fps reports/rants

natostanco

Main menu:
dx11 lowest settings / 195-205fps
dx9 lowest settings / 130-140fps

dx11 highest settings / 100-115fps
dx9 highest settings / 95-110fps

Settings impact in main menu:
Reflections / 5fps
Ambient occlusion / 30fps
Anis. Filter / 5fps
Atmospherics / 25fps
AA / 10fps
Shadows / 20fps

Sandbox mode - docking - marines side:

dx11 lowest settings / 130-140fps
dx9 lowest settings / 120-130fps

dx11 highest settings / 110-120fps
dx9 highest settings / 90-100fps

Settings impact in SB:
nope.

I suppose(guess? Imagine? Blabber about?) in the main menu the difference between high/low settings and dx9/dx11 is so huge because your are technically gpu limited since you are not in a server and there is no cpu/networking shizzles, and graphic settings and dx level are all about gpu. In sandbox mode I got those results, which shows a much tighter difference because there is most likely a hint of cpu limitation in action. In actual servers The difference between high/low settings is basically inexistent, I have seen shit drop to 40fps in combat to peak around 130fps and alfs bases with all the infestation range around 60-90fps. I was tempted to say dx11 helps more on high settings then it does on low settings but what the fuck should I know :> my numbers say otherwise, but w/e, plus since everyone in this game is basically cpu limited dx11 didnt help anyone really. (:s)
OpenGL doesn't even work.
The only time I have seen ns2 using all the cores was during load, rendering time, the stuff you do once, that stuff filled all the damn cores but did the rendering in like 10 seconds.

Am I the only one who finds it ironic that they left the most cpu dependent engine of all time (source) to basically build one as much cpu gluttonous as the one they discarded?

I might just have written a bunch of bullshit but w/e, I must justify my shattered dream of steady 120fps somehow.

System:
amd 9370-fx @ 5.1ghz
r9 290

ps: please take numbers as surrounded by a field of tildes ~~~~~

Soul_Rider

The one thing you have to remember is that the model entities in NS2 use a lot of server and client processing power. The game runs all their abilities, does hit calculations, both client and server are always updated with the animation state, models need to be rendered, and physics calculated.

The issue is that the longer a game goes on, the more models are added, and the harder the servers and clients have to work. The Spark engine is very efficient in and of itself, it is the game code that is causing the problems, there is just too much going on.

If you remember NS pushed the absolute boundaries of the HL1 engine, well unsurprisingly , NS2 pushes the absolute boundaries of the Spark engine. The engine itself is solid, NS2 is just a very big game.

IronHorse

NS2 actually reaches over 1,000 entities late game.. compare that to something like counterstrike (maybe 20 or 30 entities, tops)

Also.. do not discount your original findings with AO and atmospherics... They destroy FPS, and the former (AO) comes with inherent mouse input delay. (all AO does)
Oh and lastly, Fullscreen DX9 is the only officially supported rendering method, all else is still a work in progress. DX11 comes with better frametimes so it "feels" smoother.. but it does not match the mouse input delay of DX9. (which is equal to Quake 3's three frame delay)

ns2isgood

How do you get more FPS with DX11 than DX9? I don't get it.

lwf

Are you sure Ambient Occlusion really add input lag? All I could find with a quick look on google was basically rumours on various forums. I've never had the pleasure to play with it since it tanks performance, but it's so pretty I was looking forward to it should I ever buy beefier hardware.

IronHorse

Max (who wrote the engine) explained to me once that it was inherent in the implementation of AO after much inquiry about the obvious input delay i was experiencing.
I haven't touched a single implementation of AO that hasn't had the same issue. (even farcry 3's custom AO)

I ASSUME it's because It's an additional frame because its post processing.
If you're not sensitive to that sort of thing, then yea all that matters is beefier hardware to swallow that 20 fps cost. (though AO medium isnt that taxing)

Wiki:
"Instead of rendering 3D objects directly to the display, the scene is first rendered to a buffer in the memory of the video card. Pixel shaders are then used to apply post-processing filters to the image buffer before displaying it to the screen."
Basically from my understanding, it's using pixel depth to determine application, treating that frame like a texture and buffering it prior to modification before sending it to you. I could be way off tho.


Edit: p.s. newer algorithms are coming out and are 7x better performing and even look better. http://floored.com/blog/2013/ssao-screen-space-ambient-occlusion.html
(also, i haven't confirmed yet, but i believe the one who wrote the research paper on this at nvidia is Max's brother. :P)

Dictator93

IronHorse wrote: »

NS2 actually reaches over 1,000 entities late game.. compare that to something like counterstrike (maybe 20 or 30 entities, tops)

Also.. do not discount your original findings with AO and atmospherics... They destroy FPS, and the former (AO) comes with inherent mouse input delay. (all AO does)
Oh and lastly, Fullscreen DX9 is the only officially supported rendering method, all else is still a work in progress. DX11 comes with better frametimes so it "feels" smoother.. but it does not match the mouse input delay of DX9. (which is equal to Quake 3's three frame delay)

Why do you think all AO comes with inherent mouse input delay?

It actually depends on the type of AO from what I understand. Whether or not it uses previous frame information to generate a more accfurate depth buffer to prevent shimmering. And even then that would be rare for a game to purposefully delay a frame just to increase the temporal stability of something for the next.

Unless NS2 is purposefully waiting on the next frame drawing to draw its AO.. it would only otherwise increase input latency due to the amount of MS it takes to render a frame. But just by the amount which the frame takes longer to render. Which woulld be 30% more MS in the case of NS2 on average. (the difference in feeling of 120 fps and 90)

Someone with more technical knowledge can correct me, but I cannot think of many post processing techniques which would purposefully delay a frame... screen spacepost processing techniques exist to decrease rendering time usually. Non, post processed version of the effects are otherwise too expensive.

IronHorse

@Dictator93‌
Just going off of what Max told me..
You can definitely feel the input delay though if you test for it and are sensitive to that sort of thing.

After doing some research, I've found some interesting tid bits about how traditional AO (made by crytek) is a pixel shader and "a pixel shader is the only kind of shader that can act as a postprocessor or filter for a video stream"
But like you said there are now multiple solutions.. and the only place where I can find info about a delay is the post processing done to the high frequency noise to make it look shaded:

Without additional smart solutions, such a brute force method would require about 200 texture reads per pixel for good visual quality. This is not acceptable for real-time rendering on current graphics hardware. In order to get high quality results with far fewer reads, sampling is performed using a randomly rotated kernel. The kernel orientation is repeated every N screen pixels in order to have only high-frequency noise in the final picture. In the end this high frequency noise is greatly removed by a NxN post-process blurring step

I can only guess (probably incorrectly) that this is why medium AO is less of an issue but also more "noisy".

Omega_K2

IronHorse wrote: »

NS2 actually reaches over 1,000 entities late game.. compare that to something like counterstrike (maybe 20 or 30 entities, tops)

That's completely false, CS has several hundred entities too, and can go beyond that 1000 with a modded server or custom maps; for example, a quick test on semi-vanilla (no mods that change entites) with 24 people on css dust2 shows an entcount of 314. Yes, NS2 gets around 1000 with 24 and late game reguarily, but still that's not an excuse for poor performance.

In regards to performance, it will still be significantly better then NS2 with compareable entcount - hell the hardware that can run 24 ppl on NS can run 64 + mods in CSS, even when the entcount is similar and on roughly the same tickrate, and on the client you may even get ridiclious fps rates.

IronHorse

What? Really? Can't even begin to guess what is considered an entity in cs to make 300 of them??.. it's still 70‰ less than ns2, so I think my point still stands however..but of course lua has a hand in it too, I'm sure.

Scatter

What is the input delay differential between dx9 and 11, because I too get about 20 fps more. Although, alt tabbing is hell.

Soul_Rider

Omega_K2 wrote: »

IronHorse wrote: »

NS2 actually reaches over 1,000 entities late game.. compare that to something like counterstrike (maybe 20 or 30 entities, tops)

That's completely false, CS has several hundred entities too, and can go beyond that 1000 with a modded server or custom maps; for example, a quick test on semi-vanilla (no mods that change entites) with 24 people on css dust2 shows an entcount of 314. Yes, NS2 gets around 1000 with 24 and late game reguarily, but still that's not an excuse for poor performance.

Ent numbers are useless on their own. What do the ent's include? Dropped weapons? Fragmented models?

The difference you have to remember is that ent's in NS2 have a HUGE amount of logic. Not only do they have the physics code (As i have found out on GorgeCraft, physics code is a huge resource usage), but they also have their abilities, so every ent on the map is thinking every tick:

Crags seeing if you are within range and if you need healing
Shifts seeing if you are in range and need cloacking
Whips seeing if you are within range for a slap or bomb.
Sentries seeing if you are in range for a few bullets in the face.

I could go on, but you get the point, there are very few entities in NS2 that don't have masses of work to do every tick. You see NS2 doesn't just have entities, it has Dynamic entities, and many of them (I'm looking at you here infestation) have a lot of responsibility and take up a lot of CPU work.

Please detail the CS entities, so we can compare and see what each of those entities has to do each round. Rather than just saying this game has X vs Y number of entities.

MrFangs

IronHorse wrote: »

Also.. do not discount your original findings with AO and atmospherics... They destroy FPS, and the former (AO) comes with inherent mouse input delay. (all AO does)

What I don't get is why AO is on by default. (It still is, right?) As you wrote, it causes a big FPS decrease, and if it also adds input lag, that's even worse. How is trading in FPS and responsiveness for some eye candy a good idea?

If the standard response to "help I got crappy FPS" is turning off a default setting, something seems wrong.

MrFangs

Scatter wrote: »

What is the input delay differential between dx9 and 11, because I too get about 20 fps more. Although, alt tabbing is hell.

Try switching to windowed mode with alt-enter before alt-tabbing, then go fullscreen again after tabbing back.

It works well for tabbing out. My only issue is that I'm often left wondering if I'm actually back to fullscreen, as there are times when you can't toggle it (map loading etc), and there is no visual indication.

SUPER_SARS

running Mendasp's custom hud yourself and getting it installed on the servers you play (or all of them) will help increase your fps.