7800X3D, 7900X3D, 7950X3D..

[tepl]

5800x3d + 64gb Cl16 + RTX3080 12gb 1920*1080 High Preset

WITH MIRRORS:

 

NO MIRRORS:

[Th3ChosenOn3]

So I felt something was a bit off with the results I had previously gotten, and did a little experimenting.

I figured out that even though I had Tacview disabled in the modules menu and Reshade turned off, it was still having a massive impact on my CPU for the benchmark. I went ahead and completely uninstalled both Tacview and Reshade, then did another 5 runs at 1080p to see the new numbers. I also went ahead and did another 5 runs with mirrors disabled to see what impact it had, and I got these results:

 

Quite shocked to see how large of an impact Tacview/Reshade had on the benchmark after it was completely uninstalled. I was also not expecting mirrors to have an affect on the CPU as much as this. Looking at the FPS Graph, in the mirrorless run I'm now hitting the FPS cap of 180 at the end while averaging higher numbers across the board.

 

So yeah. It might be beneficial for you guys to run the benchmark after disabling mirrors and uninstalling Tacview/Reshade to see what kind of an impact it has on your system. Since I've gotten this extra bit of data, I'll take a deeper look to see if the 7950X3D lowers the CPU hit of having these enabled, but I'm also starting to worry if the FPS cap of 180 might be limiting the benchmark now.

Edit:

So it seems disabling Tacview in the module manager doesn't stop it from recording. Bit of an oops on my part:

Edited March 7, 2023 by Th3ChosenOn3
Added Edit

[Cheese16]

I'd be really interested to see TacView still working without reshade. Reshadw has always kinda sucked but I would surprised if TacView alone had anywhere near as much if an impact on its own.  As a general, variables should always be removed one at a time for best results. Removing TacView and reshade at the same time rather than testing their individual impact only tells part of the story.

[Th3ChosenOn3]

15 minutes ago, Cheese16 said:

I'd be really interested to see TacView still working without reshade. Reshadw has always kinda sucked but I would surprised if TacView alone had anywhere near as much if an impact on its own.  As a general, variables should always be removed one at a time for best results. Removing TacView and reshade at the same time rather than testing their individual impact only tells part of the story.

Will be doing that tomorrow when I get some time, although I suspect it to be tacview mostly.

[some1]

1 hour ago, Th3ChosenOn3 said:

So yeah. It might be beneficial for you guys to run the benchmark after disabling mirrors and uninstalling Tacview/Reshade to see what kind of an impact it has on your system.

I already posted a test with no mirrors and clean DCS previously, but here's another run in 1080p instead of 4k. (5080X3D + 4090). 

In 4k high preset, even 4090 is the bottleneck half of the time. But not by a large margin, as on average I have gained only a few FPS more in 1080p.

At 1080p, the last 30 seconds of the track simply runs at 180 FPS which is the limit of DCS itself. 

[trevoC]

Would also be interested on how much of an impact tacview has as I couldn't care less about reshade but run tacview all the time.

In my 7x3d/4090 build I'm going to start the install without tacview run the tests and then install tacview to see the difference in VR. Wouldn't have even considered this so thanks for the tests.

[okletsgo]

I ran a test or two with task manager running....it also ate quite a few frames....

I wonder if it be worthwhile trying to set the affinity of tacview, to force it onto another core, away from DCS.

My AM5 water block has finally shipped, I will be able to build my 7950x3D rig this week

 

Edited March 7, 2023 by okletsgo

[trevoC]

39 minutes ago, okletsgo said:

I ran a test or two with task manager running....it also ate quite a few frames....

I wonder if it be worthwhile trying to set the affinity of tacview, to force it onto another core, away from DCS.

My AM5 water block has finally shipped, I will be able to build my 7950x3D rig this week

 

 

Nice,

I was expecting my rad to arrive tomorrow but found it at the office this morning. Will be building mine tonight hopefully in time to let it run tonight and download DCS/MSFS etc...

[tepl]

Same test , 5800x3d, 3080, FullHD, High preset, No Mirrors, WITHOUT 180fps Limit :

[Mars Exulte]

5 hours ago, Th3ChosenOn3 said:

So I felt something was a bit off with the results I had previously gotten, and did a little experimenting.

I figured out that even though I had Tacview disabled in the modules menu and Reshade turned off, it was still having a massive impact on my CPU for the benchmark. I went ahead and completely uninstalled both Tacview and Reshade, then did another 5 runs at 1080p to see the new numbers. I also went ahead and did another 5 runs with mirrors disabled to see what impact it had, and I got these results:

 

Quite shocked to see how large of an impact Tacview/Reshade had on the benchmark after it was completely uninstalled. I was also not expecting mirrors to have an affect on the CPU as much as this. Looking at the FPS Graph, in the mirrorless run I'm now hitting the FPS cap of 180 at the end while averaging higher numbers across the board.

 

So yeah. It might be beneficial for you guys to run the benchmark after disabling mirrors and uninstalling Tacview/Reshade to see what kind of an impact it has on your system. Since I've gotten this extra bit of data, I'll take a deeper look to see if the 7950X3D lowers the CPU hit of having these enabled, but I'm also starting to worry if the FPS cap of 180 might be limiting the benchmark now.

Edit:

So it seems disabling Tacview in the module manager doesn't stop it from recording. Bit of an oops on my part:

 

Mirrors are a big pull yeah, and ReShade can be, but I'm surprised TacView would have any impact when not running.

[Number481]

3 hours ago, tepl said:

Same test , 5800x3d, 3080, FullHD, High preset, No Mirrors, WITHOUT 180fps Limit :

You make a good point about the FPS cap in graphics.lua...  anyone with a 5800x or better should be setting that higher, especially if benching at 1080p.  Or this track file needs to be made far more complex...

 

Figured I'd give this a go on my new 7950x3d (w/ 4090) - High preset, various resolutions, all full-screen - Tacview installed but disabled

Here's the full breakdown at 1080p

null

Here's a comparison at the three indicated resolutions.  I'm a bit skeptical of the 4K test.  My monitor is only 1440p, but the game should still be trying to render at 2160p.. the results are within the margin of error of being identical to the 1440p results however....

null

And, if you're interested in how Tacview being enabled affects things... (I tested with it uninstalled and saw no difference from disabled)

[trevoC]

3 hours ago, mjfrisby said:

You make a good point about the FPS cap in graphics.lua...  anyone with a 5800x or better should be setting that higher, especially if benching at 1080p.  Or this track file needs to be made far more complex...

 

Figured I'd give this a go on my new 7950x3d (w/ 4090) - High preset, various resolutions, all full-screen - Tacview installed but disabled

Here's the full breakdown at 1080p

null

Here's a comparison at the three indicated resolutions.  I'm a bit skeptical of the 4K test.  My monitor is only 1440p, but the game should still be trying to render at 2160p.. the results are within the margin of error of being identical to the 1440p results however....

null

And, if you're interested in how Tacview being enabled affects things... (I tested with it uninstalled and saw no difference from disabled)

Am I missing something.... I see just shy of 15 fps difference with tacview on vs off in above chart. What did I miss.

[some1]

3 hours ago, mjfrisby said:

Figured I'd give this a go on my new 7950x3d (w/ 4090) - High preset, various resolutions, all full-screen

Thanks, that's a first test of the new X3D  Is that with mirrors on or off? Mirrors are not controlled by graphics preset.

3 hours ago, mjfrisby said:

Here's a comparison at the three indicated resolutions.  I'm a bit skeptical of the 4K test.  My monitor is only 1440p, but the game should still be trying to render at 2160p.. the results are within the margin of error of being identical to the 1440p results however....

Most likely you're just hitting GPU limit. Even at 1080p your screenshot shows GPU limit 25% of the time, so it's only higher at other resolutions. A combination of no FPS limit in DCS and a very powerful CPU.

[Th3ChosenOn3]

10 hours ago, trevoC said:

Would also be interested on how much of an impact tacview has as I couldn't care less about reshade but run tacview all the time.

In my 7x3d/4090 build I'm going to start the install without tacview run the tests and then install tacview to see the difference in VR. Wouldn't have even considered this so thanks for the tests.

11 hours ago, Cheese16 said:

I'd be really interested to see TacView still working without reshade. Reshadw has always kinda sucked but I would surprised if TacView alone had anywhere near as much if an impact on its own.  As a general, variables should always be removed one at a time for best results. Removing TacView and reshade at the same time rather than testing their individual impact only tells part of the story.

So here's the results with Tacview Installed and uninstalled, I didn't have much time to do 5 run aggregates so ignore the 0.1% lows being all over the place. It seems Tacview was causing the FPS hit all along. Reshade when it's turned off does not affect FPS.

 

On side note, I just received my 7950X3D and all the parts to upgrade my system today, so I'll be doing the upgrade for the next few days and come back with my 7950X3D numbers.

[Number481]

2 hours ago, trevoC said:

Am I missing something.... I see just shy of 15 fps difference with tacview on vs off in above chart. What did I miss.

It probably varies based on the CPU, but you are right...  15 or even 20FPS when you're talking in the range of 180+FPS is not much of a penalty.

However, I exclusively fly in VR.  Even with the X3D and 4090, there are still situations that bring the system to its knees.

There is one campaign (Operation Cerberus North) that is an absolute killer for me.  Starting on the ground in Syria with a crap load of ground assets on the map is a painful experience.  

This is a 30 second capture in VR, just sitting cold and dark in the hangar at the beginning of one of the missions.  You can see a test I ran with the 7950x before I upgraded for comparison (I never tested with Tacview disabled before the swap).

I can assure you the ~15fps in this case makes an enormous difference.

[EightyDuce]

Dang! That is a pretty hefty uplift.

[Number481]

2 hours ago, EightyDuce said:

Dang! That is a pretty hefty uplift.

Its even more noticeable in the headset

Now, I don't want to give the impression that Tacview is this problematic across the board.  It's really only an issue in cases with a lot going on.

In a normal free flight scenario, or even on the Supercarrier PG cold start built in mission, I don't see a difference with Tacview enabled.

 

Same goes for the 7905x -> X3D comparison.  The more assets on the map, the bigger the performance improvement I got.  But, in general free-flight, the X3D is pretty much on par with the X.

 

[Vyrtuoz]

Interesting benchmarks. I would like to add a few facts, which I hope can help you:

Yes, it is true that Tacview flight recorder can take a non-negligible amount of CPU in maps with a lot of dynamic objects. The source of the problem comes from the LoGetWorldObjects() Lua function provided by DCS World. It is very slow to execute, and the only way we have, to retrieve information about the mission objects. It is called two times by Tacview flight recorder, once for most objects (that’s the big hit on FPS), and once for unguided weapons such as bullets, roquets and bombs (this second occurrence if negligeable most of the time).

This function accounts for about 98% of the CPU time taken by Tacview flight recorder. The rest is highly optimized and multithreaded C++ code.

My hope is that, with the upcoming multithreading re-architecture of DCS, this problem will disappear: ED team will hopefully put slow Lua operations in a thread working in parallel of the main game, therefore even if LoGetWorldObjects() will still take a lot of time to execute, it will not slowdown the rest of the game. Another improvement would be to create a new, more efficient API to export data, but that’s another story…

Now, how can you save CPU time then? If you want Tacview to be automatically disabled in some circumstances, you can tell it to record only single player flights and not multiplayer flights for example. You can still replay the track after your VR session to export it as an ACMI file (I know this is not perfect, but this is worth mentioning).

If you want to disable Tacview exporter without deinstalling it, go to the “Special/Tacview” setting tab, and uncheck the “Tacview Module Enabled” box. The module will still be running, but it will not call any of the slow DCS export functions, therefore, its impact should be almost zero. Please note that the addon manager “enable” function has no effect on Tacview exporter, I do not know how it is working and how (if this is possible), to support it. That explains why you thought that you had to uninstall Tacview do disable the exporter. This is not necessary, just use the special option for that, *not* the Module Manager.

Oh, a last point about processor affinity ( SetProcessAffinityMask() ) this is not a solution. A long time ago, when I was working on the Xbox360, is sister function SetThreadAffinityMask was very useful and an official way of balancing threads. Today, this it not something you want to do. With modern CPUs, the latest versions of Windows will do a far better job (by balancing dynamically the threads on each core) than the programmer would do (think about the new intel CPUs were each P & E core have a different speed!). That, plus the fact that Tacview flight recorder is part of DCS process: SetProcessAffinityMask can only target the whole process without making the distinction between DCS core and Tacview flight recorder. Tacview thread is already running in parallel: The slow down is part and internal to DCS main thread. This very last problem should be (hopefully) solved by the upcoming multithreading re-architecture of DCS world.

I hope this command can help you to better understand the issue. I you find anything abnormal, I will happy to help.

Edited March 8, 2023 by Vyrtuoz

[trevoC]

3 hours ago, mjfrisby said:

It probably varies based on the CPU, but you are right...  15 or even 20FPS when you're talking in the range of 180+FPS is not much of a penalty.

However, I exclusively fly in VR.  Even with the X3D and 4090, there are still situations that bring the system to its knees.

There is one campaign (Operation Cerberus North) that is an absolute killer for me.  Starting on the ground in Syria with a crap load of ground assets on the map is a painful experience.  

This is a 30 second capture in VR, just sitting cold and dark in the hangar at the beginning of one of the missions.  You can see a test I ran with the 7950x before I upgraded for comparison (I never tested with Tacview disabled before the swap).

I can assure you the ~15fps in this case makes an enormous difference.

Thanks for the detailed response.

I had thought you said that tacview didn't have an impact which is what I was replying to (saying that I do see an impact at 15 fps). I agree with you completely, 15 fps is huge in VR (which I also fly in).

Sorry if I misunderstood that but thanks for the detailed response... seems worse than I thought which is too bad because I use tacView on almost every flight (especially campaigns which I didn't succeed) to learn and de-brief.

Thats a big blow as I've come to really enjoy an amazing piece of software.

----

Also appreciate the detailed response from tacView. I've come to expect nothing less from this developer. His software is great and the response I've received from him through support a few years ago was just as good.

Seems like we're in a no win situation here. Will definitely run a comparison test for my build when its running with tacview not installed/enabled/disabled.

[okletsgo]

@mjfrisby Thanks for posting, wow - it looks looks like a significant improvement with the 7950x3D. Looks like we bought the right CPU. Now we just need a 4090ti (or a 5090)

 

[Th3ChosenOn3]

Alright, here are the final results between my 5800X and 7950X3D all comprised into one chart.

 

Overall it's about a 60%-75% improvement depending on what's enabled. For the 7950X3D, I've updated the frame cap from 180 to 400, as the benchmark was sitting on the frame cap of 180 for the majority of the run. If the frame cap is left on 180, the benchmarks would average 170 FPS. As for multiplayer, I'm using rough numbers but I've seen more than double the FPS compared to my 5800X. On Enigma's Cold War I used to get between 30-50 FPS average, and now I'm averaging over a 100 FPS and being GPU bound at my native resolution and graphic settings.

[okletsgo]

@Th3ChosenOn3 nice, thanks for posting. Have fun with it. I built my 7950x3D today. I haven't had time to install DCS yet though, looking forward to trying it this weekend.

 

 

 

[Anubis_94]

20 minutes ago, Th3ChosenOn3 said:

Alright, here are the final results between my 5800X and 7950X3D all comprised into one chart.

 

Overall it's about a 60%-75% improvement depending on what's enabled. For the 7950X3D, I've updated the frame cap from 180 to 400, as the benchmark was sitting on the frame cap of 180 for the majority of the run. If the frame cap is left on 180, the benchmarks would average 170 FPS. As for multiplayer, I'm using rough numbers but I've seen more than double the FPS compared to my 5800X. On Enigma's Cold War I used to get between 30-50 FPS average, and now I'm averaging over a 100 FPS and being GPU bound at my native resolution and graphic settings.

Nice, thanks for your feedback.

Do you have any hint on what ccd was used ? vcache or regular one ?

[Th3ChosenOn3]

5 minutes ago, Anubis_94 said:

Nice, thanks for your feedback.

Do you have any hint on what ccd was used ? vcache or regular one ?

All of it's on the 3D V-Cache CCD, with game bar managing the scheduling.

[Anubis_94]

8 hours ago, Th3ChosenOn3 said:

All of it's on the 3D V-Cache CCD, with game bar managing the scheduling.

Nice, thanks