More VRAM helps a lot with DCS VR performance

[Avolate]

Its a shame Nvidia cancelled the 3070 16gb version. With AMD's new GPU's coming all with 16 gb of Vram and all the Consoles also having 16 gb of Vram it seems that Nvidia is really behind.

[Csgo GE oh yeah]

Did some actual 'measurement' with some "DLL hooking" thing called 'special K' . I could not for the life of me get it to work with DCS, but i did get it to work with Resident Evil 3 remake.

The number that gpu-z, and windows show looks like allocated, not used.

 

Settings used in RE3 were 3440x1440 (+ 200% supersampling) // textures highest at '8gb' whatever that means, and all other settings set to highest as well (smaa, hboa, all the good stuff).

I only had one saved game (saved over the same game many times :( ), but in bigger areas you can get the Vram usage up to 8 GB in that game.

 

I have no idea if this translates to DCS at all, considering RE3 is a modern game, and DCS is ... not. I do know that the numbers you see with gpu-z and windows task manager are bullshit though.

 

I also do not know why in DCS, when pressing control + pause, the 'mem' thing shows 0.0 mb and not give us a number. Maybe ED don't have a clue themselves as well.

[Gunnars Driver]

The number that gpu-z, and windows show looks like allocated, not used.[/Quote]

 

Why would there be allocated memory on the GPU VRAM that is dedicated to the GPU?

 

The term allocated comes from when the gpu is sharing regular RAM/DRAM and need to define how much should be used by the GPU.

Intel HD graphics would be an example.

 

 

I do know that the numbers you see with gpu-z and windows task manager are bullshit though.

 

Windows numbers is correct after Fall Crestors update. At least this is what microsoft Direct X states on their developers blog:

 

The memory information displayed comes directly from the GPU video memory manager (VidMm) and represents the amount of memory currently in use (not the amount requested). Because these are exposed from VidMm this information is accurate for any application using graphics memory, including DX9, 11, 12, OpenGL, CUDA, etc apps.

 

https://devblogs.microsoft.com/direc...-task-manager/

 

 

 

 

 

 

 

 

 

 

 

 

 

[Csgo GE oh yeah]

That ^ article is 3,5 years old. And it's incorrectly worded. It displays the exact same number as gpu-z and the other programs.

 

""A few examples - if you look at Shadow of the Tomb Raider, Assassin’s Creed Odyssey, Metro Exodus, Wolfenstein Youngblood, Gears of War 5, Borderlands 3 and Red Dead Redemption 2 running on a 3080 at 4k with Max settings (including any applicable high res texture packs) and RTX On, when the game supports it, you get in the range of 60-100fps and use anywhere from 4GB to 6GB of memory.""

 

And with that in mind, you honestly think that DCS is 'using" 10GB of Vram, while sitting in the splash screen ? Come on man that's impossible :d

 

I've PM'd 'bignewy' , i hope he can walk up to a dev and get some ballpark number, or SOMETHING.

[Gunnars Driver]

That ^ article is 3,5 years old. And it's incorrectly worded. It displays the exact same number as gpu-z and the other programs.

[

 

For me it doesnt give the same numbers.

GPUz and fpsvr show the VRAM used by one process.

performance tab in task manager gives total use, which covers what seems to be windows use + the gpuz/fpsvr value.

 

In all of these the reported VRAM usage varies( se an earlier post in this thread. If a game would really only use 5 or 8 Gb, why would the reported VRAM say for example 16 Gb and also vary during gaming?

 

The age of the article doesnt matther. When microsoft implemented this feature to the task manager (Fall creators update; spring 2018) this feature was implemented.

I’d say a lot of this discussions has its origin in some missunderstandings.

For me, the windows developers worlds is more thrustworthy.

 

 

 

 

[Csgo GE oh yeah]

I am confident that ED will clear all this right up . (press x to doubt)

[Csgo GE oh yeah]

Since ED can't answer any questions about Vram needed for this game

So upcoming weekend, i am going to test it myself , with a 3080, and a 3090.

 

If any of you have specific scenario's they'd like to compare (3080 vs 3090) just post it in this thread.

[NCurk]

I would be really interested in your results. I am still undecided between 3080 or 6800xt, so at this moment I think I might go for the 16GB VRAM instead of 10GB as I think more VRAM is very useful in VR but having no proof. Thank you a lot in advance.

[Gunnars Driver]

Just got a link from a friend about MSI Afterburner beta being able to show VRAM usage for each process:

 

 

I havent had the time to look it trough

 

 

[RealDCSpilot]

Hmmm, i posted this: https://forums.eagle.ru/filedata/fetch?id=6677134 over a month ago. It shows dedicated VRAM usage of the DCS process from 0 to 24576 MB on very high settings and Through the Inferno Syria in DCS on OB from that time (mid of October). You can clearly see the difference between allocated VRAM and real usage.

[some1]

If any of you have specific scenario's they'd like to compare (3080 vs 3090) just post it in this thread.

 

It would be interesting to see how the same scenario looks in fpsVR. If the frametimes are really better and more stable on a card with more vram.

 

If the performance stays the same, then the whole argument about which program measures what, is rather moot.

 

 

 

[Csgo GE oh yeah]

Just a short test i did right after installing it. Frametimes are a bit better. Going from 20 to 18 , or from 18 to 15 ... In that range.

While i was flying though in noticed that 'used' (or allocated) Vram was 19 GB . I'm not kidding you . I just refuse to believe that is what DCS is actually using. BUT , let's assume it's half.

Sure i play at ridiculous settings (300% supersampling with HP Reverb) , but that's getting really close to 10 GB.

 

So in very short (30 minutes of testing) , it's a LITTLE smoother. And i'm not even sure if that's placebo or not.

Normal FPS is a little higher, but the low dips seem equally low as with the 3080.

 

I also did some multiplayer. Pressing F10 did not take as long. (it was quite fast actually, almost instant) .

Pressing F2 to see other allied aircraft was faster. Altough, after pressing F2 some more, there was still a 'hickup'.

Frametime in multiplayer was also a bit lower though.

But, there were only 21 people on the server. I should test again with 40 people to see if that has a negatvie effect. (i'm 99% sure it will :( )

 

Overall the game felt a little smoother, but honestly if you're contemplating between a 3080 and 3090 for DCS, unless you're swimming in money, i would advise against it. Even though i'd really like to believe it makes a bigger difference.

 

This card is quiet though, and it looks REALLY sweet. (gigabite "vision") . I still might keep it because lots of people want to buy the 3080 from me.

 

 

[Polo]

Just to add my very short 'test' on fpsVR reported GPU VRAM usage...

 

2080ti 11GB, Instant Action F14A take off Turkey (on the Syria map)

- with terrain textures on high I'm seeing 10.7GB GPU VRAM usage, getting 30fps with about 60-65% GPU usage and noticeable stuttering

- with terrain textures on low (only change) I'm seeing 10.5GB usage GPU VRAM usage, getting 45fps at about 80-85% GPU usage and stutter-free

 

Flying around for about 10 minutes with the Terrain textures on low and the intermittent stutter every 30 seconds or so is totally gone as well.

 

Not saying it is definitive and will still keep running on low terrain textures, but seems to have helped here.

[Gunnars Driver]

Just to add my very short 'test' on fpsVR reported GPU VRAM usage...

 

2080ti 11GB, Instant Action F14A take off Turkey (on the Syria map)

- with terrain textures on high I'm seeing 10.7GB GPU VRAM usage, getting 30fps with about 60-65% GPU usage and noticeable stuttering

- with terrain textures on low (only change) I'm seeing 10.5GB usage GPU VRAM usage, getting 45fps at about 80-85% GPU usage and stutter-free

 

Flying around for about 10 minutes with the Terrain textures on low and the intermittent stutter every 30 seconds or so is totally gone as well.

 

Not saying it is definitive and will still keep running on low terrain textures, but seems to have helped here.

 

This is exactly correlating with my experience with my 2080ti, and this was also the reason I had to tune the settings lower to get rid of stutter.

my 2080ti settings was limited ( in DCS only) by the VRAM.

 

We know 30X0 should be better to manage VRAM so 10Gb will probably worl better than 11 om the 2080ti.

 

 

 

 

 

 

[Shortyuk]

 

We know 30X0 should be better to manage VRAM so 10Gb will probably worl better than 11 om the 2080ti.

 

 

 

 

The recent 3080 v 6800xt VR comparison does seem to indicate that the 3080 is outperforming the 16gb 6800xt at the higher resolutions we need for VR in other games. We just need a DCS comparison now to be sure.

 

https://babeltechreviews.com/vr-wars-the-rx-6800-xt-vs-the-rtx-3080-15-vr-games-performance-benchmarked/

 

[RealDCSpilot]

On 11/24/2020 at 11:38 AM, Gunnars Driver said:

We know 30X0 should be better to manage VRAM so 10Gb will probably worl better than 11 om the 2080ti.

A GPU never manages VRAM. It's solely on the dev of the application that uses the GPU to implement memory management routines, up-to-date texture compression techniques, keep track of unused assets and purge them from VRAM and RAM to free space. Most "common" games do that because they are designed for current available hardware (and consoles).

Edited December 9, 2020 by Alec Delorean

[Gunnars Driver]

11 hours ago, Alec Delorean said:

A GPU never manages VRAM. It's solely on the dev of the application that uses the GPU to implement memory management routines, up-to-date texture compression techniques, keep track of unused assets and purge them from VRAM and RAM to free space. Most "common" games do that because they are designed for current available hardware (and consoles).

Well, I’m not sure but I almost think when I read it was written by nvidia. 

 

The new GDDR6X have doubled data rate per clock cycle compared the the RTX2000 GDDR6. 

Theres a new way of I/O of data /RTX IO) that offloads the CPU from decompressing work and also is ”and improving I/O performance by a factor of two.”

 

I do not remember all parts but I did read this: 

 

Quote

NVIDIA RTX IO addresses the bottleneck caused by today’s outdated file systems and storage standards. RTX IO provides new APIs for fast loading and streaming directly from the SSD to the GPU’s memory, as well as support for GPU-based loading and decompression of game assets

 

[RealDCSpilot]

@Gunnars Driver

 

The stuff you mentioned are technical capabilities, they don't help if a developer chooses to use massive amounts of polygons for 3D objects and huge textures for shaders. This will fill up VRAM in no time. As a dev, you always have to watch on VRAM consumption and implement routines to clear up free space in VRAM as soon as possible. Like when you are in MP flying one aircraft and decide to choose to fly another one. The former aircraft's cockpit model and textures do not need to stay in VRAM anymore. DCS has very poor memory management atm.

Edited December 10, 2020 by Alec Delorean

[Kercheiz]

DCS stutters because it's mostly single thread. Any operation (such as network I/O, loading textures on the fly, computing game physics, etc...) is blocking others.

 

It's possible that more VRAM causes less frequent textures loads in-flight, thus less stutters.

[Lurker]

Why does VR use up more CPU cycles? I mean for DCS and 2d gaming, you really do not need a very beefy rig to get really great frames. However in VR it seems to be more taxing on the CPU and the rest of the system by quite a margin. 

[imacken]

1 hour ago, Lurker said:

Why does VR use up more CPU cycles? I mean for DCS and 2d gaming, you really do not need a very beefy rig to get really great frames. However in VR it seems to be more taxing on the CPU and the rest of the system by quite a margin. 

Not again.  DCS is not more taxing on the CPU than GPU in VR.  It totally depends on your hardware and settings as to whether it is GPU or CPU bound.

I personally only see CPU issues on very rare occasions, like on the deck of the SC. At all other times, it's the GPU that holds things back at the settings I use.

Edited December 10, 2020 by imacken

[Lurker]

20 hours ago, imacken said:

Not again.  DCS is not more taxing on the CPU than GPU in VR.  It totally depends on your hardware and settings as to whether it is GPU or CPU bound.

I personally only see CPU issues on very rare occasions, like on the deck of the SC. At all other times, it's the GPU that holds things back at the settings I use.

 

 

Please read my post again. I'm asking why CPUs are taxed more in VR, compared to CPUs in 2d? Nowhere did I mention GPUs. 

[imacken]

3 minutes ago, Lurker said:

 

Please read my post again. I'm asking why CPUs are taxed more in VR, compared to CPUs in 2d? Nowhere did I mention GPUs. 

Apologies, I misunderstood your statement!

[S-GERAT]

1 hour ago, Lurker said:

 

Please read my post again. I'm asking why CPUs are taxed more in VR, compared to CPUs in 2d? Nowhere did I mention GPUs. 

Because the CPU need to calculate two different viewpoints for 3D

Edited December 11, 2020 by S-GERAT

[imacken]

1 hour ago, Lurker said:

 

Please read my post again. I'm asking why CPUs are taxed more in VR, compared to CPUs in 2d? Nowhere did I mention GPUs. 

Sorry, still not answering your point!

 

Edited December 11, 2020 by imacken