VRAM used by IL-2 in VR

[chiliwili69]

A recent discussion about the importance of VRAM of the new GPUs, specially the 3080 with 10GB, made me wonder how much VRAM is used (or allocated) by IL-2 in VR.

I made three runs of the same combat (Kuban map, 2vs2, the 4 planes are different, Novorossiysk low fly) with the following settings:

1st run: SteamVR SS at 130%, this is 2296x2552 per eye (so more than native resolution of Reverb). Using High settings with Remagen test settings (it uses 4K textures, FSAAx2). No clouds.

2nd run: Identical to 1st run but increased SteamVR SS to 200%. This is 2848x3168 per eye.

3rd run: It is like 2nd run but maxed out all in-game graphics settings (Ultra, Extreme or ultra everything, including HDR, SSAO, Sharpen, distant building, FSAAx4) and put averaged clouds.

 

You can see that for 1st and 2nd run the maximum VRAM used reported by MSI Afterburner is about 4Gb. So it seems that SS doesn´t affect to VRAM used.

Then in the third run the maximum VRAM ramp up to 6.7 GB.

 

So, based on that I believe that the 3080 should be OK for IL-2 VR in terms of VRAM. 

If anyone is using more than 8GB of VRAM while playing IL-2 VR, please let me know. VRAM usage is also reported in the fpsVR app.

Here the MSI Afterburner trends:

GPU temp, GPU usage, GPU VRAM

 

CPU temp, RAM usage, CPU clock.

 

 

[=RS=haikcube]

my gigabyte 1080ti's vram allocated usage starts at around 8 gigs and  locks in at 10.5gig in multiplayer servers.

my headset is an odyssey plus at steam 120% super sample(1580X1979) while the in game settings are 2xmsaa with balanced settings with ultra shadows,high clouds,medium mirrors along with distance buildings and 4k textures.

The discrepancy between our vram usages ignoring the dissimilar testing situation(controlled benchmark vs multiplayer) might be explained with us having differing driver settings

I doubt the game is actually requiring all 10.5 gigs of ram however and is just asking for it all "just in case".

 

Edited September 19, 2020 by [=PzG=]-haikcube

[Dutch2]

You are absolutely right on this afterburner, GPUz and I guess also fpsVR, are measuring the allocation of Vram. I must admit I never knew this until I saw  this at Gamer Nexus, when reading all the 3080 reviews, start around 3:20.  BTW its my source on selecting hardware, now Tomhardware is being reduced to an commercial advertising site.  

 

 

[-=PHX=-SuperEtendard]

@chiliwili69 nice, thanks for sharing it with us, have you checked how's the VRAM allocation in DCS?

[C6_lefuneste]

What about loading more than 4 planes textures, as you may seen in multiplayer ?

I wonder if gorgeous 4k texture are eating GPU memory...

[chiliwili69]

17 hours ago, [=PzG=]-haikcube said:

I doubt the game is actually requiring all 10.5 gigs

The scale of your graph for VRAM was limited to the default value 8192.

You can change that scale for the VRAM in the afterburner options. In my case I changed it to the 1080Ti VRAM memory which is 11264MB.

So you will be able to see the current usage during play game.

The Max value to 10.5 could be due to a peak or previous games you played. It would be nice if you could check it out again.

2 hours ago, -=PHX=-SuperEtendard said:

have you checked how's the VRAM allocation in DCS?

 

I played DCS in the past, when IL-2 was not in VR... more than for year ago (see below), but since IL-2 was in VR I switched again to IL-2 (my only sim).

 

[=RS=haikcube]

19 hours ago, chiliwili69 said:

The scale of your graph for VRAM was limited to the default value 8192.

You can change that scale for the VRAM in the afterburner options. In my case I changed it to the 1080Ti VRAM memory which is 11264MB.

So you will be able to see the current usage during play game.

The Max value to 10.5 could be due to a peak or previous games you played. It would be nice if you could check it out again.

 

I was not aware that there was a graph limit and assumed it automatically adjusted when reading hardware information, thank you for pointing it out to me.

Anyway similar result,fresh startup using 3 gigs, 8 gigs during game startup and hovers around 10.5 gigs while in game.

I forgot to mention that my monitors native resolution is 4k while i'm running the game client in windowed mode, that is most likely the reason for my increased vram usage.

[E69_Qpassa_VR]

On 9/19/2020 at 11:56 AM, Dutch2 said:

You are absolutely right on this afterburner, GPUz and I guess also fpsVR, are measuring the allocation of Vram. I must admit I never knew this until I saw  this at Gamer Nexus, when reading all the 3080 reviews, start around 3:20.  BTW its my source on selecting hardware, now Tomhardware is being reduced to an commercial advertising site.  

 

 

Tech jesus

[chiliwili69]

On 9/20/2020 at 4:04 PM, [=PzG=]-haikcube said:

Anyway similar result,fresh startup using 3 gigs, 8 gigs during game startup and hovers around 10.5 gigs while in game.

I forgot to mention that my monitors native resolution is 4k while i'm running the game client in windowed mode, that is most likely the reason for my increased vram usage.

 

Many thanks for running again with the corrected scale.

 

It is clear now that it is not just a peak, you VRAM is around 10.5 constantly.

 

I wonder why you require more VRAM. I could see four reasons:

 

1.- The resolution of the monitor window while you are in VR. I also have a 4K monitor but the window is 1920x1080 (and not full screen). Perhaps when developer said to reduce this size for optimum performance they were referring to VRAM usage bottleneck. (I didn´t see any performance bump by going from 1920x1080 to 1024x768, so I use 1920x1080)

 

2.- The WMR software (for your Odyssey) require more VRAM than than the software of the Index (just SteamVR)

 

3.- The MP require more VRAM than SP.

 

4.- There were more variety of planes in your test than in my test. (more textures loaded in GPU).

 

Who can determine which is the reason?

[chiliwili69]

Yesterday I had the chance to do a bit more testing to know what of the four reasons above influence the VRAM usage.

For all tests I used the Remagen test Settings (High preset, 4K textures, etc and 130% SS for Index).

 

The first test was just QMB in Kuban with the monitor window at 1920x1080 (not full screen) in my 4K monitor.

Initially 2vs2 planes  and then 8vs8 planes. Reaching 4.2 GB at the end. So it is aligned with my previous test.

 

Then I switched to 4K resolution for the monitor view and run 3 series of 8vs8 using different planes each time. Here I saw an increase to 4.8GB, but that increment (I believe) is due to changing the planes, not due to using 4K resolution since the first 8vs8 used the same VRAM than the previous 8vs8 in FullHD.

 

Then the last test I went to MP mode in CombatBox dogfight just at the ground airfield with a lot of smoke from a nearby explosion. Here the VRAM rampup to 6.1 GB.

 

Conclusion:

The reason 1 above does not influence VRAM usage (or allocation).

The reason 3 is true. MP seems to require a bit more (about 1Mb). But not sure if this is because I was at ground.

The reason 4 is slightly true, but very small effect for every new plane textures.

 

But I am still far from the 10GB.

 I wonder why haikcube had 8GB just at game start-up (maybe msaa?). Is anyone also using 8GB VRAM at game startup?

 

 

here also show other details like RAM usage which also increases in MP mode.

[Charlo-VR]

In CombatBox last night with around 40 players I typically used 6 to 7 GBs of of my Titan X Hybrid’s memory while flying around, with spikes up to about 9 GBs that seemed to correlate to between missions. Oculus Quest, 1.1 SS in OTT, ASW off, 4K textures, 1920 x 1080 windowed, 4 x FXAA

Edited September 23, 2020 by Charlo-VR
Added 4 x FXAA

[dburne]

When it comes to Vram I think of it much like system ram.

One can never have "too much", while certainly one can have "not enough".

 

I will always prefer to have more than really needed. Plus as much as I love and enjoy IL-2 GB, it will never be the only game I play.

[SCG_Fenris_Wolf]

Thanks chili.

 

These tests basically show that we may be much better off aiming for an RTX 3080 (instead of a 3090) and then probably to use the funds to upgrade to a well overclocking CPU/MB/Memory instead.

[Alonzo]

On 9/23/2020 at 3:49 PM, SCG_Fenris_Wolf said:

These tests basically show that we may be much better off aiming for an RTX 3080 (instead of a 3090) and then probably to use the funds to upgrade to a well overclocking CPU/MB/Memory instead.

 

Yep. For me, I have a Valve Index and an RTX 2080. If I had $1500 USD to spend, I think my best bet would be a nicely performing 3080 ($800), a Reverb G2 ($600) and $100 on maybe an airflow-oriented case.

 

Or spend zero money and just enjoy what I have.

 

Actually, serious question - is it possible to demo a Reverb? Even the original? I'm trying to understand what the resolution upgrade and FOV decrease look like.

[chiliwili69]

19 hours ago, Alonzo said:

Actually, serious question - is it possible to demo a Reverb? Even the original? I'm trying to understand what the resolution upgrade and FOV decrease look like.

 

You can see what I saw when comparing them one year ago:

 

I think the images really reflects what I saw with the device. You can download the raw images. Those images were taken at the centre, so it will be the same in G2.

 

After a very difficult decision process, I kept the Index.

 

Now with G2 three of the reasons should be gone: Sweet spot,  audio and performance (solvable with a better card now, for example 3080).

 

The only thing will be the better FOV of the Index. I will need to check myself I will kept the Index or not. I will know that in November hopefully.

[SCG_Fenris_Wolf]

21 hours ago, Alonzo said:

 

Yep. For me, I have a Valve Index and an RTX 2080. If I had $1500 USD to spend, I think my best bet would be a nicely performing 3080 ($800), a Reverb G2 ($600) and $100 on maybe an airflow-oriented case.

 

Or spend zero money and just enjoy what I have.

 

Actually, serious question - is it possible to demo a Reverb? Even the original? I'm trying to understand what the resolution upgrade and FOV decrease look like.

 

Unless you'll want to upgrade to 3080 + Reverb G2 in one go (which both are delayed), better stay with the Index at this point I'd say. Maybe beginning of next year, when there are decent numbers floating around.

[Charlo-VR]

On 9/23/2020 at 2:49 PM, SCG_Fenris_Wolf said:

Thanks chili.

 

These tests basically show that we may be much better off aiming for an RTX 3080 (instead of a 3090) and then probably to use the funds to upgrade to a well overclocking CPU/MB/Memory instead.

 

In my tests a few weeks ago earlier in this thread I was using an i7-6700K Skylake OC @ 4.7 GHz on an ASUS ROG MAXIMUS VIII HERO motherboard and 32 GBs of G.SKILL Ripjaws 4 Series DDR4 3000 RAM running at 2,666 MHz. At that time in Combat Box with around 40 players I typically used 6 to 7 GBs of of my Titan X Hybrid's memory while flying around, with spikes up to about 9 GBs that seemed to correlate to between missions. Oculus Quest, 1.1 SS in OTT, ASW off, 4K textures, 1920 x 1080 windowed, 4 x FXAA.

 

The other day I upgraded to an i7-10700K OC @ 5.1 GHz on an MSI MPG Z490 Gaming Carbon WIFI motherboard, and 32 GB's of G.Skill Trident Z Royal DDR4-4000 RAM running at 4,000 MHz. I am still using my Titan X Hybrid that has 11GBs of VRAM.

 

I decided to see if my machine can now handle 4x MSAA, and it can, though sometimes my Quest's frame rates dipped to 34fps this evening in a few different maps in Combat Box with around 40 - 70 players. I'm lucky that those frame rates don't make me sick, and I was able to fly a few hours today like that.

 

Of even greater relevance to this thread, GPU-Z indicated I was now using all 11GBs of of my Titan X Hybrid's memory while flying around in Combat Box. Oculus Quest, 1.1 SS in OTT, ASW off, 4K textures, 3840 x 2160 full screen, 4 x MSAA.

[chiliwili69]

On 10/5/2020 at 4:33 AM, Charlo-VR said:

3840 x 2160 full screen, 4 x MSAA

 

Thanks for this info. It is very relevant.

The changes in CPU/Mobo/RAM should not affect the amount of VRAM used in your TitanX.

So, you basically changed 1902x1080 with 4xFXAA to 3840x2160 with 4xMSAA with the same Quest at 1.1SS.

According to my test, going from fullHD to 4K for the monitor view while you are in VR doesn´t increase the VRAM usage (only very little. about 0.5Gb).

So, probably the biggest impact is the 4xFXAA to 4xMSAA, about 4GB more in your case.

 

Can anyone else just confirm this as well?

[=SFG=capt_nasties]

it will be a while but I will definitely do some benchmarking in VR (against my current 2080) when I get a 3080 ( probably holding out for the 20gb version) with my reverb g1.  This is a great thread @chiliwili69 is always coming up with good topics and data to discuss here!  

Edited October 14, 2020 by =SFG=capt_nasties

[Ribbon]

So rtx 3080 10GB won't be enough to run "full" potential of g2?

[E69_Qpassa_VR]

33 minutes ago, =VARP=Ribbon said:

So rtx 3080 10GB won't be enough to run "full" potential of g2?

We don't know unless we have g2 on our hands, I will test with my 1080ti too

[SCG_Fenris_Wolf]

41 minutes ago, =VARP=Ribbon said:

So rtx 3080 10GB won't be enough to run "full" potential of g2?

Yeah, we don't know. From any other application we know though, that it's only allocation. 

 

Based on what I know about the usage and experience benchmarking and gathering stats is that VRAM requirements don't rise, nor are games requiring such insane amounts, or even use it. See Half Life Alyx - it scales its allocation to the available VRAM. So it'll reserve more VRAM on a 3090 than on a 3080, but without using it. Hence, waiting for a 20GB version of the 3080 will be a) wasted time b) wasted money in my opinion.

 

Ultimately, we'll see very soon anyway. 

 

Just, remember, try to avoid using cheapskate programs like GPU-Z. The logging functions are bad, limited, and it doesn't even tell you that it's just allocation, and not use, nor logs throughput properly along time. That is why it is unfit to benchmark or gather statistics properly.

[Ribbon]

1 hour ago, E69_Qpassa_VR said:

We don't know unless we have g2 on our hands, I will test with my 1080ti too

 

1 hour ago, SCG_Fenris_Wolf said:

Yeah, we don't know. From any other application we know though, that it's only allocation. 

 

Based on what I know about the usage and experience benchmarking and gathering stats is that VRAM requirements don't rise, nor are games requiring such insane amounts, or even use it. See Half Life Alyx - it scales its allocation to the available VRAM. So it'll reserve more VRAM on a 3090 than on a 3080, but without using it. Hence, waiting for a 20GB version of the 3080 will be a) wasted time b) wasted money in my opinion.

 

Ultimately, we'll see very soon anyway. 

 

Just, remember, try to avoid using cheapskate programs like GPU-Z. The logging functions are bad, limited, and it doesn't even tell you that it's just allocation, and not use, nor logs throughput properly along time. That is why it is unfit to benchmark or gather statistics properly.

Ok, thnx guys drifting away some doubts i have!

I also preordered G2 and was intend to upgrade pc (gpu, cpu, psu and mobo) but this "out of stock" fiasco with 3080 and scalpers made me postpone it for a while till it get's in stock and within reasonable price range.

If 3080 10gigs will be enough i'll go with it instead waiting for 20gig version for increased price taking in consideration higer price and 20GB (if) overkill is not worth for someone who upgrade pc frequently.

By the time 20GB vram is needed in games 3080 could/wpuld be obsolete card.

Same with intel's rocket lake....i also hope amd and ati will help this situation!

[Alonzo]

I think it's super unlikely that IL2 will require large amounts of VRAM. It's just not a texture-heavy game like some of the AAA shooters out there. For those games maybe you'd worry about 10GB VRAM at 4K or above, but not for IL2.

 

That said, if you can wait another two weeks for AMD's announcement, it's definitely worth waiting. The rumor is for a 16GB nearly-3080-performance card for slightly less money, and a possibility for a better-than-3080 card if they push their clocks and power higher. What AMD do may even affect NVidia's pricing on the 3070, which is deliberately timed to be announced 24 hours after AMD's event, just in case they need to scramble and drop the 3070 price a little.