CPU and GPU analysis with DCS World

[@mnd479]

these benches you make are crucial to 99.9% of simmers, finally someone is doing real world cpu gpu bench in dcs in vr ! , most of us that need that extra performance and still not sure on what cpu pulling the trigger on, this will help a lot!

[@flxdrv5020]

Well you know which one to pull the trigger on, too bad there isn't any in stock at a decent price lol

[@abetts69]

So glad to see someone doing VR benchmarks in simulation. Would love to have seen RX7900XTX in this comparison like in your iracing vid.

[@jcugnoni]

Cool, this is so much valuable! We need a proper benchmark basis for DCS in VR. Instant subscribe to your channel.

[@sailyourface]

DCS benchmarks? Yes please! Perfect timing for my build upgrade thanks!

[@carymcrawford]

Most excellent foray into DCS bench marking. Looking forward to your subsequent reports.

[@mapleschnitzel2473]

Great job, I hope you get your hands on 5090 I need to know the VR performamce. This is the best place to get that info.

[@andreyzhavoronkov6746]

Hey, thank you very much for this! Sim benchmarks are rare to come by and while MSFS does get some attention nowadays, same cannot be said about DCS. So a big hats off from me personally and dare I say, DCS community, for doing this!

[@mnd479]

benching dcs is best when you fly over a city say Damascus with some say 50 ground and air assets moving (the more assetes \groups move on a longer path the more cpu heavy it is)

[@staubsauger2305]

Correct. Damascus low (but not ultra low where horizon line-of-sight clipping would reduce the polygon count) where you would really differentiate between systems. Also, 4k display also pushes systems harder, or even dual-4k in high-end VR headsets. One of the biggest factors worth benchmarking for DCS is the effect of increased VRAM (video card memory), as moving textures along the bus between RAM and VRAM affects performance more than most people understand. The extra VRAM accounts for a lot of the performance difference between Nvidia's GPUS (eg. 4090 to 4080) and as someone who programmed a bit of OpenGL you were always aware of this and tried to keep as much on VRAM as possible (compressing textures, using indexed vertex buffers etc). Thanks for doing DCS Benchmark Odysseys the reward for getting good (simulated) flying is worth the effort to get there.

[@baststar]

What a gread Benchmark - and for VR. Thank you very much for this!!!!!! This is so rare on KZbin!

[@taubkin]

Excellent work. Can't wait for more. Multiplayer is definitely a great follow up. There are also some "stress test" benchmarking tracks with lots of AI units and AI action running around the internet that can further expand on CPU comparissons.

[@terrylaven]

So good to see. Thank you for taking the time to do this. I am also interested in the effect of memory speed. i.e. difference between DDR4 vs DDR5 memory, 32GB vs 64GB on multiplayer servers. Its my understanding that the 32 vs 64GBs become important in multiplayer, but would be nice to see the actual numbers. Liked and subscribed!

[@jcugnoni]

Ps: by the way, it would be more realistic to have some head movements also in VR as it has a strong influence on the CPU work that needs to load / animate a much larger number of objects. Maybe it could be possible to playback the camera movement too by disabling the head tracking input from the headset...

[@Onehighcat]

Thank you for this. This is very helpful for the community!

[@Chloiber]

What would be extremely interesting is VRAM comparison, e.g. 12v16v24 or so.

[@HamVR]

Welcome to the rabbit hole of benchmarking DCS 😂A couple of things to consider. The Intel 12/13/14th gen CPU have 8 P-cores but they also have additional 8 to 16 E-cores, and this game is optimised to use those too. It's also important to record the build number of the DSC version you are benchmarking, as I've seen large swings in performance from build to build, when ED make performance tweaks, which is often...

[@BenchmarkOdysseys]

Thanks for having me ;) I did try disabling Ecores on 13700K to see if it reduced the FPS dips I noticed, but it did not. Thread usage is high so yes I can see Ecores being an asset. The build tested was 2.9.11 and it looks like ED released an update during the time it took me to produce the video. That's always a risk with these 'live' service simulators and a burden we have to benchmark them, compared to the single player titles that many benchmarkers use and don't have to worry about that.

[@hustlinmagic]

@@BenchmarkOdysseys DCS certainly had issues with 13th and 14th gen Intel. Check how its spreading the workload over the cores. There were issues with it maxing out one or two cores and barely touching the others , which led to stutter and frame rate dips. The solution at the time was using Lasso to disable the cores it was maxing out, and it would then spread the workload more efficiently over the remaining P cores. I didn't really look into the how's and whys, but it certainly worked to improve performance.

[@mnd479]

keep up the good job m8 we really apricate it a lot ! (and need it)

[@mnd479]

can you do a bench without dlss\upscaling (rasterized) and with say x2 msaa ?

[@3gun_squadfather]

Excellent start to benching DCS with your awesome graphs for vr! Hopefully some people come through with track files. Also please do some tests without dlss turned on, higher textures and shadows. The settings you’re using are sort of compromises to get vr to run solid, which sort of goes against the idea of benching right? You’ve shown that those settings run dcs more than acceptably in vr and pancake, so let’s see a little less compromised settings.

[@BenchmarkOdysseys]

Thanks :) Yes the settings are more 'optimized' but I was still surprised by the GPU demand it created in VR. Even with such a low VR resolution, it was still too much for a 12GB 3080Ti (which is similar to 4070 family of cards).

[@gamerbvv7414]

Great.👍 finally THE game😊

[@gamerbvv7414]

could you one day compare the cpu ms with a big resolution (crystal) 4300x5100 per eye? 🙏

[@gamerbvv7414]

for DCS in multi, I'm not sure that a .track shows the "problem" I don't think it's only because of the number of units, I'll check

[@BenchmarkOdysseys]

That's an incredibly high resolution! And probably outside the reach of my 4080 Super 16GB. But I suppose the Crystal has DFR with eyetracking and Quadviews?

[@gamerbvv7414]

@@BenchmarkOdysseys yes, but with regular games, this resolution is fine in 120fps with 4090 and i9 9900k 😬

[@gamerbvv7414]

@BenchmarkOdysseys c is strange but I've seen benchmarks that might suggest that with huge resolutions like the one of the crystal lle 14900k could be faster than the 9800x3d, especially in frame time cpu and 0.1% low. What do you think about that, i will change my old 9900k for the best cpu possible very soon 🙏

[@TalkieT]

2 things. Once you identified that the 1080p behaviour was odd for the top CPUs and even suggested that it may be because you were breaking the default 300FPS threshold - why didn't you retest at 1440P to see if you could see normal behaviour between CPUs return? Second. Is there a 5090 in your immediate future? There's so much hype around it but I would love to see how it will actually benefit iRacing and short of buying it myself, I think you're probably the only reviewer I'd trust to test it cleanly.

[@BenchmarkOdysseys]

Great questions! I did complete a few runs with 1440p and 4K using 9800X3D and 7600X (same graphic settings). This definitely introduced a GPU constraint and closed the distance between the two CPUs, and at 4K, they delivered the same average FPS. Dips in frametime were still present, but affecting the 7600X more. I figured I'd explore this in more detail on the next set of DCS benchmarks, where I use a track file that is more CPU intense than the ED mission in this video. As for the 5090, I think those are going to be impossible to find for awhile. Heck I think AMD is calling Nvidia's bluff with the 5080, which will probably have underwhelming performance and also limited supply. Rumors are Radeon 9070 won't even launch until late March.

[@Rico-yw1jf]

This is quality content! Very interesting. Maybe put the VR headset on rotating fan to have consistent head movement, just an idea. Also, do you have some AMD GPUs to throw into a mix ? Like every game, some prefer AMD and some NVIDIA, I'd like to know which card manufacturer does better in raster VR DCS.

[@jaeloo]

Yeeeeeesssss!

[@mnd479]

hes doing the right bench at the right time before 9950-9000x3d's coming out, right on time hammer on nail this guy !. they should pay you m8. tens of thousands of dcs players are now on the hunt for new cpu to relief our 4k-8k+ vr experience , we all really bothered by this every day !

[@mnd479]

i have a 7900xtx and 5800x3d 64gb ram and i get max 70-75 fps in the situation stated below, now am sure if i had a better cpu say 9950x3d (16 cores) it will run better, how much better ? well we dont know until you make a bench on it ;) (also 9800x3d-7800x3d bench in dcs will help a lot to decide on what cpu upgrade should i go for , and not just me).

[@mortlet5180]

I have a TL;DR explanation following the question, but could you please do a run where you disable frame generation/interpolation both for the NVIDIA GPU and the Valve Index, then run it in its max 144Hz mode (at a lower pixel density to remove the GPU bottleneck) and report on the frametime distribution that a fully unlocked (power limit, power saving, C-states, etc. all disabled in the BIOS) and OC'd 9800X3D can achieve, especially when looking around rapidly with the headset? I'd really love to be able to play DCS, but after years of trying to run it, it seems like the engine itself is incapable of smooth, low frametime frame pacing, especially when flying low and looking around rapidly, and I've just given up trying and wasting time and money on someting that seems like it's never going to improve, especially with everyone that keeps saying that 90-100fps is perfectly smooth and anyone that wants more fps is just looking to brag and can be ignored. *TL;DR* For whatever reason, frame rates 300W, even at idle), cooled by a custom loop running on chilled (~10 °C) water with direct-die cooling the delidded CPU and a custom block on the GPU, all in a cold (~16°C) room.

[@Munky332]

Would like to see these tests, except with something a bit newer/harder to run -- ie the most popular headset right now, the Quest 3. The quest 3 has some more unique quirks being that it has to route through the GPU encoder (NVENC for Nvidia)

[@BenchmarkOdysseys]

Higher end headsets definitely demand more of the graphics card than CPU. So a user has to dial in their resolution and DCS settings for their GPU first with VR. So in theory, if I dialed my settings for Quest 3 with 4080 Super, the performance headroom between CPUs (as shown in the graphs) 'should' be similar as what I showed with Index. Unfortunately the streaming nature of the Quest 3 disguises proper performance analysis, so I cannot produce the same data sets.

[@alpenfoxvideo7255]

Unfortunately Multiplayer tracks don't usually work, crashing the client on playback. But I could cook you a mission in Syria with several units doing stuff and some scripts running? Unfortunately i can't track it for you as I don't own the F-18. Do you also have the F-16 or F-14 or F-4? Hello, I can for sure say that on difficult situations the CPU is a huge bottleneck. I have a 4090 and 13700K running at 80hz (quest3). Sometimes the frames are not in time and i see my CPU had 1 or 2 cores pegged to 100%. And when that happens sometimes the frames missed are several, like 5 or 10 frames missed. Otherwise my GPU is never at maximum usage, but at around 90-92% due to the frame cap. I am gonna buy a 5090 very probably and I will be frustrated to see my pc not able to push more than 80hz anyway, so your work is very useful to me

[@BenchmarkOdysseys]

Thank you for the feedback! I do not know how to script a mission, so I do appreciate that. Currently I only have F18 but I can pick up another aircraft next sale. Pushing a 5090 to it's limits will demand the fastest CPU available. The 13700K is very capable but I think you will see the same FPS dips if you replace 4090 with 5090.

[@alpenfoxvideo7255]

@ i’ll see if I can edit the mission this week and i’ll send that to you. The goal would be to have a scenario where you just need to fly trough it and has timed actions happening

[@BenchmarkOdysseys]

No rush, I won't be cycling back to DCS until later in February.

[@mnd479]

oo and another thing i almost forgot , most of us run QUADVIEWS, fixed foveated rendering, now this my friend take a lot of load from the gpu but kills the cpu, Literally kills it!!! so , our only refuge from dcs killing our gpu's is to use quadviews with high end cpu , and thats it folks , we just need the right cpu and were done with the never ending struggle!

[@audunskilbrei8279]

8:48 My setup is a 12700k 64 GB Ram 6900XT and I also observe low wattage on the GPU while playing DCS even though the game itself claims to be GPU limited.

[@philhealey2520]

I'm currently looking to upgrade from 2060super to a 9800x3d and either 4080super, 4090, 5080 or 5090 all subject to reviews and costs for a full new PC. Will be watching as it's mostly for DCS VR that I'm upgrading

[@mnd479]

can you also add to dcs bench a rasterize test with no dlss \upscaling, with say at least x2 msaa ?

[@RSharpe]

These benchmarks are already running at very low resolutions, we shouldn't be benchmarking with DLSS which lowers the internal resolution further.

[@mnd479]

further more , dcs bench when plane on the ground near grass and trees that move by the wind and jet blast tank the cpu sometimes heavily ,

[@RSharpe]

DCS is often data bandwidth limited. SSD->system RAM->VRAM. Do you have specs and settings for your system RAM? Maybe interesting to test bandwidth/latency

[@BenchmarkOdysseys]

Yes, I noticed with VR that the system memory usage shot up to over 32GB and started using a paging file. So I did this benching with 64GB DDR5 6000 CL30 and 64GB DDR4 3600 CL18. I am using Gen4 NVMe drives from WD and Samsung.

[@Funky85Mess]

Hey man, I'm going to record some tracks with the Kiowa, Apache, F-18 and I'll send em to you if you'd like.

[@BenchmarkOdysseys]

I'd appreciate that! Thank you :)

[@bigbangerz5856]

So many reviewers have just completely dodged PCVR performance, complete bias towards pancake 😢

[@mnd479]

few things when benching on dcs , dcs has 4 modes \versions, single thread and MT=MULTI THREAD, openxr and steam vr , most of us use MT with openxr to get the best performance, 4k+ i usually the res we have , say hp g2 runs on openxr at min 3100+3100 res per eye as openxr default,

[@madigorfkgoogle9349]

I believe all DCS is now MT by default.

[@mnd479]

@@madigorfkgoogle9349 i sure hope so

[@T595955i]

I appreciate your testing, many thanks. However, the video should be named instead as "CPU and GPU analysis with DCS F/A-18C module on Syria map". What I mean is, it's incomplete, testing with different modules and maps, and their different combinations, will give different results - now that would be an "analysis with DCS World" like you named it. For example, you focused on the F/A-18C Hornet and that one is, perhaps, the best optimized FF module of DCS. It is just one among dozens of FF modules for DCS. Now try that again but with the F-4E Phantom II, or F-14A/B Tomcat, or AH-64D Apache, or CH-47F Chinook, or Mi-24P Hind, or KA-50 BlackShark (the newest version), and etc. You'll probably see significantly different results, which get even more dependant on the combination of certain modules with certain maps/terrains. Syria is now (finally) a well optimized terrain/map, as is the default Caucasus (by far the most performant), it's a world appart in terms of resources requirements if you get into something like the Afghanistan, or Iraq, or Kola, or South-Atlantic maps.

[@BenchmarkOdysseys]

Thank you for taking the time to explain this. I appreciate your feedback. I am familiar with the difficulties of benchmarking racing simulators, and I think there is crossover with flight sim. Variables like weather conditions, locations visited, time of day, number of opponents and the user's proximity to them, type of vehicles being used, speed travelled, size and length of the event, etc. Each factor influences performance. Simulators like iRacing also charge per DLC, and there are 136 cars and 126 tracks, of which only 12% are included free. Optimization of DLC is also a tricky thing to navigate. Too new or under-developed, and I'm testing poor code rather than hardware. I picked Syria because it is widely enjoyed in Multiplayer, and I picked F18 because several guides suggested it was the most versatile full fidelity aircraft. Turns out 'most versatile' doesn't equate to popular. My best effort is to pick something that reasonably equates to DCS World experience. It may not match your favourite craft or flight scenarios. Just like how I use GT3 road cars at international race circuits to evaluate iRacing, but thousands of racers only drive NASCAR ovals in southern USA. Then there's the different resolutions and aspect ratios of monitor(s) people use. Or all the different VR headsets and software packages. Also, I only have 10 CPUs and 6 GPUs at my disposal, while hundreds of different models sold over the last few years. The combinations of sim variables + personal preference + individual PC configuration are infinite.

[@T595955i]

@@BenchmarkOdysseys Yes, for sure, I understood it, and by no means am I underrating your efforts here, which are worth of praise. What I mean is, how the results you got here will be perceived by both newcomers and veterans of DCS. What you get from here is, very likely, "the best case scenario", and that's why (IMO) it should be interpreted with caution, and why I call it "incomplete". Instead, and if the choice was there, I'd rather see these tests done with other combinations of popular module/map of DCS (and even more so in a populated MP server), because the increased demand on resources that such does bring is a better representation of "what to really expect across the board".

[@jamesmcd71]

I'm running a 7950x with a 4090. 1440p and 4k are no problem, I can run 165 fps with a 162 1% absolutely butter. Yet both the quest 3 and Cristal light will bottlenecke my CPU. We're talking about it drops GPU usage down to 35- 40%. And that's with an extremely choppy image. In a single player it will do OK at around 45 to 50 fps with a slight stutter. But in a multi player server I have to drop the graphics down to potato just to maintain a jagged 45 fps. It's unplayable. The 7950x is just not fast enough because it's forced to use ram as cash.

[@BenchmarkOdysseys]

Interesting! Thanks for sharing. DCS becomes a performance hog with VR, it's a totally different beast than single screen performance.

[@jamesmcd71]

@BenchmarkOdysseys I decided I was going to switch to the 9800X3D. Then I found out there are none available without paying a insane markup to a CPU scalper. I ordered a 7800X3D because it seems like they are almost gone as well.

[@unotoli]

Very good depth. But width in terms of meaningful GFX settings tweaks toward max would be very welcomed. Your "hardware" graphs, combined with "software" (settings) study like this - kzbin.info/www/bejne/hICzhqOBip2cb9U would be a magical/dream combo!

[@baststar]

[@cks5148]

Agree , buying Intel new Ultra CPUs is worse than AMD. 9700X is still better than Ultra 7 265K . Not to mention the high cost of 8400M/T CUDIMM , without that you are much worse than a 12th gen intel CPU . IF price is right , 9800X3D should be the king for DCS .

[@-ugly-Derby]

Thank you so much for doing some proper testing. I can't offer a Syria track but I might be able to provide you with a track on Marianas next week. There won't be much happening in terms of AI but it's going to be a bunch of F/A-18C's, F-16Cs and other aircraft operating around Andersen AFB (we are doing ATC training with the squadron on Thursday). It's not a very intensive mission though. The track would be F/A-18C. There are a bunch of statics at Andersen, an American CSG, a Russian carrier group, some Tankers and an AWACS but not much else. Unfortunately most of my multiplayer flying features quite long transit times, often exceeding 30min, and it's usually not in modern aircraft as most of my time is spend flying Warbirds on the Project Overlord WWII Server or flying the F-4E Phantom with the the Squadron. I am waiting for the 5090 release because I fly DCS in VR and plan on upgrading the VR Headset this year. Unfortunately my 3080ti does show its limits these days although my recent upgrade to a 9800X3D with 96GB DDR5-6000 made frametimes much more consistent which is also shown in your benchmarks (I upgraded from a 5900X with 64GB DDR4-3600).