Jump to content

Recommended Posts

Posted

Okay, so...at a loss here. Screenshot will probably help more than anything.

 

Specs:

3570K OC'd to 4.2 (yup, living in the past still and okay with it)

Z77-HD3 mobo

1060 GTX 6 GB (I think drivers are from April)

 

Oculus Rift CV1 and/or 3 monitor setup at 5760 x 1080.

 

Been playing with this for a day or so. VR is garbage at the moment, and it makes no sense. We could all agree that I'm due to upgrade...but at the same time things don't add for that being the current reason why my VR frames suck. When I put myself over downtown Vegas at 5760 x 1080, I'm sitting on 70 FPS with my current setup...not shabby at all for a 8 year old processor. I switch to VR, struggling for 30. Initially tried without an overclock vs overclocking to 4.2, still the same result. Screenshot tells the story on usage. MSI AB monitoring utility isn't saying anything different really. That wasn't a low spike, there's still consistent overhead there and I'm stuck at garbage frames.

 

Anyone have any bright ideas on a cause or a fix?

297391549_DCSVRperformance.thumb.png.c356520b6df0a30d4a9607d5ff560cda.png

Posted (edited)

GPU utilisation is not a reliable indicator of bottlenecks. Its one number that shows something. I have never found what it shows exactly, but it most likely shows shader unit utilisation rate (most common bottleneck and performance factor in games), while the bottleneck could be elsewhere, especially for DCS its more likely in the texturing mapping units or maybe even rasterizers. The number task manager shows is not proof the GPU isnt your bottleneck.

 

The better way to figure out your bottleneck is underclocking, like you did with the CPU. Try it with the GPU, underclock it, see if it makes a difference. You could even vary core and vram clocks separately if you want. If performance drops almost linearly with clock speeds, then you found your bottleneck. If performance barely changes, then something else is bottle-necking you.

 

But my guess is your main bottleneck is neither the CPU or GPU, but your DDR3 DRAM. See my post here:

https://forums.eagle.ru/showpost.php?p=4378621&postcount=102

Edited by Vertigo72
Posted
GPU utilisation is not a reliable indicator of bottlenecks. Its one number that shows something. I have never found what it shows exactly, but it most likely shows shader unit utilisation rate (most common bottleneck and performance factor in games), while the bottleneck could be elsewhere, especially for DCS its more likely in the texturing mapping units or maybe even rasterizers. The number task manager shows is not proof the GPU isnt your bottleneck.

 

The better way to figure out your bottleneck is underclocking, like you did with the CPU. Try it with the GPU, underclock it, see if it makes a difference. You could even vary core and vram clocks separately if you want. If performance drops almost linearly with clock speeds, then you found your bottleneck. If performance barely changes, then something else is bottle-necking you.

 

But my guess is your main bottleneck is neither the CPU or GPU, but your DDR3 DRAM. See my post here:

https://forums.eagle.ru/showpost.php?p=4378621&postcount=102

 

Thanks for the input. I have been using MSI AB for testing as it has a really nice in-game overlay for seeing usage and just about anything else you might want. Unfortunately, it’s absolutely useless in VR as it won’t display unless your in 2D. Using the logging tool, I could verify that both were giving pretty much the same numbers, the win monitoring was just the only way to get a real-time screenshot.

 

Changing the clocks on my video card would be super easy, so I may mess with that a bit and see if the usage numbers go up as a result.

 

The RAM deal crossed my mind. There’s just one thing that bugs me, and that’s the huge disparity between running in 2D vs 3D. If we’re talking video rendering, the pixel count should be almost a 1/3 higher in the 5760 res vs the VR 1920x2...unless I’m missing something. In the end, data being data, why such the hiccup? If it’s getting slowed by the RAM, why only in 3D? My RAM is descent stuff, so there’s a little room to mess with it, just not sure if there would be any way to differentiate results.

Posted (edited)

VR performance is always going to be slower as it involves a number of additional steps, like rendering 2 eye camera's, warping / projecting the image and frames need to be rendered at a higher FoV, so rendering a ton of stuff you;ll never see, before you can warp them, and really it should be rendered at a higher resolution too or it will look rather terrible. You are basically doing a lot of work "for nothing". I dont know what you set your MSAA and SS at, but those are pretty much required for VR and will bring any current PC to its knees in DCS.

 

Now whether the above has any bearing specifically on dram speed, I have no clue. That 70FPS sounds like a lot to me, but its hard to compare without replicating your exact settings and I dont have Nevada either. To give you an idea, using medium preset, instant action, P51, caucasus, cold start, I get about 60 FPS on triple HD monitors and 140 on just one. Thats with a ryzen 2600, 16gb DDR4-3200 and gtx 1070. I dont think our cpu's will be very different, gpu shouldnt be a major bottleneck on 1 screen, so if your performance is close to mine, ram isnt likely your problem; but it never hurts to test.

Edited by Vertigo72
Posted

Yeah, I dunno either. It is DDR3 1600...but it only shows about 11/16 usage using MSI AB. I would think it would flood more data to the RAM to make up for any slow speed issues, but that’s merely speculation on my part.

 

Any chance the mobo could be bottlenecking data? Anyone know of any way to check the data transfer amounts across it?

Posted (edited)
I would think it would flood more data to the RAM to make up for any slow speed issues,

 

ahem.. no, thats not how it works :)

 

Any chance the mobo could be bottlenecking data? Anyone know of any way to check the data transfer amounts across it?

 

Thats not really how it works either. But yes, I already told you how you can help determine what is primarily bottle-necking your performance: under clock. If you underclock your ram and performance decreases linearly, then you are heavily dram speed limited. If it makes no or almost no difference, then its not bottlencking you. Same with GPU speed. If lowering the clock speed of your 1060 by say 20% results in (close to) 20% lower frame rates, then you are GPU bottlenecked, regardless of what task manager or afterburner or anything else says. You already determined you are not primarily cpu (frequency) bottlenecked, since lowering your cpu clock speed made no difference. The logical conclusion of that, is increasing the clockspeed isnt going to improve things either.

 

The only things this does not help determine, is if there is a capacity constraint somewhere. Like running out of video ram which could result in the GPU having to send massive amounts of data back and forth over the PCI-E bus and that could tank performance. And that could also result in task manager showing the GPU as being "50% idle" because 50% of the time its waiting for data from main memory.

 

Lowering and increasing texture quality and changing the "cockpit display resolution" should change the amount of vram you need to some extend and might help identify that as a bottleneck, but its not a "direct" test and I have no idea if even the lowest settings might exceed the 6GB of your card in VR mode. Looking at afterburner wont tell you this either, it will use all 6GB. It will use 11GB if you have 11GB and I bet DCS will use 24GB if someone makes a 24GB card. The question is how much is actually needed to prevent massive PCI traffic, and I cant tell you that. But I wouldnt be surprised if its more than 6GB in VR.

Edited by Vertigo72
Posted

I know VR is more of a strain here and different VR drivers.

 

But the OP can run at 60 or 70 fps at 5760 x 1080 (3 monitors) generally? with X game settings.

 

That's 6,220,800 pixels

 

With a Riift CV1 pixel density (PD) of 1.5 = 3240x1800 resolution = 5,832,000

 

Understanding Pixel Density - Making sense of it

i7-7700K OC @ 5Ghz | ASUS IX Hero MB | ASUS GTX 1080 Ti STRIX | 32GB Corsair 3000Mhz | Corsair H100i V2 Radiator | Samsung 960 EVO M.2 NVMe 500G SSD | Samsung 850 EVO 500G SSD | Corsair HX850i Platinum 850W | Oculus Rift | ASUS PG278Q 27-inch, 2560 x 1440, G-SYNC, 144Hz, 1ms | VKB Gunfighter Pro

Chuck's DCS Tutorial Library

Download PDF Tutorial guides to help get up to speed with aircraft quickly and also great for taking a good look at the aircraft available for DCS before purchasing. Link

Posted

 

With a Riift CV1 pixel density (PD) of 1.5 = 3240x1800 resolution = 5,832,000

 

you are only counting pixels that end up on your screen. With a flat screen, every pixel that is rendered is eventually shown on your monitor. With VR, because of the lens distortion, you need to warp a rectangular flat projection that our GPUs produce, in to some weird shape, stretching it at the borders and compressing it near the center, which throws away a massive amount of pixels.

 

Im also a little skeptical about the 70FPS claim on triple HD with a 1060. Im not getting that with my 1070. And I could get somewhat acceptable frame rates in SP when I had a rift-s.

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...