Gaming benchmarks: NVIDIA Physx performance

[Artas1984]

Greetings fellas!

This time around i've decided to check something, that has got little popularity and variety in the internet: NVIDIA Physx performance evaluation for gaming. Once in a while a random thread or video pops in the internet displaying how NVIDIA Physx algorythm runs a certain game with random video cards, but that is it.

The purpose of this benchmark is to find out if NVIDIA Physx middleware helps to improve performance in terms of FPS in games, which use any particle physics effects.

For this test i have selected only those games, in which i can clearly see the usage of particle physics being rendered on the screen. You should have an idea what those are by now: any weather effects, explosion debris, fire, cold, dust, smoke, interacting small 3D objects like vegetation, gibs, fragments of material...

I've created 4 different popular scenarios how NVIDIA Physx is being used by gamers of the world:

1. Physx is set to CPU, while main rendering is set to single high-end GPU.

2. Physx is set to a powerful mid-range gaming GPU, while main rendering is set to single high-end GPU.

3. Physx is set to a fast low-end gaming GPU, while main rendering is set to single high-end GPU.

4. Physx is set to to a single high-end GPU.

-----------------------------------------------------------------------------------------------------

Let's see how does this work i my testing rig, which has Core i7 5775C CPU, Gigabyte Z97X-Gaming 3 SYS, 2X8 GB DDR3 1600 MHz CL8 RAM, Samsung Evo 500 GB SSD, GTX980 Ti, GTX970 and GTX650 Ti Boost video cards.

1. Physx is set to Core i7 5775C 4 GHz {1 MB L2 + 6 MB L3 + 128 MB L4 caches}

2. Physx is set to Zotac GeForce GTX970 AMP Omega Core 4 GB.

3. Physx is set to Palit GeForce GTX650 Ti Boost 1 GB.

4. Physx is set to Asus GeForce GTX980 Ti Strix 6 GB.

-------------------------------------------------------------------------------------------------------------------

I am sorry for making this benchmark quite late. Asus GTX980 Ti Strix is no longer the top dog of GPU world, even though according to Techpowerup it is faster than R9 Fury X, Titan X Maxwell and R9 295X2.

I've tested 17 games at 1920X1080 resolution with all effects maxed out, but with no anti-aliasing as i think this is the most popular resolution and it does not bottleneck GTX980 Ti so much. Some games have their own benchmarks, for others i use custom Fraps MIN/MED/MAX benchmark tool, different time-bench for each game. But the most important thing is that every tested game has particle effects on screen - whether that game is officially supported by NVIDIA Physx i do not care.

We will begin this test with an "official" (not really) NVIDIA Physx benchmark tool called Fuildmark - if nothing else this tool will surely show if any additional hardware improves FPS when handing particles.

VIDEO PRESENTATION

FLUIDMARK 1.5.2

Holly crap daddy! According to this scenario no way in hell should you use your CPU for Physx. There is a colossal difference between GTX650 Ti Boost and GTX970 handling Physx as the single GTX980 Ti sits somewhere between them in terms of performance. Just makes me wonder what SLI GTX980 Ti could do..

While this was all very impressive and mouth dropping, lets do real world game testing, because it is nothing like the synthetic test...

ASHES OF THE SINGULARITY

Made this test 5 times in a row for each scenario to get the best result from each. There is so much particle buzz going on in this official game benchmark, that it is easy one of the most CPU intense games out there. Not much difference between the contenders, but clearly GTX650 Ti Boost lags behind the rest.

BATMAN ARKHAM ORIGINS

Made this test 3 times in a row for each scenario to get the best result from each. Again GTX650 Ti Boost clearly lags behind the rest, but CPU performance is absolutely abysmal, just like in Fluidmark.

BATTLEFIELD 1

Made this test 5 times in a row for each scenario to get the best result from each. This game uses a lot of particles on screen - explosions happening and debris falling, fire burning and ash pouring, yet there is no difference in performance. Fraps 15 second benchmark.

BORDERLANDS 2

Made this test 15 times in a row for each scenario to get the best result from each. Clearly Physx is supported in this game and you can see cold weather effects. GTX970 pulls the best average and maximum FPS, while GTX650 Ti Boost falls short in minimum FPS. Amazingly the CPU pulls the best minimum FPS over all. Even with that in mind, it is my opinion that it is not worth to allocate any additional hardware for Physx in this game. Fraps custom 30 seconds benchmark moving in one direction.

COMPANY OF HEROES 2

This is one of the most CPU intense games on the market and you can clearly see that this game is full of heavy weather and massive explosion effects which result in the CPU falling short in maximum and medium FPS, while GTX650 Ti Boost falls short in minimum FPS.

CRYOSTASIS THE SLEEP OF REASON

Single benchmark is enough, since all the rest yield the same results. An older yet incredibly demanding or perhaps poorly optimized game that uses a horse-load of cold weather particle effects on screen and was one of the first ever CUDA optimized Physx games. Unfortunately all that Physx support was just for the looks as there is no benefit in performance. Fraps 15 second benchmark.

CRYSIS 3

Made this test 5 times in a row for each scenario to get the best result from each. Heavy weather effects clearly on screen, but small difference in FPS is probably random. Fraps 15 second benchmark.

DYING LIGHT

Made this test 5 times in a row for each scenario to get the best result from each. For testing this game i have selected a custom zombie map which has a lot of particle effects in it - moving vegetation, ash flying in the sky, smelly vapor pouring from zombies, fountain with water flushing in the center of the map (not everything is visible in the screenshot). However it seems there is no almost no performance advantage in any of the contenders with the exception of GTX650 Ti Boost perhaps trailing behind... Fraps 25 second benchmark.

DRAGON AGE INQUISITION

Made this test 5 times in a row for each scenario to get the best result from each. Even though some cold weather effects like snowflakes and freeze haze are visible, all the contenders have identical performance. Fraps 15 second benchmark.

KILLING FLOOR 2

Made this test 5 times in a row for each scenario to get the best result from each. The game recommends GTX980 or higher for Physx effects, yet GTX650 Ti Boost managed to score more than the rest... WTH? Huge disappointment and a shity benchmark no less... Fraps 10 second benchmark (when waiting longer zombies appear).

LORDS OF THE FALLEN

Made this test 5 times in a row for each scenario to get the best result from each. Finally a good example how additional hardware improves performance, although how come GTX650 Ti Boost managed to beat single GTX980 Ti contra-indicating Fluidmark results is bizarre... The CPU field is empty because CPU does not support Physx in this game at all. Fraps 15 second benchmark.

METRO LAST LIGHT REDUX

Made this test 5 times in a row for each scenario to get the best result from each. There is so much wind, ash, rain and vegetation interaction in this game, yet the lazy NVIDIA bastards put no effort in optimizing performance as all the contenders score the same FPS. Fraps 15 second benchmark.

MIDDLE EARTH SHADOW OF MORDOR

Made 3 to 4 benchmarks in a row, since after the fourth benchmark the results would not improve. During the benchmark we see a lot stuff happening, but the results are inconclusive.

MIRROR'S EDGE CATALYST

Made this test 5 times in a row for each scenario to get the best result from each. Some rain is visible on the screen. It seems GTX650 Ti Boost falls short here, even though it is the same engine tested as in Battlefield 1 - Frostbite 3. Fraps 15 second benchmark.

RAINBOW SIX SIEGE

Made this test 10 times in a row for each scenario to get the best result from each. Explosions occur during testing which account for particle effects, yet the performance is almost identical, except for CPU falling short in maximum FPS.

RISE OF TOMB RAIDER

Made this test 10 times in a row for each scenario to get the best result from each. I only summed up the results from the first test - mountain benchmark seemed to be the most stable and reproducible. I am not sure if GTX650 Ti falling slightly short is a coincidence or not...

WITCHER 3 WILD HUNT

First of all i've made this test from different testing places - one near the fire, another in the outside like in the picture, so that all the clouds, vegetation and birds would amount for physics particles, and then i summed up the results from both test places. Made this test 5 times in a row for each double scenario to get the best result from each. However, as you can see no contender offers any performance value. Fraps 15 second benchmark.

-------------------------------------------------------------------------------------------------------

You can all make your own conclusion, but if had to make ones:

1. If you have a single high-end GPU, no additional GPU for Physx will be needed.

2. You should never allocate Physx for CPU, even though for most cases the CPU will not hinder performance.

3. Allocating Physx for low-end GPU hinders performance more than using Physx with single high-end GPU, but for most cases the low-end GPU will not hinder performance.

4. Allocating Physx for mid-range GPU improves performance more than using Physx with single high-end GPU, but for most cases the mid-range GPU will not improve performance.

 

[P4-630]

1. If you have a single high-end GPU, no additional GPU for Physx will be needed.

So I'm fine then, thanks!

 

[silentbogo]

The main problem is - there are no games that use physx that much.
Ever since Cellfactor died, the most we get in games today is cloth and hair simulation, which can be outsourced to GT730 (at most).
I want destructible objects, simulated particle effects, physically-based wind and weather effects etc, but apparently game devs only care about better textures, higher res and more blur/flares/grain/other shit.

 

[RejZoR]

You should test Killing Floor 2. Btw, most listed games don't really use PhysX beyond what Havok was, a basic CPU based physics engine.

 

[Artas1984]

I've updated the thread with video presentation.

You should test Killing Floor 2. Btw, most listed games don't really use PhysX beyond what Havok was, a basic CPU based physics engine.

Killing Floor 2 has been tested in several different maps with no performance benefit in any of the tested hardware parts. In the end i chose the map with the biggest amount of snowfall. And yes, most of these games, it seems, don't support Physx despite having many physics particles.

 

[rtwjunkie]

You are definitely dedicated! This was a massive amount of work that must have taken at least 20 30 hours to do, although you'll likely surprise me and say it was way more.

What I wasn't clear on was were all these game physics-using games or PHYS-X games? The reason is because there were quite a few where there appeared to be little difference between any combination of hardware.

I appreciate your conclusion that a higher-end GPU really has more than enough to handle PHYS-X on its own. It made me think though, if a mid-tier GPU would benefit more not only from a second GPU for phys-x or even offloading to the CPU? My guess is in that case CPU offloading would not hinder as much as it did with the 980Ti.

Again, great work!

 

[jboydgolfer]

certainly must not have been easy....

 

[RejZoR]

I've updated the thread with video presentation.



Killing Floor 2 has been tested in several different maps with no performance benefit in any of the tested hardware parts. In the end i chose the map with the biggest amount of snowfall. And yes, most of these games, it seems, don't support Physx despite having many physics particles.

Erm, but have you actually enabled Flex and Fluid features? Because this game doesn't use them by default, but does offer them in graphics settings. It's also focused on guts, so if you just run around the map, you'll not see any change since you're also not seeing effects used by these two features.

 

[Toothless]

Erm, but have you actually enabled Flex and Fluid features? Because this game doesn't use them by default, but does offer them in graphics settings. It's also focused on guts, so if you just run around the map, you'll not see any change since you're also not seeing effects used by these two features.

Does Flex and Fluids actually fully use a single physx card? I'm tempted to turn off SLI and try the theory.

 

[EarthDog]

Tha ks for the i fo... but do some of these games even support hardware accelerated physx???
https://en.m.wikipedia.org/wiki/List_of_games_with_hardware-accelerated_PhysX_support

I mean the conclusion is right from what we've seen tested in the past (you love testing things already tested!!).. but many of those games aren't hardware phyx accelerated...no?

 

[Artas1984]

Tha ks for the i fo... but do some of these games even support hardware accelerated physx???
https://en.m.wikipedia.org/wiki/List_of_games_with_hardware-accelerated_PhysX_support

I mean the conclusion is right from what we've seen tested in the past (you love testing things already tested!!).. but many of those games aren't hardware phyx accelerated...no?

Right, most of the games officially might not support NVIDIA Physx in my benchmark, but there are many games with Physx support that are not included in the official Physx support list! My purpose was to find out these "sleeper games" which perhaps do support NVIDIA Physx despite not being mentioned as supportive! That being said, if you look at Rise of Tomb Raider, Mirror's Edge Catalyst, Company of Heroes 2 and Ashes of Singularity benchmarks - you will notice that GTX650 Ti Boost provides weaker results versus the rest of the contenders, and i've tested these benches at the very least 5 times in a row. Example: in Ashes of Singularity GTX650 Ti Boost would never score over 46 FPS in minimum range, but GTX970 would always score 48-49 FPS! So perhaps these mentioned games are the sleeper games! Obviously Lords of the Fallen, Killing Floor 2, Cryostasis, Batman Arkham Origins and Borderlands 2 do support Physx.

Erm, but have you actually enabled Flex and Fluid features? Because this game doesn't use them by default, but does offer them in graphics settings. It's also focused on guts, so if you just run around the map, you'll not see any change since you're also not seeing effects used by these two features.

Yes, i did enable Flex and Fuild features, but as you said, i needed a scene with those things rendered, and since this game is a dynamic game with zombies always hunting you, the benchmarking options are very limited.

 

[EarthDog]

Ok.. the 'point' of the thread changed from the title and first post. Noted.

 

[Ferrum Master]

Alice: Madness Returns is the only physx heavy title where dedicated card gave a huge FPS boost.

Thats because of the smoke simulation.

 

[TheHunter]

Great testing and all, but 70% of the games DON'T use physx, at least not hardware so it doesnt make any difference.

no physx:
ASHES OF THE SINGULARITY
BATTLEFIELD 1
COMPANY OF HEROES 2
CRYSIS3
DYING LIGHT
DRAGON AGE INQUISITION
MIDDLE EARTH SHADOW OF MORDOR
MIRROR'S EDGE CATALYST
RAINBOW SIX SIEGE
RISE OF TOMB RAIDER

Not sure about Wicher3, other then it uses nv own hairworks. EDIT: it uses it for APEX for cloth and destruction and simulations, but in software.

For killingfloor2 you need to have both flex and fluids enabled
*flex for rigid bodies - guts, fluid - for blod.


*here is a list of all HW physx based
https://en.wikipedia.org/wiki/List_of_games_with_hardware-accelerated_PhysX_support
http://physxinfo.com/wiki/Upcoming_GPU_PhysX_games


by most with old physx 2.8.x code it can run like crap, lots of gpu stalling issues and crippling gpu with low gpu usage, newer physx3 is better but still not ideal.

Latest nv thing is Flex and its getting a lot better.
https://developer.nvidia.com/gameworks-visualfx-overview

https://developer.nvidia.com/gameworks-physx-overview

 

[rtwjunkie]

by most with old physx 2.8.x code it can run like crap, lots of gpu stalling issues and crippling gpu with low gpu usage, newer physx3 is better but still not ideal.

For those older ones, install the legacy physX available on Nvidia's own download site. Your system will use it if a game calls for it as it doesn't replace the current physX.

http://www.nvidia.com/object/physx-9.12.1031-legacy-driver.html

TW3 seems to have the full gamut of Nvidia proprietary stuff, including PhysX.

 

[TheHunter]

For those older ones, install the legacy physX available on Nvidia's own download site. Your system will use it if a game calls for it as it doesn't replace the current physX.

Yes I had to install it for hmm, let me think - Stoked BigAir or it wouldn't start at all.

But overall it doesn't speedup, its the code that's inefficient, perfect example was Alice2, Dark Void, Borderlands, UT3 physx maps - was also driver dependent, some had more fps drops then others. But never ideal.

I remember Alice2 and why they capped it to 30fps, if uncapped at 60 when high smoke turbulence happened gpu usage dropped and fps dropped to ~ 35fps, which made sporadic fps all over the place. This was by any hw 2.8.x game, I tested each game with newer gpus and improved context switching, but never was ideal enough. Fermi 580 >> Kepler 780 >> Maxwell 980ti..

 

[rtwjunkie]

Yes I had to install it for hmm, let me think - Stoked BigAir or it wouldn't start at all.

Velvet Assassin is another one that requires the legacy physX driver.

 

[FordGT90Concept]

NVIDIA supposedly open sourced PhysX but this page gives me a very strong impression that AMD still can't touch it with a ten foot pole (you can't even view it unless NVIDIA says you can--very closed). Until NVIDIA decides not to be an ass about it, it can only serve as an afterthought. Using PhysX the way it is meant to be used would mean breaking games on Intel and AMD graphics. Broken games don't sell and, if they do, they get refunded.

 

[TheHunter]

NVIDIA supposedly open sourced PhysX but this page gives me a very strong impression that AMD still can't touch it with a ten foot pole (you can't even view it unless NVIDIA says you can--very closed). Until NVIDIA decides not to be an ass about it, it can only serve as an afterthought. Using PhysX the way it is meant to be used would mean breaking games on Intel and AMD graphics. Broken games don't sell and, if they do, they get refunded.

Apparently they were or are working on direct compute variant, this was back in 2015, but so far nothing changed, yet.

http://forums.guru3d.com/showpost.php?p=5030362&postcount=9

 

[Toothless]

Assassin's Creed Black Flag uses it

 

[Artas1984]

Assassin's Creed Black Flag uses it

It uses it, but do hardware components benefit from it in terms of FPS performance? Even if so, i would require a benchmark rendering screen with physics effects present "in your face". If i have to play this game for over an hour just to see some smoke or rain, than it ain't worth the effort. What can you tell me about it? I have Assassins Creed Syndicate, but i did not test it, since i saw no physics particles.

 

[Folterknecht]

Great job.

Though that only reinforces my opinion, that a dedicated PhysX card just causes problems (space, noise, heat, power draw) while contributing next to nothing to the gaming experience. Better sell your old GPU and use the money for something that makes more sense, like a few cases of beer.

 

[NavyDiver]

Killing Floor 2 with Physx Fluids @ 2560 x 1440 (GTX 1080, GTX 1080 with GTX 1080 dedicated physx, GTX 1080 SLI).

Theoretically, as long as the dedicated Physx card is equivalent to the primary video card, there is nothing to loose but electricity. But here is the FRAPS data on 16 minutes of KF2, a game that actually stresses modern video cards with Physx. The map is Tragic Kingdom, solo HOE short, Firebug for rounds 1-4, Gunslinger for boss round.

Bottom line up front: SLI is currently broken on KF2 and has been for some time. Despite FPS counters reporting higher framerates, the mouselook in SLI mode feels around 30-40 fps and is unplayable. Regarding the dedicated Physx card, based on the objective data there is a significant difference averaging 30 fps. In my case I probably won't notice anything above 70 fps, but other players may. The best video card option for KF2 full Physx Fluids is 2 identical cards, but 1 for dedicated Physx instead of SLI. Such as a 1080ti plus 1080ti. Regardless of what is available in 2017 on the consumer market, I'm confident that fps drops into the 20s is unavoidable when the Firebug unleashes into a crowd with the microwave gun, etc.


GTX 1080:

GTX 1080, with additional GTX 1080 as dedicated Physx card:

GTX 1080 SLI (Nvidia HB SLI bridge):

 

[FordGT90Concept]

How about GTX 1080 with GPU PhysX disabled? Frankly, PhysX isn't worth the compute cost.

 

[kgame]

Tha ks for the i fo... but do some of these games even support hardware accelerated physx???
https://en.m.wikipedia.org/wiki/List_of_games_with_hardware-accelerated_PhysX_support

I mean the conclusion is right from what we've seen tested in the past (you love testing things already tested!!).. but many of those games aren't hardware phyx accelerated...no?

Unity3D & Unreal Engine uses PhysX as physics engine by default.
https://developer.nvidia.com/physx-sdk