F14 frame rate thread

Im curious how other CPUs and GPUs perform, particularly with the F14, and you might be too if you are looking for an upgrade, so Im hoping many of you want to do a short simple test and post your results to have some rough ideas. Im particularly keen to find out of there is a noticable difference between AMD and Intel cpu's at similar frequencies, and how AMD GPUs fare compared to otherwise largely equivalent nVidia cards.

- Set your resolution to 1920x1080

- Configure DCS graphics settings to the preset HIGH (ensure vsync is off)

- Restart DCS, and to be sure your new settings are applied, someone suggested you need to delete the fxo and Metashader2 folders.

- in your gpu driver settings, ensure you are not overriding MSAA or FXAA or anything else relevant

- Load the "HB Tomcat Defend The Fleet' mission

- disable your headtracker, or ensure you are looking straight forward

- Zoom out all the way (this lowers FPS and looks silly, but increases consistency)

The viewpoint should look like this:

The base of the stick should still be in frame

- Note your framerates (Lctrl+pause, or use fraps or afterburner to have an averaged value)

- Now enable your mirrors ("M" by default) and note your frame rates again.

post your results along with your hardware specs. Here are mine:

DCS 2.5 open beta 2.5.4.30038

Ryzen5 2600X stock speed, water cooled.

3.6Ghz base clock, 4.2GHz boost, actual clock ~4.1 GHz

16 GB DDR4-2133

Geforce 1070 stock speed

M2 NVME SSD

Without mirrors: 51 FPS

With mirrors: 43 FPS

update:

Since no one else tested this yet, and I still dont have my ram, I just tried to overclock my old ram. From the default 1066 Mhz (DDR4 2133) to 1200 Mhz (DDR4 2400), which is about as high as it will go. CAS latency only took a small hit.

update2: DDR4 3200 arrived

Without mirrors: 51 FPS @2133 ram

Without mirrors: 56 FPS @2400 ram

Without mirrors: 60 FPS @3200 ram CAS 16

Without mirrors: 61 FPS @3466 ram CAS 17

With mirrors: 43 FPS @2133 ram

With mirrors: 47 FPS @2400 ram

With mirrors: 51 FPS @3200 ram CAS 16

With mirrors: 52 FPS @3466 ram CAS 17

• DCS 2.5 open beta 2.5.4.30038

• i7 860 @2.8GHz (Dell Vostro 430/stock cooler)

• 12GB DDR3 (PC3-10700, 667 MHz)

• Gigabyte GTX1050Ti 4GB

• 1TB SanDisk SSD, 4+2TB WD HD

Without mirrors: 24 FPS

With mirrors: 19 FPS

Note: Running with custom Medium settings, does not increase the average FPS but does reduce RAM use and stop stutter during A2A combat, etc.

I'm at work right now, but in VR, with most settings to high (accept MSAA is at 2X), I get a solid 45 FPS when on deck or on the ground with other players (20+) near by. Once in the air, I sometimes can hit upwards of 60 FPS.

my 9700K is overclocked to 5 GHz. Other specs are below. I will see about getting a in game screenshot for you, if I remember to.

I did a little extra testing to see how much this scenario is CPU or GPU bound, at least on my system. To check that, I underclocked my CPU down to 2 GHz and measured with and without mirrors

CPU clock: FPS no mirrors/FPS with mirrors:

2.0   27 /23
2.5   34 /28
3.0   41 /34
3.5   46 /40
4.0   51 /43

The correlation is almost linear! To double check, I underclocked my GPU as much as Afterburner let me (Ill need to search how to underclock it further)

GPU core speed:
1544 51 /44
1709 51 /44
1924 51 /43

GPU Mem:
3500 Mhz 51 /43
4000 Mhz 51 /43

Min clock for both mem and core:
51 /43

No correlation whatsoever.

We all know DCS is CPU hungry but I didnt quite expect this. Certainly not with "high settings"

Things may be different above land, I will need to test that later, but in this scenario and with these settings, I seem to be 100% "cpu" limited. I put that in quotation marks, because it might be broader than just the CPU, in particular memory speed might be a factor, and I do have some slow leftover DDR4. I have some DDR4 3200 on order, so soon I will be able to test if that make any difference.

@Strikeeagle345 I would really love to see exact measurements from you, as you happen to have the same GPU and an intel cpu !

I5-8600k @ 4,9 GHz

1080 @ 2050 MHz

32 GB ddr4-3000 ram

60 FPS Without mirrors.

55 with mirrors.

We all know DCS is CPU hungry but I didnt quite expect this. Certainly not with "high settings"

 

Things may be different above land, I will need to test that later, but in this scenario and with these settings, I seem to be 100% "cpu" limited. I put that in quotation marks, because it might be broader than just the CPU, in particular memory speed might be a factor, and I do have some slow leftover DDR4. I have some DDR4 3200 on order, so soon I will be able to test if that make any difference.

 

@Strikeeagle345 I would really love to see exact measurements from you, as you happen to have the same GPU and an intel cpu !

It is very CPU hungry. More so over land. I doubled my frame rates going from a 3rd gen i7 to a 9700k. Both using the same video card.

I5-8600k @ 4,9 GHz

1080 @ 2050 MHz

32 GB ddr4-3000 ram

60 FPS Without mirrors.

55 with mirrors.

Interesting. Since I think we can discount the fact you have a faster GPU, we basically get identical performance per GHz for our CPUs. I was expecting recent Intel i5s to have a slight edge over AMD. Now Im even more curious if, or how much RAM speed matters, as you do have faster ram than me.

If you're curious to see how fps is impacted by CPU throughput (drawcalls): place 50 F-18s on the tarmac, put the camera 500m away, note FPS- and do the same but replace F-14's with Hornets.

The F-14 should yield much superior FPS, despite more triangles and heavier textures.

Optimization is a complex topic. The F-14 will lose 1:1 with everything else due to geometric complexity and how heavy handed we were on the quality, but at the same time it will win in a scenario where many aircraft are on screen at once.

Hello again, Vertigo72.

I've already started a similar thread ages ago:D. Maybe you'll find some useful information there: https://forums.eagle.ru/showthread.php?t=236968&highlight=oddities

But one thing I can say: Every time you change the graphic-settings, you have to delete the fxo and metashader2 folder, otherwise changed settings will have (almost) no effect.

Without mirrors: ~54 FPS

With mirrors: ~47 FPS

Only less than 10 FPS difference, but in this range it makes it unplayable for me.

*Intel Core I5 4690 @ 4.4 Ghz

*Asus B85M-E Mainboard

*Asus Strix RX580 8GB (OC Mode)

*Crucial 24GB DDR3-1866 Modules, running at 1600 (Mainboard limitation)

*Samsung SSD 850 EVO 250GB

*32" Philips Cineos 32PFL9603 TV (1920*1080)

Hello again, Vertigo72.

 

I've already started a similar thread ages ago:D. Maybe you'll find some useful information there: https://forums.eagle.ru/showthread.php?t=236968&highlight=oddities

 

But one thing I can say: Every time you change the graphic-settings, you have to delete the fxo and metashader2 folder, otherwise changed settings will have (almost) no effect.

Good info. I will rerun the tests after deleting those files. I had already noticed some settings didnt seem to take hold, unless I restarted DCS. Maybe thats not enough. Edit: no difference in my case

I cant replicate you vsync problem or alt+enter issue. That does seem to be AMD specific. Its not clear to me if you already tried disabling vsync in the game and forcing it through the driver? I have encountered games/drivers where this was needed or otherwise I would end up with a "double applied" vsync and thus framerates locked to 30FPS.

As for your results; assuming things dont change after deleting those files, its interesting you beat me in this test. Clearly its not the RAM. Nor is it the GPU as such. So either it is the core i5 thats faster than a ryzen per clock (which sounds reasonable and I would have expected that, but contradicts some other results posted so far), or perhaps the AMD drivers incur a lower CPU overhead?

I guess if we get more results we'll learn more.

If you're curious to see how fps is impacted by CPU throughput (drawcalls): place 50 F-18s on the tarmac, put the camera 500m away, note FPS- and do the same but replace F-14's with Hornets.

 

The F-14 should yield much superior FPS, despite more triangles and heavier textures.

Optimization is a complex topic. The F-14 will lose 1:1 with everything else due to geometric complexity and how heavy handed we were on the quality, but at the same time it will win in a scenario where many aircraft are on screen at once.

Im not here to criticize the F14. I love it too much :)

But it does seem like rendering the cockpit requires significant resources, particularly CPU resources. In external views the framerates are much much better. They may indeed be better than a F18, but the reality is we spend most of our time inside the cockpit and thats where the framerates suffer. And gorgeous as it is, I am curious if you have any insights on why its so cpu dependent? High res textures or high polygon count would mostly stress the GPU, so is it AI, or physics or the modelling of all the internal systems, or is it the rendering pipeline somehow?

^^^ What Vertigo72 said...

While I notice a slight drop in FPS vs the F18 what I really notice, and is the worst part, is how, for some reason, the sim keeps 'going to disk' when in the Tomcat. Seemingly 10x more than the Hornet. Worst in PG map, but I still see it in the NTTR. I've tried it several times (same missions). It's becoming a bit frustrating. Happens seemingly no matter what settings I use, from low -> high.

My specs for those that will ask:

32gb RAM, i7 6700 4ghz

8 Gig dedicated Vram (nVidia 1080)

I didn't run the test exactly like Vertigo72, but hopefully my results can be of some use.

DCS Settings as seen in attachment below...

1440p

i5 7600K overclocked to 4.6GHz

1080Ti stocks speeds

32GB DDR4 3200 RAM

SSD

I set up the Defender of the Fleet mission as described by the OP

Zoomed out

Mirrors on: FPS 60, CPU 46%, GPU 73%

Mirrors off: FPS 70, CPU 46%, GPU 66%

Normal Zoom

Mirrors on: FPS 78, CPU 47%, GPU 73%

Mirrors off: FPS 96, CPU 47%, GPU 81%

RAM usage was 8.5GB

One thing I don't understand is it seems like I have CPU and GPU usage to spare, why isn't my FPS higher and my usage at 100% to try and get the max FPS that my hardware can deliver?

Im not here to criticize the F14. I love it too much :)

 

But it does seem like rendering the cockpit requires significant resources, particularly CPU resources. In external views the framerates are much much better. They may indeed be better than a F18, but the reality is we spend most of our time inside the cockpit and thats where the framerates suffer. And gorgeous as it is, I am curious if you have any insights on why its so cpu dependent? High res textures or high polygon count would mostly stress the GPU, so is it AI, or physics or the modelling of all the internal systems, or is it the rendering pipeline somehow?

I meant it more as a curiosity. :)

The likely primary reason that the F-14 is more CPU bound on the render thread is due to it having a significantly higher object count.

The F-14 cockpit in the pilot's position (with manual culling) incurs 130+ drawcalls, while the Hornet is closer to 45.

This is due to things like the pilot models and higher actual cockpit complexity (switches, knobs, total size even, et al.) and can't really be optimized away.

The likely primary reason that the F-14 is more CPU bound on the render thread is due to it having a significantly higher object count.

The F-14 cockpit in the pilot's position (with manual culling) incurs 130+ drawcalls, while the Hornet is closer to 45.

 

This is due to things like the pilot models and higher actual cockpit complexity (switches, knobs, total size even, et al.) and can't really be optimized away.

Mr Cobra,

First, I LOVE this module. Really. I'm a VR-guy (waited 30 years for this and I ain't going back).

Several times while admiring the beauty of the Tomcat cockpit (my wife thinks I look hilarious twisting and turning in my chair), I've thought it would be nice to have a lower-detail version? Please don't take that as any kind of slight to your (and your teams) collective work. Just an option for performance, sorta like tuning down texture res. or limiting draw distance...

Thanks

Mr Cobra,

 

First, I LOVE this module. Really. I'm a VR-guy (waited 30 years for this and I ain't going back).

 

Several times while admiring the beauty of the Tomcat cockpit (my wife thinks I look hilarious twisting and turning in my chair), I've thought it would be nice to have a lower-detail version? Please don't take that as any kind of slight to your (and your teams) collective work. Just an option for performance, sorta like tuning down texture res. or limiting draw distance...

 

Thanks

Thanks for the kind words! :)

Setting textures to medium will load the next mipmap in the texture mip pyramid, significantly lowering vram usage, memory bandwidth and cache misses.

Beyond that, however, there isn't really a mechanism in DCS for us to deliver a lower detail cockpit through the options.

In general, it would also take us a lot of time to create such a cockpit (we'd have to rebake all of our normals).

We're pretty happy with performance at present. I know it can be painful on mid-range and in VR, but it was a real challenge to deliver quality while at the same time having it be a big, complex (shape and texture wise) aircraft.

That unfortunately pushes hardware requirements up though.

One thing I don't understand is it seems like I have CPU and GPU usage to spare,

You dont. CPU usage is very misleading and borderline meaningless. It shows the average of all 4 or 6 cores, but a bottleneck will form if one single core cant execute one thread fast enough. On a 6 core / 12 thread machine this could look like 8% cpu usage, yet you are completely CPU bound.

More over, windows will shuffle threads across cores (for good reason, dont stop this). Assume DCS is only 1 thread that uses 100% of a core all the time, but if windows moves that 10x or 100x per second, on time scales of seconds it may look like all cores are underutilized all the time.

Finally there may be limits to the parallelism of the CPU and GPU execution. Imagine a frame is prepared by the CPU, than handed off to the GPU for rendering and while the GPU is doing its thing, the CPU has nothing to do. Lets say it takes equal time on CPU and GPU. You would never see more than 50% utilization on any core on the CPU, and yet you are both CPU and GPU bound. Of course reality is much more complicated than that, but you get the idea.

As for GPU utilization; Im not sure what is being measured. I suspect its shader utilization ("cuda cores"). DCS GPU pipeline may not be shader bound, but texel fillrate bound. So that measurement could be pointless too.

Instead of looking at utilization rates, its much more telling to do as I did; underclock CPU or GPU. If you see no difference, then the bottleneck is clearly elsewhere. IF performance drops linearly with the underclock, as is the case with my CPU in this test, then you are entirely bottle necked by that component. CPU usage is still low though, because I have 6 cores /12 threads, most of which arent doing anything. Adding more cores, the utilisation rate would drop further, but is going to do nothing to performance. increasing clock speed would result in pretty much linear performance boost. At least in that particular test.

The F-14 cockpit in the pilot's position (with manual culling) incurs 130+ drawcalls, while the Hornet is closer to 45.

This is potentially good news! Being CPU bottle necked, I was worried we where pretty much stuck performance wise for the foreseeable future as cpu's just arent getting significantly faster anymore, all we get is more cores. Single thread CPU performance scaling is all but dead. GPUs do still get better every generation, and because of their parallel nature, can be scaled almost arbitrarily. So being CPU bound, I didnt quite see how we would ever be able to use future next gen super high res VR sets with DCS, or even current ones at good FPS.

However, if drawcalls is a significant bottleneck, the Vulkan API may be a game changer. Its supposed to dramatically reduce the overhead for drawcalls by a factor of 10x or more.

OTOH, even 130 doesnt sound like a lot. Of course, this is a purely synthetic test that only measures the API calls:

https://www.anandtech.com/show/11223/quick-look-vulkan-3dmark-api-overhead

But its measuring millions of drawcalls per second. If that is what our CPUs are capable off (doing nothing else) and DCS only does 100s of drawcalls per frame, so on the order of 10000 per second, that still only accounts for less than 1% of the cpu time, and even if that gets sped up by a factor 10x, it wouldnt be measurable.

Oh well, time will tell I guess/

With an 8700K @ 4.8Ghz, 1080 ti @ 1658MHz/5954MHz, and 64GB DDR4 3000, I get 50-60 with mirrors, and a more solid 60 without.

Can anyone with DDR4-3200 try and see if changing dram speed makes any real difference?

If you can, set it to 2133 in the bios, and rerun the test.

My dram order has been backordered, and now Im curious if its worth spending to replace my current 16GB 2133 with 32gb 3200 or if I just should just add 16GB for now and run everything at 2133 (need the extra ram for other apps, may or may not need faster ram for DCS, I have no idea)

Can anyone with DDR4-3200 try and see if changing dram speed makes any real difference?

If you can, set it to 2133 in the bios, and rerun the test.

 

My dram order has been backordered, and now Im curious if its worth spending to replace my current 16GB 2133 with 32gb 3200 or if I just should just add 16GB for now and run everything at 2133 (need the extra ram for other apps, may or may not need faster ram for DCS, I have no idea)

I didn't run the test, but when i built my machine, i forgot to enable the XMP profile. my DDR4 was running at 2133. FPS was good in DCS. I went and enabled the XMP profile to 3200, and the DCS load times were reduced and my FPS went up about 5 fps. Others mileage may very, but this was very interesting non the less.

I don't have a super machine by any means yet the Tomcat runs AND looks absolutely AMAZING!!!...compared to the next best thing (the hornet), the Tomcat literaly blows it out of the water in both performance and visual quality ratio...

Is obvious that HB just made a step forward and rised the bar doing a "future proof" module so you can spect a little bit of an impact due to that but just in a few months/years as hardware improves the Tomcat will still look and run amazingly (not the case for others)...

that said, I do repeat that I haven't noticed any particular performance issue with the Tomcat except a slight one when flying as RIO, which makes perfect sense as there is a lot more things to load from the rear cockpit as the pilot cokpit is also being partially rendered, but nothing to do with HB.

HB just made the FIRST plane to date that, imho, REALLY feels and looks as the real thing, it is simply the most photorealistic cockpit I have ever seen in a simulator (in general, not only Flight simulation)...Absolutelly every single square inch in the cockpit AND outside is filled with lots of detail, quality and love...they made every single little thing come to life with extremely detailed textures, materials (Oh my, the roughmets are soooooooooo goooooooood :notworthy: ), meticulously accurate normal maps and again love, huge amounts of love:)...look at basically everything in the cockpit and you won't find a single thing not looking exactly as it should and not responding to light exactly as it should, is simply out of this world!!:worthy:

To all this you have to take into account the many new and unique features like Jesters AI, the carrier qualification chart, the several effects like turbulences, wing vibrations, etc. All this have an impact, a very welcome one imho :P

Nothing comes even close to the HB's Tomcat at the moment, so I think the little price to pay for this is well worth it.;)

Good info. I will rerun the tests after deleting those files. I had already noticed some settings didnt seem to take hold, unless I restarted DCS. Maybe thats not enough. Edit: no difference in my case

 

I cant replicate you vsync problem or alt+enter issue. That does seem to be AMD specific. Its not clear to me if you already tried disabling vsync in the game and forcing it through the driver? I have encountered games/drivers where this was needed or otherwise I would end up with a "double applied" vsync and thus framerates locked to 30FPS.

 

As for your results; assuming things dont change after deleting those files, its interesting you beat me in this test. Clearly its not the RAM. Nor is it the GPU as such. So either it is the core i5 thats faster than a ryzen per clock (which sounds reasonable and I would have expected that, but contradicts some other results posted so far), or perhaps the AMD drivers incur a lower CPU overhead?

 

I guess if we get more results we'll learn more.

I've disabled vsync in both game and Amd settings. But I think it's an AMD issue. Still wondering that no one else has come forward.

Strange that deleting the shader folder had no effect on you. Without deleting the shader I can go from "low" to "high" and the appearance and FPS will remain the same. To make that clear again: It's about leaving the game after changing the settings and then deleting the folders. Again, I am surprised that no one else seems to benefit from this. Or it is simply not being noticed...

[...]

Also, I'd reiterate deleting the fxo and metashaders folders on every graphics change. We're not sure why the precompiled shaders are behaving so erratically in DCS.

 

[...].

Regarding single core performance and higher FPS, I made a quick benchmark with CPU-Z:

F-18 7500 ms, 163 drawcalls;

F-14 1720 ms, 91 drawcalls.

https://forums.eagle.ru/showpost.php?p=3900154&postcount=68

In game:

61/71 fps

In pause:

73/84 fps