PhysX causing low GPU usage (GPU PhysX)

[MushBrain]

Hi everyone I own a GTX 580, I have been having strange issues with Batman Arkham Origins, if I disable PhysX altogether the 580 runs the game on max settings flawlessly sans MSAA (use FXAA), if I enable high PhysX, in the benchmark with no effects even in use the FPS will drops to 40-30 FPS and my GPU usage sags rather low to like 50-75%.

Is this normal? I have never had a PhysX game do this even Borderlands 2 does not do this nor either of my previous Batman games.

I have also overclocked the 580 from 772mhz to 945mhz stable, it's around the GTX 960 in performance.

 

[MushBrain]

I'm really annoyed with how poorly made the high PhysX setting is in this game, I ran the bench with normal PhysX and scored not far off the same as without PhysX with all other settings maxed.
Though I only saw 62FPS as my minimum, not sure how it recorded it as 46 unless it's from transitioning scenes.

 

[photonboy]

1) I don't know how you monitor GPU "usage" but what you want to observe is if the GPU frequency is dropping which indicates either:
a) thermal throttling, or
b) CPU bottleneck

**WTF????

I just tested and got the EXACT same FPS using HIGH vs MEDIUM PHysX (that was NOT the case when I bought it). I have VSYNC off. I can tell you my GPU isn't throttling based on my card fan speed. It's running higher than normal.

I'm going to RETEST in case I have to restart the program, so I'll update if I find out anything.

Other:
With my GTX680 + i7-3770K, I got 128FPS average at 1920x1080 mostly maxed (motion blur is off, possibly a couple other settings dropped).

The benchmarks is almost exactly TWICE what I get in-game because i carefully tweaked to achieve 60FPS most of the time in-game. I really, really hate benchmarks that don't indicate in-game performance. I think it's SHAMEFUL because the only reason i can think is to mislead people into buying the game who have weaker systems.

Running Batman on a rooftop is not the same as fighting a whole bunch of enemies in a different location.

 

[MushBrain]

1) I don't know how you monitor GPU "usage" but what you want to observe is if the GPU frequency is dropping which indicates either:
a) thermal throttling, or
b) CPU bottleneck

**WTF????

I just tested and got the EXACT same FPS using HIGH vs MEDIUM PHysX (that was NOT the case when I bought it). I have VSYNC off. I can tell you my GPU isn't throttling based on my card fan speed. It's running higher than normal.

I'm going to RETEST in case I have to restart the program, so I'll update if I find out anything.

Other:
With my GTX680 + i7-3770K, I got 128FPS average at 1920x1080 mostly maxed (motion blur is off, possibly a couple other settings dropped).

The benchmarks is almost exactly TWICE what I get in-game because i carefully tweaked to achieve 60FPS most of the time in-game. I really, really hate benchmarks that don't indicate in-game performance. I think it's SHAMEFUL because the only reason i can think is to mislead people into buying the game who have weaker systems.

Running Batman on a rooftop is not the same as fighting a whole bunch of enemies in a different location.

What? that's the end screen of the benchmark it does several scenes, in game i'm always above 60FPS.

 

[photonboy]

1) I don't know how you monitor GPU "usage" but what you want to observe is if the GPU frequency is dropping which indicates either:
a) thermal throttling, or
b) CPU bottleneck

**WTF????

I just tested and got the EXACT same FPS using HIGH vs MEDIUM PHysX (that was NOT the case when I bought it). I have VSYNC off. I can tell you my GPU isn't throttling based on my card fan speed. It's running higher than normal.

I'm going to RETEST in case I have to restart the program, so I'll update if I find out anything.

Other:
With my GTX680 + i7-3770K, I got 128FPS average at 1920x1080 mostly maxed (motion blur is off, possibly a couple other settings dropped).

The benchmarks is almost exactly TWICE what I get in-game because i carefully tweaked to achieve 60FPS most of the time in-game. I really, really hate benchmarks that don't indicate in-game performance. I think it's SHAMEFUL because the only reason i can think is to mislead people into buying the game who have weaker systems.

Running Batman on a rooftop is not the same as fighting a whole bunch of enemies in a different location.

What? that's the end screen of the benchmark it does several scenes, in game i'm always above 60FPS.

I know how the benchmark works, and it's a running average. It does not reset for each scene. Just look at the min/max/avg and it carries over from the roof to inside the prison.

I drop from 90FPS at the start and slowly drop to 66FPS (1080p, I think max everything except no motion blur as I think there was a problem with that).

*I had to RESTART the game or else the benchmark used the same MEDIUM PhysX settings. So now i get this for my GTX680 + i7-3770K->

1080p, almost max, PhysX HIGH:
66FPS avg

1080p, almost max, PhysX MEDIUM:
128FPS

In-Game (PhysX MEDIUM):
*I don't know what to call the average as it varies too much. It's rough 40FPS to 125FPS. In some areas I'll start fighting at 90FPS and never drop below 60FPS, and in other areas I drop down to 40FPS (probably a bit below what's reported in reality).

Anywhere that PHYSX isn't used at all I'm sure to get the same performance regardless of PhysX setting, but according to the benchmark I'll see a huge drop when HIGH PhysX feature are used.

For me the game stutters a lot if VSYNC is OFF, or if I have it on but drop below the refresh (below 60FPS for 60Hz monitor).

So...
Long story short the benchmark would suggest I'd get 66FPS using PhysX HIGH, and in reality there would be times I'd drop to around 20FPS likely.

So like most games with benchmarks I run the benchmark, tweak thinking it may work, then play the game, get stutters, low FPS and have to tweak things.

Unfortunately, many people aren't comfortable with that and just suffer through stuttering. (theoretically a console is the way to go as the company tweaks for you, but many games drop under 20FPS at times with 30FPS max so that's not an option either).

Just FYI, but Adaptive VSYNC works great for several of my games. Not sure if it's helping Batman AO though. I think something in how the program is setup is preventing that working properly.

For Assassin's Creed Brotherhood for example, I tweak to maintain 60FPS about 90% of the time, maybe higher. I then force on Adaptive VSYNC and it toggles VSYNC ON or OFF (the main point is to avoid stuttering if you can't output fast enough to synch with the monitor because then your frame time doubles causing mixed frame times.).

Mixed frame times are also why games don't appear smooth sometimes. Hence the frame time analysis to replace frame RATE (FPS). You might have 60FPS reported and have an experience closer to 30FPS. Go run a game like Half Life 1 and it can appear incredibly smooth at 60FPS VSYNC ON.

 

[MushBrain]

I managed to even it all out, it's also very smooth now with no harsh frame drops, it stays at 30-45FPS, I just turned everything up including MSAAx8 and maxed PhysX runs better than it does with lowered settings, UE3 is not a great engine.

 

[MushBrain]

Here is a PhysX heavy scene on the high PhysX setting, no AA and everything else maxed, it is stuck at 30FPS whilst GPU usage is at 65%.

So putting all the other GPU processing power to MSAA is the only way I could even it all out, the times where PhysX was not so heavy the GPU usage goes up and so does the FPS causing a very rapid change in frame rates making the game very uneven.

This is 8xMSAA on top of it all.

So if you have some GPU power left over after PhysX work I suggest you just crank it all up, I still never drop below 30FPS, it's a weird game.

 

[photonboy]

It's not a "weird game", you just don't understand how the GPU works well enough.

You can NOT raise any setting and have the frame rate increase. Something's wrong there.

Also, as I said I had to REBOOT the game for my settings to stick so if you change the settings you may not be benchmarking what you think you are.

Suit yourself, but I found the game pretty stuttery if I didn't tweak it to a solid 60FPS VSYNC'd experience. VSYNC OFF was not smooth even when well above 60FPS, and dropping below 60FPS (60Hz monitor) was also not smooth.

If it was ME I'd drop to Medium PhysX, 4xMSAA etc and attempt a more fluid experience.