[result] rtx 3080 for VR

[speed-of-heat]

I think there are definite performance merits in supporting multi core... what we don’t really know for example is how much of the bug thread is currently dealing with graphics, what Vulcan doesn’t do for example is deal with AI logic ... and we know that as unit count increases cpu workload increases, and we have seen that unit pathing can have a tremendous impact on cpu perf.

 

So whilst Vulkan will improve performance, I suspect that unless other aspect of the engine are also multi threaded as well,and mostly areas around AI logic, then performance will not substantially increase in scenarios with high unit counts.

 

I stress this is speculation, and I also stress that I do think Vulkan will offer significant improvements, to the rendering performance. But that itself will not allow full multi core utilisation of the entire game engine, as it is bigger than just rendering.

[Headwarp]

You should disable motion reprojection to see what framerates you're actually capable of getting or you're just going to see 45fps anytime you dip below 80-90 fps. With reprojection, gpu usage will also be lower than without as the gpu is being frame limited. I would do that both at the resolution you currently play at, as well as increasing the resolution to a point where you are mostly gpu limited just to get a general idea of what the gpu is capable of. Also worth running DDU if you haven't, as well as making sure power settings are set correctly in NV Control panel.. as well as boosting the fans via afterburner or something since Nivida cards run better with lower temps.

 

It might be worth seeing what happens if you disable the meltdown/spectre patches as well, as they've reduced the performance of at least 8th gen and older intel CPUs. if you're up for the risk.

 

At some point, newer cpus will have to break 5.5ghz.. lol.. isn't hardware fun?

 

There's simply more to benchmarking than just throwing a new piece of kit into a system and running software. I mean, it might still end up a disappointment but tests without covering the bases don't really offer much to go on.

Edited September 27, 2020 by Headwarp

[Eldur]

whilst the engine is an issue... if you look at the OP's original post they are getting 80-90FPS, over ground in VR on pretty high settings ... that's pretty good

 

I wouldn't even remotely call flat shadows and terrain object shadows off high settings, especially given the fact he's running a OG (what does that even mean, I understand that as 1st gen) Vive which pretty much has the same specs as a CV1 regarding how much it tanks the hardware. I get 45fps in the same scenario with low-high shadows (no difference at all), flat terrain object shadows and just running a very much dated 3570k and a 1080 nonti and it does not look like early 2000s lighting. Turning the shadows to flat only and terrain shadows to off makes it look much worse, but doesn't increase the performance at all, regardless of flying offline in a rather spartan map or online on an empty Through The Inferno hosting server, just with the difference the online experience is tanked by mostly just 15fps, sometimes even dropping to single digit, maxing out at 22 in less dense areas like offshore.

 

It's so much definately an engine issue, especially WRT MP. I tried running my own training maps off of a dedicated server locally (different rig, 955, 8GB, 6950, console only) and joining that just to find out my render performance then goes down by at least half of the frames where I expected the server to take off some load of my client that wouldn't have to do the AI calcs and so on anymore. It's outright ridilulous. And no new hardware would ever fix that. Not even a 4090Ti.

[Headwarp]

In many Dx11 titles, less than 4k resolution was still showing a cpu bottleneck @5ghz with a 1080Ti (per other peoples benchmarks, I went from a 980Ti to a 2080Ti, and had similar results with the 2080Ti). @3440x1440 i could see a single core of my 8700k hitting 99% while still getting 98-99% gpu utilization.. pretty much riding right on the edge of a cpu bottleneck in DCSW.

 

With my Odyssey and 160% SS, I often get the light blue indicator showing motion reprojection being engaged due to cpu limitations, yet at the same time show the dark blue indicator indicating gpu limitations with my 2080Ti (WMR specific feature). Same riding right on the edges of bottlenecks, but lower FPS overall in VR as one would expect. If I bump up to say 200% SS, which is like 2024x2520 per eye with my odyssey, I get just the dark blue indicator showing being gpu limited (what we strive for) but, ioften I'll see the red indicator meaning that the system can't even maintain 45fps.

 

So, without going to a higher resolution, or getting a good deal above 5ghz on a single core of your cpu, a gpu with more power than a 1080Ti, or even 2080Ti isn't going to show an improvement.

 

This is a big reason we're looking forward to Vulkan. I wouldn't recommend holding your breath waiting for vulkan.. a complete graphics API rewrite, especially one that changes the single threaded nature of DCS on DX11 to a multi-threaded process is a hell of a task.

 

In the meantime, for dx11 and lower titles, especially sims, increased resolution is about the only way to measure performance between nvidia gpus from pascal to now.. other than increasing cpu clock frequencies, which 10th gen k series I7 come stock with a 5.1Ghz Turbo boost.. intel's kind of up against a wall pressing the limits of their current architecture.

 

Currently I run at 1808x2252 per eye and still hit cpu limitations with my 2080Ti @4.9Ghz, and I wasn't seeing a real difference at 5Ghz. Thing being though, my bet is the 30 series will have an easier time at maintaining the minimum 45fps for reprojection at higher resolutions. For index users, at higher resolutions you might be able to maintain that 60fps minimum for 120hz refresh rates.

 

But testing with reprojection on isn't going to net results that tell you much of anything. Bump your SS% up to Reverb G2 resolution or more.. and do all testing with reprojection off imo. Otherwise upgrading to a 3080 from a 1080Ti isn't really going to be beneficial with current hardware and DCS on DX11. In the end, your still basically looking for the highest resolution you can get while maintaining the minimum framerate that still allows reprojection to function.. or sacrificing graphics quality while attempting to hit 90fps.

 

The results don't really surprise me considering the nature of the testing, and the nature of DX11 which doesn't allow multi-threaded processing anywhere near the capability of Vulkan or DX12. I will note that I prefer to run with msaa 2x in DCS, as I just can't stand the shimmers without it.

Edited September 28, 2020 by Headwarp

[SchumiF399]

Here I was thinking a new rtx 3000 series card would fix my frame rate on triple monitors. It seems like it might not be the case. :(

 

Starting to think a lot needs to be done in the optimization department.

[Headwarp]

Here I was thinking a new rtx 3000 series card would fix my frame rate on triple monitors. It seems like it might not be the case. :(

 

Starting to think a lot needs to be done in the optimization department.

 

If you're making your decision based on this user's experience alone, reading about CPU and GPU bottlenecks, the differences in DX11 vs newer API's like Vulkan and DX12, and the effect resolution has on each of these is probably much more valuable than a new piece of hardware can be and will help you make such decisions in the future. Here's an example of how such information can help.

 

If you're hitting 99% on a single CPU core while flying in DCS, which is currently a DX11 title, then you are cpu limited and a faster GPU won't help, where a CPU able to achieve faster clock speeds will. It takes a higher resolution to become GPU limited so you can tell what your GPU is actually capable of. (you have to look at each individual core.. using a quad core for example, one core maxed out is only 25% total cpu usage) If instead you are consistantly using 99% gpu, and none of your cpu cores are anywhere near being maxed out, a faster gpu will likely increase performance.

 

It's going to be a different story when ED is on par with the Vulkan API.. but currently, treat DCSW as a program that relies on single threaded performance up to the highest obtainable cpu clock speeds until you reach a certain resolution.

 

Trying to judge what experience you'll get from someone using a completely different display and resolution, as well as being frame limited by VR specific features intended to reduce the risk of motion sickness isn't going to offer you much to go on. Understanding how hardware bottlenecks work, however, might keep someone from spending money on something that won't offer much of an improvement to their current setup.

Edited September 28, 2020 by Headwarp

[Delareon]

Okay.... so GPU is idle and CPU is almost always at maximum...

 

Its time that ED shows us that they actually spend development time in improving VR experience. e.g. multi threading, Vulcan and other stuff. Hardware can do only so much.

[stormrider]

Okay.... so GPU is idle and CPU is almost always at maximum...

 

Its time that ED shows us that they actually spend development time in improving VR experience. e.g. multi threading, Vulcan and other stuff. Hardware can do only so much.

 

 

Spot on.

[imacken]

Not spot on at all in so many circumstances.

An example https://forums.eagle.ru/showpost.php?p=4501894&postcount=33

My comment there was sarcasm, btw.

[Mars Exulte]

Its time that ED shows us that they actually spend development time in improving VR experience. e.g. multi threading, Vulcan and other stuff. Hardware can do only so much.

 

Nice ''talking point''. It's not a hotfix style solution. Multithreading, engine rewrite, that's a years long process with no ''halfway'' points. They already began, it will likely be several more years. Get comfortable or play something else, those are literally the options. The process has begun, it ain't gonna be finished anytime soon.

[Headwarp]

Not spot on at all in so many circumstances.

An example https://forums.eagle.ru/showpost.php?p=4501894&postcount=33

My comment there was sarcasm, btw.

 

 

While I understand the sarcasm and appreciate that you seem to get it.. have you tried to set some custom fan curves for your GPU? :) IDK about Palit, but my 2080Ti tends to stay just around 61C @ 99% usage. Though the sound of it sometimes scares me and causes me to crank my neck checking 6 in WW2 birds as the fans spin up.

 

There's been a time or two where a driver update has caused me to need to set up my fan curves again and the gpu getting above 80C had a pretty big effect on performance.

 

Anyway, just some food for thought. My fan curves are way aggressive. I like to get the air flowing well before it's started getting towards that 60C temp range.

Edited September 28, 2020 by Headwarp

[STRYC]

My friend you have your preload radius on 100%, which is choking your ram...

This is what causes your stuttering. For dcs you need at least 32gb.

 

 

I was running 16Gb on my system and saw with DCS running I was using almost all of the 16GB. So I bought 32Gb of 4000mhz ram. I tested frame rate before and after and saw NO change in performance. Now with DCS I use about 20 to 25 gigs of the 32, but no change in performance.

[imacken]

While I understand the sarcasm and appreciate that you seem to get it.. have you tried to set some custom fan curves for your GPU? :) IDK about Palit, but my 2080Ti tends to stay just around 61C @ 99% usage. Though the sound of it sometimes scares me and causes me to crank my neck checking 6 in WW2 birds as the fans spin up.

 

There's been a time or two where a driver update has caused me to need to set up my fan curves again and the gpu getting above 80C had a pretty big effect on performance.

 

Anyway, just some food for thought. My fan curves are way aggressive. I like to get the air flowing well before it's started getting towards that 60C temp range.

 

Thanks for that. I wasn’t looking for help, I just get fed up with some people stating that DCS is CPU bound per se.

[hansangb]

Thanks for that. I wasn’t looking for help, I just get fed up with some people stating that DCS is CPU bound per se.

 

Not always. Different HW specs have different bottlenecks. always have, always will.

 

But DCS fires up 50-64 software threads. All of which runs on one core (maybe two, I'm told the audio portion runs on a separate core). And resource meter bears this out.

 

But again, I understand the complexity of getting the scheduling right for multiple threads to coordinate. I'm afraid when I took Operating Systems, multi-threading wasn't de rigueur! So I have no way of judging what amount of effort it takes to move the software threads to separate physical cores.

[imacken]

Not always. Different HW specs have different bottlenecks. always have, always will.

 

And different software settings. That has always been my point. DCS is not CPU bound per se.

[Katj]

Not always. Different HW specs have different bottlenecks. always have, always will.

 

But DCS fires up 50-64 software threads. All of which runs on one core (maybe two, I'm told the audio portion runs on a separate core). And resource meter bears this out.

 

But again, I understand the complexity of getting the scheduling right for multiple threads to coordinate. I'm afraid when I took Operating Systems, multi-threading wasn't de rigueur! So I have no way of judging what amount of effort it takes to move the software threads to separate physical cores.

It's not that all threads run on the same core, the operating system will see to that, it's that almost all the important work is done on only one thread.

 

Think of it this way; a supermarket has 50-64 cash registers, each with a queue, (threads) and 8 cashiers (cores). The cashiers run between the cash registers to serve any arriving customers. Do the number of cash registers or cashiers matter much when there is a guy with an Eagle Dynamics t-shirt that directs 99 % of the customers to the same cash register?

 

The answer is no. Only one cashier at a time will be able to do any meaningful amount of work while the others update their Tinder profiles on their phones.

 

It's directing the customers (the work to be done) to different cash registers (threads) that is hard, because it turns out it's black friday and they're all trying to buy the same TV.

[Headwarp]

Thanks for that. I wasn’t looking for help, I just get fed up with some people stating that DCS is CPU bound per se.

 

I get you, just thought it might be worth checking out. ;) The nVidia default fan curves favor lower decibel operation as opposed to performance.

Edited September 29, 2020 by Headwarp

[Thinder]

I wonder how many players complaining about DCS being an issue are aware of the limitation of their own PCs?

 

I just checked on the setting of one and I find this:

 

Your graphic card is too weak for this processor on 2160p/4K resolution

 

Average bottleneck percentage:

 

6.54% at 1080p

 

8.54% at 1440P

 

10.09% at 2160p/4K.

 

Two thing: First I will not commit to Nvidia "new" generation of GPU, there is little new in the technology but a huge increase in size and power use.

 

Second, Instead of trying to squeeze more performance out of a system by buying new gear, I'm focused in fine tuning what I already have.

 

Meaning before buying anything, reducing the bottleneck to a minimum at the resolution I'm planning to use it, that would be 7.57% at 2160P/ 4K with the GPU O.C at 105%. This in view of playing DCS in VR with the HP Reverb G2.

 

The Ryzen 5 3500X work well with DCS, the weak point here is my 1080 Ti 11GB which is no slouch itself so since I've got it second hand at relatively low cost, it's gonna be the subject of my next upgrade.

 

That's not going to be Nvidia.

 

 

......

[A2597]

@Thinder, similar place here. R5 3600, GTX 1080 (non Ti).

There's a small part of me that wants to get a GTX 1080ti and Reverb G2, but I think I'm going to keep saving for a little bit longer, and see what happens over the next year, both Headset wise and graphics card wise.

[Thinder]

@Thinder, similar place here. R5 3600, GTX 1080 (non Ti).

There's a small part of me that wants to get a GTX 1080ti and Reverb G2, but I think I'm going to keep saving for a little bit longer, and see what happens over the next year, both Headset wise and graphics card wise.

 

Right now, I'm limited by the CV1 and I can't really tell if the 1080Ti is capable of handling the G2.

 

I'll make my decision once I got the G2 and by then, the new generation of AMD GPU will be out, when I upgrade I'll base my choice on the best bound for the R5 3600X.

 

If it's not done already, try playing around a little with Ryzen Master, there are Game modes and Auto Overclocking modes that makes it easier and safe to fine tune your system.

 

Simplest setting: 4,500MHz Maximum.

 

 

......

Edited October 2, 2020 by Thinder

[speed-of-heat]

I run the g1 reverb on a 1080ti my settings are below. A lot will depend on your cpu frame times but I get acceptable, by no means perfect,m performance on a 1080ti The g1 and g2 drive the same number of pixels and so you will get similar performance.

[Thinder]

I run the g1 reverb on a 1080ti my settings are below. A lot will depend on your cpu frame times but I get acceptable, by no means perfect,m performance on a 1080ti The g1 and g2 drive the same number of pixels and so you will get similar performance.

 

 

:thumbup:

 

That's good news, thanks!

 

At the moment I run tests after tests (3D Mark and Heaven Benchmarks) to figure the optimum settings for my 1080 Ti, the 3600X is sorted.

 

 

......

[Csgo GE oh yeah]

What are you guys doing running steam supersampling @ only 150% with a 3080 ?

 

You're not going to see much of a difference indeed like that.

 

Put that slider on 300% , textures high, visibility extreme, Msaa on .

 

Then test with your 1080 ti first, and then with the 3080.

 

edit* I can see that in multiplayer or missions with A.I the performance gain will be a lot less. Holy crap just getting into a very simple multiplayer pvp mission with not a lot of A.I at all will put SO much load on only 1 core. It's a little sad actually :(

Edited October 2, 2020 by Csgo GE oh yeah

[Baldrick33]

edit* I can see that in multiplayer or missions with A.I the performance gain will be a lot less. Holy crap just getting into a very simple multiplayer pvp mission with not a lot of A.I at all will put SO much load on only 1 core. It's a little sad actually :(

This has always been the concern, the results depend a lot upon what you like to use DCS for. Free flight, AI furballs and multiplayer will have very different outcomes depending on just how much the CPU is stretched.

 

Of course there is nothing wrong with having the best hardware and there will always be gains to be made but I think we just need to manage expectations.

[boedha68]

It's in! MSI gaming trio 3080 card came in this morning with the Postman. At last. So time to test myself. :D Can I do Magic? We'll see... :D

 

First impressions.

 

Not the ultimate solution!

 

yes, you can take dcs settings little further like MSAA and distance view, full shadows.

 

But FPS is not much higher then before.

 

Statement!

 

Please ED go for another game engine that makes vr great again. :D

 

thanks.

Edited October 3, 2020 by boedha68