and the monitor is 4K, I just wondered, what is the technological reason for this?

Isn't it because in VR your PC has to produce 2 of those images? One for each eye.

It renders two separate viewports, one per eye. Thats the main fps eater. On top of that, I guess a bit of power goes to the warping of each viewport, on a shader level I guess.

With a monitor you need to calculate the scene from a single viewpoint. With VR you need to calculate the scene from two viewpoints. This costs a lot of extra cpu cycles and basically causes DCS to become CPU bound. You'll find that your GPU usage drops in VR while your CPU is maxed out on 2 cores.

Yes stereoscopic and you can also freely look around in a simulated 3D environment, takes some number crunching. :) :thumbup:

With a monitor you need to calculate the scene from a single viewpoint. With VR you need to calculate the scene from two viewpoints. This costs a lot of extra cpu cycles and basically causes DCS to become CPU bound. You'll find that your GPU usage drops in VR while your CPU is maxed out on 2 cores.

Sent from my Redmi 4 using Tapatalk

Not my experience, 1080ti easily goes to 100% with relatively low graphics with 8700k.

 

Sent from my Redmi 4 using Tapatalk

You're probably: running enough PD, playing single player, using MSAA. Try again on a heavy MP server and you'll see what I mean.

Actually i have tried with 1080Ti and I5-9600 with HTCVivePro with SS between 1.5-2.0 and GPU is what is on 100% and CPU is rarely goes to 40% (but actually that is not per core statistics as i am looking for software which would show individual core performance in graph format)

Interesting info. I was just wondering this last just. I have a i7 8700 with 1080ti and 32 gig and my frame rate with Odessey + is 30fps. With just monitor I’m getting 70-90.

I’m nervous to overclock my CPU. I’m not experienced in doing that. Wondering if overclocking would help.

Sent from my iPhone using Tapatalk

You're probably: running enough PD, playing single player, using MSAA. Try again on a heavy MP server and you'll see what I mean.

Change Graphic settings occasionally, normally PD 1.5, shadow flat or low, MSAA x2 or none.

Fly both SP and MP , when I test performance and check utilization I do it in SP but I don't see how my GPU utilization will decrease in MP vs SP with the same settings.

If you load too many AI units, sure that will utilize more CPU so some servers will load CPU due to high amount of AI calc, but VR is pushing the GPU to the max and not CPU in my case.

My old system was GTX970 with 4790 and the result was pretty much the same, CPU cores were not going 100 but the GPU was always maxed out, I also could clearly notice the slight increase of frames by overclocking the GPU.

Not saying it to argue, just sharing my experience. Now for sure, if one has RTX2080TI and I3 CPU might bottleneck but with equal generation CPU and GPU, from my experience in VR the bottleneck is the GPU.

Why is VR FPS so poor in comparison to monitor?

typically VR puts additional load in the CPU (GPU’s too of course, but VR resolutions are so low that any modern GPU can render those display sizes easily).

then again, you can easily up the load on the GPU itself by increasing your SS or PD values. so much so, that you often fall below the 90 fps design target

Change Graphic settings occasionally, normally PD 1.5, shadow flat or low, MSAA x2 or none.

 

Fly both SP and MP , when I test performance and check utilization I do it in SP but I don't see how my GPU utilization will decrease in MP vs SP with the same settings.

 

If you load too many AI units, sure that will utilize more CPU so some servers will load CPU due to high amount of AI calc, but VR is pushing the GPU to the max and not CPU in my case.

 

 

 

My old system was GTX970 with 4790 and the result was pretty much the same, CPU cores were not going 100 but the GPU was always maxed out, I also could clearly notice the slight increase of frames by overclocking the GPU.

 

 

 

Not saying it to argue, just sharing my experience. Now for sure, if one has RTX2080TI and I3 CPU might bottleneck but with equal generation CPU and GPU, from my experience in VR the bottleneck is the GPU.

What is MSAA?

Sent from my iPhone using Tapatalk

typically VR puts additional load in the CPU (GPU’s too of course, but VR resolutions are so low that any modern GPU can render those display sizes easily).

And you have experience of how many different HMD and how long time? Do you consider Asus GeForce GTX 1080 Ti ROG-STRIX-GTX1080TI-O11G-GAMING as modern GPU?

Isn't it because in VR your PC has to produce 2 of those images? One for each eye.

No, that’s a total for both eyes on the Vive Pro.

Fov is higher therefore more scenery and objects to be rendered simultaneously.

Sent from my Redmi 4 using Tapatalk

Why is VR FPS so poor in comparison to monitor?

Do you consider Asus GeForce GTX 1080 Ti ROG-STRIX-GTX1080TI-O11G-GAMING as modern GPU?

i do.

no modern gpu has trouble rendering a 1920x1080 display or two 1,080 x 1,200 pixel displays in the case of VR. its not a lot of pixels

scene creation, object positions, geometry calculations, culling, all that is done on the CPU and it’s a CPU intensive operation. and in the case of VR, you have to do it (2) twice while tracking the users bobbing head. oh, and the VR apis take time to call too.

in the GPU, you do shadows, shaders, lighting, things that make use of the simple but highly parallel stream compute units. this is pretty easy work and what the cards are designed for.

what kills the GPU in VR is all the extra stuff we do, like raising the SS. and sharpening, and greatly increasing the pixel count (often exceeding 4k resolutions!), to try and see things on the low resolution display. things you would never do on a monitor display.

i do.

no modern gpu has trouble rendering a 1920x1080 display or two 1,080 x 1,200 pixel displays in the case of VR. its not a lot of pixels

 

scene creation, object positions, geometry calculations, culling, all that is done on the CPU and it’s a CPU intensive operation. and in the case of VR, you have to do it (2) twice while tracking the users bobbing head. oh, and the VR apis take time to call too.

 

in the GPU, you do shadows, shaders, lighting, things that make use of the simple but highly parallel stream compute units. this is pretty easy work and what the cards are designed for.

 

what kills the GPU in VR is all the extra stuff we do, like raising the SS. and sharpening, and greatly increasing the pixel count (often exceeding 4k resolutions!), to try and see things on the low resolution display. things you would never do on a monitor display.

Ok so i did two different runs in Caucasus map. First i have SS set to 1.0 which is VivePro native resolution and on 2nd run i have set SS to 1.6

On both cases you can see GPU utiliziation on picture circled in blue and highest it is 97 and all the time over 80% I try to keep graphic settings so that GPU utilization dont go into 100% because then it is not possible to play in VR.

On both cases CPU utilization beside few spikes where it goes to 40% it stays on avarage between 25-20% on all cores.

So i do not know where do you get your ideas but in my setup GPU is bottleneck even without exceeding or pushing resolutions and struggles to maintain even native resolution. CPU is hardly heating at all.

So modern GPU dont handle and draw VR easily and without problems at least not when person is using HTCVive Pro/Odysseus or coming Pimix displays.

at ss 1.0 the graph is much smoother for the GPU (the left blue circle)

at 1.6 there is a lot more spiking in the graph (its not as smooth)

this is the CPU struggling to feed the gpu fast enough. the GPU spikes as it gets fed and then waits.

so at 1.6 you are being CPU limited to a degree. before you hit 100% GPU.

and its not the cpu being overutilized, its not fast enough..

no modern gpu has trouble rendering a 1920x1080 display or two 1,080 x 1,200 pixel displays in the case of VR.

Or 2 x 1440 x 1600 in the case of Vive Pro or Odyssey.

With my i9 9900k, CPU utilization can go up to 90% running hornet tutorial missions but only in one core. So if you only look at average CPU utilization the number would be much lower which can be misleading. My 2nd core utilization is typically around 40% while the rest (6 other cores) are below 10%. Apps such as HWINFO64 can gives lots of detail info regarding separate core temperature, percent use, voltage, clock speed, etc.

Sent from my iPad using Tapatalk

at ss 1.0 the graph is much smoother for the GPU (the left blue circle)

at 1.6 there is a lot more spiking in the graph (its not as smooth)

this is the CPU struggling to feed the gpu fast enough. the GPU spikes as it gets fed and then waits.

 

so at 1.6 you are being CPU limited to a degree. before you hit 100% GPU.

and its not the cpu being overutilized, its not fast enough..

GPU utilisation dont support this as graph in bigger picture shows that CPU is mainly used in 20-30% utlilization

GPU utilisation dont support this as graph in bigger picture shows that CPU is mainly used in 20-30% utlilization

DCS is single threaded, windows tries to spread the workload across all the cores (so long as your affinity is set to do so). So if you take the number of cores and multiply DCS CPU% you will have a much more true number... example, 4 cores 25% = 100% cpu, and there for CPU bound not GPU bound. DCS is generally CPU bound on high end system people use for VR (your 1080/Ti builds)

4 cores 25% = 100% cpu !? No way I can figure out this. Not the first time I read that so it might mean something… but I don't know what.

DCS is single threaded, windows tries to spread the workload across all the cores (so long as your affinity is set to do so). So if you take the number of cores and multiply DCS CPU% you will have a much more true number... example, 4 cores 25% = 100% cpu, and there for CPU bound not GPU bound. DCS is generally CPU bound on high end system people use for VR (your 1080/Ti builds)

I try to understand this logic but I find it little bit hard

If I monitor CPU usage right from monitor window and it shows 30% and we know that DCS is using mainly single core so I do not understand what multiplaying core load with core numbers brings into this. Or are you saying that monitoring softwares which shows core load do not snow it correctly.

Sorry but I do not understand your explanation :(

I try to understand this logic but I find it little bit hard

 

If I monitor CPU usage right from monitor window and it shows 30% and we know that DCS is using mainly single core so I do not understand what multiplaying core load with core numbers brings into this. Or are you saying that monitoring softwares which shows core load do not snow it correctly.

 

Sorry but I do not understand your explanation :(

Sorry DCS uses 2 threads, one for sound, one for everything else... so there will be a bit more CPU % here as well, so 5% for sound then...

Windows will try to spread the workload as much as it can on a single threaded application, so it will try and make your machine not use just 1 core for a single threaded app, adversly a multithreaded app could potentially use 1 thread per cpu and peak each cpu core at 100%. Hope this helps explain that, if not I can try again ;)