Nvidia teases a 21 day countdown to unveil or release of the RTX 3000

[chiliwili69]

7 hours ago, Jaws2002 said:

If anyone makes a useful track, please share it. 

 

It would be nice if all of us can use the same track, so we can find the best combination of hardware.

 

Well, the problem with tracks is that they could stop working with new updates, so all previous work is not comparable.

 

This end of the year is quite special with the arrival of exciting CPU (Intel vs AMD), exciting GPU (AMD vs NVidia) and exciting VR devices (G2, 8KX, should I say Quest2?).

 

Many of us are looking at them and want to max performance/eye candy for our own particular budget. Once we acquire the hardware we also want to be sure it will work as expected, so a benchmark will be useful as well. And also to tune our settings by adjusting the most fps-eater options.

 

I think it would be good to know if a simple in-game benchmark would be in the future roadmap of IL-2 . I think we could openly ask this to Jason or Dev team just to know if this is something they would consider for a short-term, mid-term or perhaps never.

 

Two years ago I created a poll for that: 

 

I very recently made a post in suggestions: https://forum.il2sturmovik.com/topic/66380-create-a-benchmark-to-study-best-hardwaresettings-for-this-game/

 

 

@Alonzo, @dburne, @Jaws2002, @Voyager, @SCG_Fenris_Wolf, @Gambit21  How we can kindly ask this to Jason? Should we create an specific post for this or better a PM? a formal hand written letter coupled with a good bottle of wine?? what is the best way just to ask the question?

 

I am know how limited is the dev team and how many things are demanded by community and I will understand whatever answer they may give us. I only wish to know what is their idea about that.

Edited November 6, 2020 by chiliwili69

[HunDread]

@Alonzo since I know you are a pro with the mission editor is it possible to create a bench mission? I'm thinking about something like an airstart with a bunch of friendly bombers around without any orders than to fly low and straight over an enemy AF stuffed with enemy AA guns, then maybe into a big city.

 

Not trying to put any pressure on you that you should do it, just curious whether this could work.

[Ribbon]

8 hours ago, Bernard_IV said:

I think if you have a 3080 or better it should be enough to constantly pin a reverb at 90fps at max, or near max, settings so long as the CPU will allow it.

I hope so, i have reverb g2 on preorder coming by end of the month while still waiting on pc upgrade......my original plan was i7 10700k with rtx3080 but now i'm considering amd aswell...what ever can run g2 on high settings!

[dburne]

In my case I am not sure a bench of in game in VR will tell a  lot.

Currently with my rig and my Rift S, I am typically maxed out at 80 fps for the bulk of the time only dropping into ASW occasionally.

So let's say the new 3090 provides me constant 80 fps and that does not represent much of an increase.

 

Of course when Reverb G2 gets here I expect it to be more demanding and little lower in the performance department than Rift S.

[HunDread]

1 hour ago, dburne said:

In my case I am not sure a bench of in game in VR will tell a  lot.

Currently with my rig and my Rift S, I am typically maxed out at 80 fps for the bulk of the time only dropping into ASW occasionally.

So let's say the new 3090 provides me constant 80 fps and that does not represent much of an increase.

 

Of course when Reverb G2 gets here I expect it to be more demanding and little lower in the performance department than Rift S.

 

That is the point of a benchmark. To create something that would tax any system much more beyond normal use so they cannot run max FPS. Then you would be able to test how much you have with your current setup and the new one.

[dburne]

1 hour ago, HunDread said:

 

That is the point of a benchmark. To create something that would tax any system much more beyond normal use so they cannot run max FPS. Then you would be able to test how much you have with your current setup and the new one.

 

I personally think it is better to do any benching in 2d Mode on the monitor, rather than in VR itself.

Because with VR we are dealing with ASW or Reprojection, which does not cover a full fps range but rather a limited fps range.

Even if one reaches full non-asw or non-reprojection, fps is capped at the device refresh rate. On monitor it is not.

One can typically get a clearer picture IMHO by using 2d on monitor and getting the full rage of fps during the bench.

Typically one can gauge an app VR affect as it usually is around half of the fps on 2d monitor.

Edited November 6, 2020 by dburne

[HunDread]

1 hour ago, dburne said:

 

I personally think it is better to do any benching in 2d Mode on the monitor, rather than in VR itself.

Because with VR we are dealing with ASW or Reprojection, which does not cover a full fps range but rather a limited fps range.

Even if one reaches full non-asw or non-reprojection, fps is capped at the device refresh rate. On monitor it is not.

One can typically get a clearer picture IMHO by using 2d on monitor and getting the full rage of fps during the bench.

Typically one can gauge an app VR affect as it usually is around half of the fps on 2d monitor.

 

This is true. Benchmarking in 2d is more accurate. Doing it in VR can still show some good data but there are more dependencies so it's less accurate. 

[Bernard_IV]

Benchmarking in 2d won't do it as the VR really swamps the processor.  The new AMD chips look to be getting to 5ghz or better so I think they will perform extremely well.  The new AMD GPUs look extremely impressive on top of that.  They are killing it.  I think with any of these new high end parts and a 5ghz processor you will be very happy with the Reverb.  I ran the reverb on a 1080TI and it was quite good with clouds set to low and no antiailiasing.  You really don't need the AA with the impressive resolution of the reverb but I think you might be able to have it all with the new graphics cards.  I know the 3080 is almost double the throughput of  the 1080TI.

[Alonzo]

6 hours ago, HunDread said:

@Alonzo since I know you are a pro with the mission editor is it possible to create a bench mission? I'm thinking about something like an airstart with a bunch of friendly bombers around without any orders than to fly low and straight over an enemy AF stuffed with enemy AA guns, then maybe into a big city.

 

Not trying to put any pressure on you that you should do it, just curious whether this could work.

 

I don't feel any pressure, feel free to ask! ?

 

But to answer the question, I did a "benchmark" mission a while back. My intention was to have some action and some Ai planes, and have the player hit "autopilot" at the start of the benchmark. Then the Ai would fly their plane as well as all the others. Unfortunately due to the randomness of Ai (which is a good thing for the game!) each 'run' of the mission produced different FPS results. You could use this kind of thing for setting up a single rig, because the goal there is to tune and hit 90 FPS 100% of the time, and it could help you with that, but it doesn't produce a repeatable benchmark that can compare different systems.

 

What we could do is make a mission like that and use a 2 minute recording from it as a benchmark track. Then as long as the track was playable, it would be a good benchmark. But that's kind of similar to Chili (or whoever) making an 'unofficial' benchmark track for each patch of the game.

[SYN_Vander]

It's also possible to have a mission without a player, just AI. All you have to do is then setup some timers and cameras so you can focus on certain areas of the map (high, low, from cockpit, flyby, fixed camera) etc. You probably want to avoid a dogfight as indeed randomness will then give different results, but two formations flying in- and out of each other should be possible and repeatable.

[Gambit21]

I was planning on making a benchmark mission on the Rhineland map since I'm in the middle of gathering parts for a new system.

Here's what I'm thinking.

 

A 10-12 minute or so flight with a player aircraft (which you will set to autopilot) amongst several formations of AI B-25's flying over Bonn. 

Double decker clouds just overhead, (or with the flight between layers)  lots of AAA, lots of smoke. All aircraft set to invulnerable, no external cams, no enemy unit cams etc. (everything set to "NOICON")

You're either in the cockpit or better yet IMO, chase cam so that more things are visible.

 

This setup will yield some consistency over the course of 10 min or so since no fighters/air combat is involved.

 

Further, subtitles set to go off at 2 min, 5 min, 8 min, 10 min so that I can check time dilation, which is a more important thing to measure than FPS in most cases.

However the mission will be heavy enough that it should be an FPS indicator as well.

 

Now for you guys to benchmark FPS against each other, that's more problematic (and not really the intent of the mission) because that means everyone would have to remain with the same camera activated the entire time. Cockpit, vs chase cam etc, and the exact same graphics settings. However you can compare time dilation easily enough. 

 

It's not a promise as I have a lot going on right now, but I can take the Hell Hawks mission 3 which already has most of these things in place and just delete a bunch of crap that's not needed, change the cloud deck, create an air start near Bonn, add more AAA and smoke and I'm pretty much set. So doable.

 

 

 

I could also record a track of it, then update the track later if needed rather than requiring everyone to run the mission.

This would help with consistency I guess.

[SYN_Vander]

but with AI only and automated camera views you *could* have a benchmark that can be compared between different setups right?

[Gambit21]

23 minutes ago, SYN_Vander said:

but with AI only and automated camera views you *could* have a benchmark that can be compared between different setups right?

 

Correct, cameras with animation operator MCU's I believe. I've never done it, but possible.

I'm not sure if there's a way to slave the players view to a particular static camera otherwise, someone else might know.

[SYN_Vander]

2 minutes ago, Gambit21 said:

 

Correct, cameras with animation operator MCU's I believe. I've never done it, but possible.

I'm not sure if there's a way to slave the players view to a particular static camera otherwise, someone else might know.

Yeah, you can also set the camera to cockpit view if that is what you mean, but ...... don't use trackir or VR because it may not be centered, but looking back or something

I'll see if I can throw together a skeleton mission.

Edited November 6, 2020 by SYN_Vander

[Alonzo]

1 minute ago, Gambit21 said:

Correct, cameras with animation operator MCU's I believe. I've never done it, but possible.

I'm not sure if there's a way to slave the players view to a particular static camera otherwise, someone else might know.

 

It's also possible (I don't know without measuring) that the in-cockpit view is more demanding than a chase plane view, since you're rendering the cockpit instruments in detail and might LOD them out with an outside camera. I do like the idea of a mostly-repeatable benchmark mission though. Maybe a low-level flight as part of a big formation, with as you said lots of AA and puffy flak and smoke and ground fire and stuff.

 

With a player plane, if you set it to auto-pilot it'll follow mission waypoints and stuff, right? Would it even fly in formation if you linked the 'Player' plane to a flight lead? That could be pretty good, having a player in a bomber formation maybe one or two planes back.

[Gambit21]

Yep - the player AC will fly formation in auto-pilot. That was the plan.

 

You may be correct regarding the cockpit view being the better idea.

[dburne]

I will do a 2d benchmark using a DX11 benchmark.

If I run a premade mission I will run it with the settings I am now, and compare the 2080 Ti to the 3090 in 2d.

This will give a good comparison in average fps between the two

The main thing is to see the improvement between the two. IL-2 is DX11 so that should give a good comparison.

 

As I said I believe a VR run just will not give a good representation. With the same settings , VR is on average around half.

Not going to spend too much time on it, as I will want this 2080 Ti out of there and the 3090 in as soon as possible.

[Ribbon]

8 minutes ago, Alonzo said:

 

It's also possible (I don't know without measuring) that the in-cockpit view is more demanding than a chase plane view, since you're rendering the cockpit instruments in detail and might LOD them out with an outside camera. I do like the idea of a mostly-repeatable benchmark mission though. Maybe a low-level flight as part of a big formation, with as you said lots of AA and puffy flak and smoke and ground fire and stuff.

 

With a player plane, if you set it to auto-pilot it'll follow mission waypoints and stuff, right? Would it even fly in formation if you linked the 'Player' plane to a flight lead? That could be pretty good, having a player in a bomber formation maybe one or two planes back.

Yup, highest load you'll achieve from cockpit view and that is inside he111 cockpit (at least in VR)!

To avoid AI randomization maybe spawning player in he111 (cockpit view) as a leader of 8×He111 or more flight without any enemy plane.....low alt near the city or between Kuban hills.

Thats how i test my settings vs. performance!

 

[Gambit21]

No need to start the player as a leader.

[SYN_Vander]

Okay, here is what I have. Unfortunately, you can't have a cockpit camera without a player (makes sense), but you can have other fixed camera views.

 

 

Benchmark_example.zip

 

[Gambit21]

3 minutes ago, SYN_Vander said:

Okay, here is what I have. Unfortunately, you can't have a cockpit camera without a player (makes sense), but you can have other fixed camera views.

 

 

Benchmark_example.zip 199.83 kB · 0 downloads

 

 

Right, you can place static cameras all day long and cycle between them at will. However doing this will lead to inconsistent results.

 

Thus it’s best to stick with a single camera/cockpit I think.

 

A test mission with a single external, fixed camera and some large formation flybys is also a possibility.

[SYN_Vander]

True: the flyby camera for instance seems to do something different every time you run it.

[chiliwili69]

3 hours ago, Gambit21 said:

Further, subtitles set to go off at 2 min, 5 min, 8 min, 10 min so that I can check time dilation, which is a more important thing to measure than FPS in most cases

Sorry for my ignorance but, what is time dilation? and why it is important to measure?

3 hours ago, Gambit21 said:

I could also record a track of it, then update the track later if needed rather than requiring everyone to run the mission.

This would help with consistency I guess.

 

What are the pros/cons of using the mission vs. the track for benchmark?

I understand that a mission can "live" along the future version of IL-2, right? (Not really familiar with mission builder)

[Alonzo]

14 minutes ago, chiliwili69 said:

Sorry for my ignorance but, what is time dilation? and why it is important to measure?

 

What are the pros/cons of using the mission vs. the track for benchmark?

I understand that a mission can "live" along the future version of IL-2, right? (Not really familiar with mission builder)

 

Time dilation is (I think) where simulation time starts diverging from wall-clock time. I think Gambit has some specific approaches to test for this, as he builds very big single player missions with a lot of moving pieces. For me in multiplayer, the equivalent is probably failing to hit 50 simulations per second on the server.

 

A mission, in theory, will always work, no matter if the game is updated. It could also run a lot longer than a recording. If we use external cameras it also may be more reliable/repeatable than an in-cockpit view, and it might work better with VR.

 

I know several folks in this thread have said benchmarking in 2D is better, and I won't disagree that yes, you'll probably get more reproducible numbers. But VR is the only reason I'd bother benchmarking or tuning, you're going to hit 100+ FPS on any reasonable rig in 2D, and the two modes are going to demonstrate very different scaling and bottlenecking behaviors. "3090 is 2x faster than 2080ti in 2D" is sort-of useful I guess, but I'd really prefer to know the VR numbers.

[thermoregulator]

Managed to get EVGA 3090 FTW Ultra. According to fpsvr, I am probably still gpu limited with ryzen 5900x and HP Reverb, 118% steamvr SS.

[Bernard_IV]

Don't run super sampling on a Reverb, there is no benefit.  I got my 3080 but have sold my previous reverb so I'm waiting on the new one.  These things are super fast!

On 11/6/2020 at 1:13 PM, chiliwili69 said:

Sorry for my ignorance but, what is time dilation? and why it is important to measure?

 

 

People experience time differently based on where you are in the word, how much gravity you are experiencing, and how fast you are going!  A notable example is GPS satellites that have to constantly adjust their clocks to match the time on earth, it moves faster up there.  Trippy stuff but true!

[chiliwili69]

I have tested the mission benchmark provided by SYN_Vander, I have run 3 tests at 4.4 GHz and another 3 tests running at 4.8 GHz.

The results are:

 

At 4.4GHz
Frames: 4589 - Time: 42000ms - Avg: 109.262 - Min: 84 - Max: 145
Frames: 4524 - Time: 42000ms - Avg: 107.714 - Min: 72 - Max: 144
Frames: 4598 - Time: 42000ms - Avg: 109.476 - Min: 83 - Max: 145

 

At 4.8GHz
Frames: 4963 - Time: 42000ms - Avg: 118.167 - Min: 95 - Max: 155
Frames: 5050 - Time: 42000ms - Avg: 120.238 - Min: 94 - Max: 154
Frames: 5016 - Time: 42000ms - Avg: 119.429 - Min: 93 - Max: 150
 

The mission takes 42 seconds and I used fraps.

I run it at 1080p with everything max out except clouds at low. Just to stress only CPU/RAM.

 

The plot is:

 

There is some small variability. I think it is explosions nearby.

I don´t know how to limit the random effects to decrease the variablity.

 

We can run a longer track but I like having a short quick benchmark. This one is 42 seconds. Perhaps 90 second would be fine.

 

Another option would be to run it 3 times and calculate average.

[SYN_Vander]

2 minutes ago, chiliwili69 said:

I have tested the mission benchmark provided by SYN_Vander, I have run 3 tests at 4.4 GHz and another 3 tests running at 4.8 GHz.

The results are:

 

At 4.4GHz
Frames: 4589 - Time: 42000ms - Avg: 109.262 - Min: 84 - Max: 145
Frames: 4524 - Time: 42000ms - Avg: 107.714 - Min: 72 - Max: 144
Frames: 4598 - Time: 42000ms - Avg: 109.476 - Min: 83 - Max: 145

 

At 4.8GHz
Frames: 4963 - Time: 42000ms - Avg: 118.167 - Min: 95 - Max: 155
Frames: 5050 - Time: 42000ms - Avg: 120.238 - Min: 94 - Max: 154
Frames: 5016 - Time: 42000ms - Avg: 119.429 - Min: 93 - Max: 150
 

The mission takes 42 seconds and I used fraps.

I run it at 1080p with everything max out except clouds at low. Just to stress only CPU/RAM.

 

The plot is:

 

There is some small variability. I think it is explosions nearby.

I don´t know how to limit the random effects to decrease the variablity.

 

We can run a longer track but I like having a short quick benchmark. This one is 42 seconds. Perhaps 90 second would be fine.

 

Another option would be to run it 3 times and calculate average.

 

Mine was just an example. We need: other location (stalingrad has random explosions), skip the flyby as it also has some randomness, longer duration and a lot more vehicles and aircraft to increase load.

[chiliwili69]

5 minutes ago, SYN_Vander said:

Mine was just an example. We need: other location (stalingrad has random explosions), skip the flyby as it also has some randomness, longer duration and a lot more vehicles and aircraft to increase load.

 

OK, great then!.

You know well what locations could be good.

Last Remagen track was 180 seconds and I think it was a bit long. If there is no variability it doesn´t need to be very long. I think 90 or 120 seconds would be more than fine.

 

Thanks so much.

[SYN_Vander]

Here is version 2. I chose a free map so anyone can test.

 

Wait for all the buildings to load before hitting 'P'.

 

Benchmarkv2.zip

[chiliwili69]

Thanks again for this new mission benchmark. It is good to have it in a Free map, so everyone can run it.

I see now it is much more dense in terms of buildings and planes, and fire and smoke. I think it is quite dense, perhaps too much but OK.

It is also nice the lenght, 46 seconds, so it is good when people do multiple runs to compare settings.

 

I have run it 9 times in the following way:

 

1. CPU at 4.4GHz, all max out except clouds (Low) and no AA. So GPU is not loaded.

Frames: 2697 - Time: 46000ms - Avg: 58.630 - Min: 21 - Max: 96
Frames: 2778 - Time: 46000ms - Avg: 60.391 - Min: 30 - Max: 95
Frames: 2728 - Time: 46000ms - Avg: 59.304 - Min: 29 - Max: 95

 

2. As 1 but CPU at 4.8GHz
Frames: 2910 - Time: 46000ms - Avg: 63.261 - Min: 33 - Max: 100
Frames: 2945 - Time: 46000ms - Avg: 64.022 - Min: 34 - Max: 102
Frames: 3018 - Time: 46000ms - Avg: 65.609 - Min: 34 - Max: 100

 

3. As 2 but with no shadows, no mirrors (there is no mirrors in the views), no HDR&SSAO (they don´t affect CPU)
Frames: 3534 - Time: 46000ms - Avg: 76.826 - Min: 36 - Max: 131
Frames: 3499 - Time: 46000ms - Avg: 76.065 - Min: 43 - Max: 124
Frames: 3530 - Time: 46000ms - Avg: 76.739 - Min: 44 - Max: 122
 

All of them still show some variability, now a bit less than in the first example (no ramdom explosions now).

This is the plot:

 

 

If I do the average of the three runs of each group then we can clearly see the different performance:

 

 

In the past we also had some variablity (no more than 1 fps up or down) using recorded tracks.

 

We never knew what was causing this variability.

 

[SYN_Vander]

If you want to avoid the spikes I can do only one camera from start to finish. I'm think about the external camera that locks onto another plane.

And should I get rid of the 'huge' smoke perhaps?

 

Here it is:

Benchmarkv3.zip

Edited November 8, 2020 by SYN_Vander

[chiliwili69]

Hey!, Thank again for this new mission.

Yep, it is better to have the same camera view to avoid the transition spikes. Having less fire is also better. This is getting better and better.

 

I run 3+3 test at different OC with all max out except Clouds (low) and no AA, no HDR, no SSAO. It is 1080p.

The GPU load for my 1080Ti is 30% reaching 40% during some seconds in the middle of the mission.

 

4.4GHz
Frames: 2775 - Time: 46000ms - Avg: 60.326 - Min: 32 - Max: 99
Frames: 2798 - Time: 46000ms - Avg: 60.826 - Min: 38 - Max: 100
Frames: 2772 - Time: 46000ms - Avg: 60.261 - Min: 37 - Max: 98

 

4.8GHz
Frames: 3010 - Time: 46000ms - Avg: 65.435 - Min: 41 - Max: 108
Frames: 2949 - Time: 46000ms - Avg: 64.109 - Min: 42 - Max: 105
Frames: 2954 - Time: 46000ms - Avg: 64.217 - Min: 41 - Max: 103

 

At 4.4GHz the variability in avg values is really small (just 0.5 ffps).

At 4.8 we have a bit more.

 

The inital high altitude view produce very high number (105fps) and there is a big step down when switching to the IL-2 view. You put the high view to see when all buildings are loaded but then we have this big step. Is there any way to see when all buildings are loaded form the IL-2 view?

 

Also, is there any way to prevent random bombing from planes?

.

Also, another thing. Could you make the mission a little bit longer. The variability effects will afect more in a short track.

Something around  80 seconds instead of 47.

[chiliwili69]

17 hours ago, chiliwili69 said:

Is there any way to see when all buildings are loaded form the IL-2 view?

 

I have seen that the bridges are the last objects to be loaded, so perhaps the IL-2 external view could be looking towards the bridge should be and then people know when they can press the unpause "P". Just a thought.

[SYN_Vander]

3 hours ago, chiliwili69 said:

 

I have seen that the bridges are the last objects to be loaded, so perhaps the IL-2 external view could be looking towards the bridge should be and then people know when they can press the unpause "P". Just a thought.

 

I can start the camera behind the plane and then wait there before starting. It think the big train depot building is one of the last to load; it should be visible I think. There is no random bombing: I use a script that sets off a series of explosions by destroying dummy vehicles. This *should* not be random.

[dburne]

It has arrived.

 

[dburne]

2080 Ti FTW3

Fire Strike Extreme - 17182.

IL-2 47 second Benchmark - FPS - min 48, max 151, avg 85

 

Will test the same when I get the 3090 in.

I would also note this is with my standard GFX settings which are mostly Ultra and High with 2x msaa.

Edited November 9, 2020 by dburne

[dburne]

She is in and humming nicely now. 

 

[dburne]

I don't know gang, something seems off here.

 

Ok first my comparison on Fire Strike Extreme DX 11 benchmark.

2080 Ti FTW3 - 17182.  3090 FTW3 Ultra - 22188 ( GPU stock speed/boost)

 

Now the strange part on the IL-2 Benchmark mission

2080 Ti FTW3 - min 48  max 151  avg 85

3090 FTW 3 Ultra - min 46  max 153  avg 86

 

But - viewing the GPU clock and temp in PX1 - with the 3090, it did not boost at all during the IL-2 run ( it did with 3d Mark) , and the GPU temp stayed at like 32c for the whole benchmark in IL-2 . Running it with the 2080 Ti it did not have this odd behaviour. In 3d Mark my GPU temp was around 60c and the GPU clock was boosting properly.

So not sure why it did this with the 3090 but behaved more as expected with the 2080 Ti.

Something very weird there that I can not explain. 

So I guess this is maybe not any help yet. 

 

Anyway I will just have to see how it does with IL-2 playing in VR, will likely be tomorrow before I can get to it.

[chiliwili69]

7 hours ago, SYN_Vander said:

I can start the camera behind the plane and then wait there before starting. It think the big train depot building is one of the last to load; it should be visible I think.

 

Yes please, we wil see then when all building are loaded.

3 hours ago, dburne said:

IL-2 47 second Benchmark - FPS - min 48, max 151, avg 85

Congrats for your new beast!

 

The 47 seconds benchmark (which is temporary, SYN_Vander will provide final version) can be run in two modes: Stress CPU or stress GPU.

If you are going to test the 3090 vs 2080Ti, you should use the setting to stress GPU.