Nobody knows for sure, but if you want to ask someone with a good sense of the past and future of game development, Epic founder Tim Sweeney would be a fine place to start. So the fact that he said recently that we’ll need GPUs capable of 40 teraflops to reach photorealistic detail levels in games, gives us a rough idea of what we can expect.
To give that statement some context, the recently released GTX 1080 is capable of putting out around 9 teraflops. If you’re thinking putting four of these together would future-proof yourself for the next few decades, Nvidia shot that idea down emphatically.
That is around double what the GTX 980 was capable, and while we can’t expect that sort of jump every generation, it may be that a dual-GPU setup with the right optimized game could potentially deliver photo-realistic graphics within a few generations.
However, as Sweeney pointed out in his chat with GameSpot, visual fidelity on that sort of level is possible with today’s hardware through static scenes. It becomes much harder to render when you factor in players (especially multiples of them) with reflective surfaces, or more complex, busy environments.
Even NPCs throw a spanner in the works, as having them move and react realistically is much more resource intensive — potentially — than realistic-looking static scenes. “Simulating real humans,” will be the next big challenge, he claimed.
“If you gave us an infinitely fast computer, we still don’t have the algorithm [to simulate human interaction],” he said, pointing out that the problem runs much deeper than just the hardware available: we simply haven’t made software smart enough to do it yet.
So how far away does he think that sort of AI development is? Decades at best, he thinks — so get your PlayStation XXVIII pre-order ready!
