RETHINKING VIDEO CARDS

The nVidia RTX cards have some logic on the die for ray tracing and artificial intelligence. The additional logic is of limited value for gaming which depends on the primary parallel ALU.

Recall Crytek with their ray traced demo on a Radeon Vega 56 card. This shows that a modern graphics card has enough TFLOPS to handle more sophisticated shaders. The ray tracing shader is simply more demanding.

The PhysX cards had roughly the performance of a GeForce 9800 GTX which means it can generally be handled by a simple compute unit on a modern video card rather easily. Recently nVidia released PhysX as open source so it can now be forked for integration into games more easily while running on any conventional graphics card.

ARTIFICIAL INTELLIGENCE

AI has special needs that require a lot more performance than the few registers found on nVidia RTX cards. IBM has some neural logic chips but they are rather exotic and its not clear if they can be integrated into a gaming product. The AI used in automated cards uses more conventional CPU logic with some advanced programming to handle the sensors and to made decisions.

In March 2018, a self-driving Uber Volvo XC90 operating in autonomous mode struck and killed a woman named Elaine Herzberg in Tempe, Arizona. Obviously for self driving cars the need for better proximity sensors to detect pedestrians is needed. Nearly 40,000 people die on US roads every year, and human error causes more than 90 percent of crashes but it may a a long time before changes are possible. 

Self driving cars are still a long way from being autonomous enough to handle drunk passengers. Still, GM has been retooling the Chevrolet Volt with no steering wheel or pedals in anticipation that self driving cars will be approved within a few more years.

Expectations with Waymo’s self driving cars has been high that self driving cars with detailed maps can stay out of trouble spots. Waymo still uses safety drivers at present.

Driverless cars could represent as much as $7 trillion to the world’s economy. Such cars would revolutionize the supply chain.

Legal liability issues will require legislation to set the ground rules etc. So far very little has been done mostly to limit liability. Humans tend to bend rules which is an anathema to automated systems. This is forcing developers to pressure legislatures to set rules for the industry.

AI IN GAMES

AMD and nVidia have powerful hardware that can handle a lot of tasks but AI has specialized needs which represent a serious challenge to developers.

AI in games is a comparatively mature technology now and many games now have “bots” which can fill the role in a coop action game while the game server awaits new players to join. Generally specialized hardware is not needed to handle games which depend mostly on the developers design more than anything.

Havoc has been offering packages for a long time but with PhysX being open source it may finally allow games to have more realistic rain, water and even better destructible game worlds. Driving games can benefit with better car crashes etc.

So the AI on nVidia RTX cards does not have much to offer for gaming as opposed to PhysX. Autonomous cars are also open source which is probably the way of the industry generally looking forward. Open source is open for anyone to contribute.

GAME SDK PACKAGES

Both AMD and nVidia have a lot of software to help game developers. Some of this is tied to hardware rather tightly as game benchmarks show. More packages are being developed but open source has advantages that make it more valuable, it can be compiled for any platform.

4K GAMING

Playing games at 3840×2160 is very demanding on the core parallel ALU in video cards. A few recent games are really pushing the VRAM on the GTX 1060.