Do Dedicated PhysX Cards Improve Performance?

Godlygamer23 · November 25, 2015

Hello everyone. I recently came across a thread where someone had asked if a PhysX card was worth it when paired with a 980 Ti. The general response was no, it's not worth it, and I see that as a common response to queries like that. I got curious. What would happen if I used my system as a test rig to see what kind of performance improvements I could get by using a dedicated PhysX card?

What is PhysX?

Well, it all goes back to a physics simulation engine known as NovodeX, which was developed by NovodeX AG. Another company called Ageia bought them out and developed hardware technology that could accelerate physics calculations - the idea was to take some of the load from the CPU. Ageia had called this hardware piece PhysX "Physics Processing Unit" or PhysX PPU for short. In 2008, NVIDIA bought Ageia out and added PhysX acceleration to its line of video cards. As time went on, NVIDIA phased out the PPU in favor of using its own video cards. People who were interested in PhysX but liked AMD cards would buy a cheap NVIDIA card for PhysX and a high end AMD card for the actual game rendering. NVIDIA caught onto it and disallowed it by detecting AMD cards in the system and disabling Advanced PhysX altogether.

Test System

CPU: Core i5 3570K @ 4.2GHz

CPU cooler: Kraken x60

Motherboard: ASUS P8 Z77 LK

RAM: G.Skill 12GB @ 1333MHz

GPU(s): Gigabyte Windforce 970 @ 1.46GHz and ASUS DirectCUII 660 Ti @ 736MHz(underclocked)

Storage: Intel 520 240GB SSD; 3 WD Caviar Blue HDDs

Power Supply: Seasonic X-Series 750W Fully Modular

Display(s): LG IPS234 LED LCD; ASUS VG284QE

Operating System: Windows 10

The games I tested at this point are Metro Last Light and Batman Arkham City respectively. I was originally going to test Batman Arkham Asylum, but the benchmark function has been removed apparently and testing it in-game won't really be useful as the frame rate is locked at 62FPS, even if I miss with the INI file. I plan to add to this post by testing with Assasin's Creed Black Flag if I can find an appropriate testing area.

Please bear in mind that these games are not rendering on the 660 Ti. They're still being rendered by the 970. The tests are for PhysX performance only.

Batman Arkham City

I ran the integrated benchmark and the settings are as follows:

Resolution: 1920x1080

V-Sync: Off

Anti-aliasing: FXAA(high)

DX11: Disabled

Detail Level: Extreme

Dynamic Shadows: Yes

Motion Blur: Yes

Distortion: Yes

Lens Flares: Yes

Light Shafts: Yes

Reflections: Yes

Ambient Occlusion: Yes

PhysX: High

Tests:

660 Ti(Test one):

Minimum: 40

Maximum: 128

Average: 82

660 Ti(Test two):

Minimum: 39

Maximum: 126

Average: 82

970(Test one):

Minimum: 35

Maximum: 119

Average: 73

970(Test two):

Minimum: 36

Maximum: 119

Average: 73

Metro Last Light

I ran the benchmark tool and the settings are as follows:

DirectX: 11

Resolution: 1920x1080

Quality: Very High

Texture Filtering: AF 4X

Motion Blur: Low

Tessellation level: Very High

PhysX: Advanced

Scene: D6

Tests:

660 Ti(Test one):

Minimum: 10.27

Maximum: 67.37

Average: 37.61

660 Ti(Test two):

Minimum: 17.15

Maximum: 67.09

Average: 37.69

660 Ti(Test three):

Minimum: 17.47

Maximum: 68.41

Average: 37.67

970(Test one):

Minimum: 9.18

Maximum: 58.35

Average: 31.78

970(Test two):

Minimum: 15.04

Maximum: 60.59

Average: 31.60

970(Test three):

Minimum: 15.40

Maximum: 68.41

Average: 31.55

One might notice that my Settings sections for each game are not consistent. That's because it's in the order it displays in on the respective panel. If you would like me to change it to maintain consistency, message me or post it in this thread and I'll take care of it. If you would like to see other games in this post, please let me know of your request and I'll try to get it in.