Epic Games CEO thinks we need 40TFLOPs for photo real gaming

[HEXUS]

And more for realistic dynamic human responses; emotions, expressions and so on.

Read more.

[3s-gtech]

That photo of Tim Sweeney is terrible and looks stretched - can it be resized in the article?

[darcotech]

And his photo has aliasing also, can you please use better GPU so we can get realistic graphics ? ;-b

On the side note, we are close to 10TFlops,in theory they could already test it with four cards, in two years we hope to double the performance so we will need two cards....so should be on the shelf for everybody in about 10 years?

All we need is great programmers to make it happen.

[kalniel]

That photo of Tim Sweeney is terrible and looks stretched - can it be resized in the article?

Not enough FLOPs

[theleader]

There must be more then enough FLOPs as we are all living in a virtual universe anyway! SD,HD or UHD version don't know!

The end of the universe will not end in a violent event, just a blue screen with error indication and a Press any key message to reboot!

You may ask how I know that we all live in a virtual universe, it's because, early in the morning after a heavy night, everything looks pixellated!

[peterb]

Non issue surely? As more people live in a VR they will forget how to interpret the subtle emotional clues so they won't matter any more...





/fetches coat

[outwar6010]

Or a fraction of that and a good dev....

[Chaoss]

This article is very mis-leading and it looks written by someone not well informed in how graphics and GPU's work or the different ways 3d graphics can be created.

We have, right now as I type this the power to render a perfectly photo realistic scene in VR at the full 90fps, what's more is even a NVidia GTX 980 (not Ti) has the welly to render this. The 'problem' with this is that the environment would be captured using photogrammetry which means the scene would look very static and stale and the shadows would be un-moving, meaning adding things to the scene in real time means they'd cast no shadows that match the environments.

The 2 giant elephants in the room when trying to get photo-realistic environments that can be interacted with are real-time lighting (complete with global illumination and soft-shadowing) and fluid simulations. Massive strides are being made towards both of these with lighting taking a small lead ahead of fluid simulation. We could probably, with some 'John Carmack' level of genius render this on current-generation technology (GTX 1080/1070). with 40TFLOPS. We can currently render humans and their facial expressions with almost near perfect accuracy and we crossed the uncanny valley back in 2013, the work required to put this into games is a lot and many developers don't have the time or funds. Hair is however still a bit of an issue but the rest of it is there.

We need more universal solutions to come out (like a program that specialises in making human characters only and exporting them rigged, with animations and ready to go for both indie's and AAA studios) and similar solutions for lighting. Right now we have many very bright minds trying to compete and it is to fragmented.

[Mblaster]

Whatever happened to ray tracing which was once touted as the future of super realistic graphics? Have we still not reached the required GPU power or did some other method displace it?

[kalniel]

Whatever happened to ray tracing which was once touted as the future of super realistic graphics? Have we still not reached the required GPU power or did some other method displace it?

Still too expensive, but some elements are now ray-casted like global illumination.

[HW90]

Whatever happened to ray tracing which was once touted as the future of super realistic graphics? Have we still not reached the required GPU power or did some other method displace it?

Still too expensive, but some elements are now ray-casted like global illumination.

Expensive how? Because there have been real time ray tracing benchmarks done on mobile GPUs let alone top of the line desktop GPUs

[kalniel]

Expensive how? Because there have been real time ray tracing benchmarks done on mobile GPUs let alone top of the line desktop GPUs

How is easy to explain - it takes lots of calculations and memory to fully bounce rays around a typical gaming scene with all the different properties and there hasn't been as much investment in hardware acceleration for this compared to rasterisation. Even the most demanding of point based approaches like ambient occlusion are roughly three times faster than a simple ray trace and you don't have to worry about jittering tricks for AA etc.

[Mblaster]

Expensive how? Because there have been real time ray tracing benchmarks done on mobile GPUs let alone top of the line desktop GPUs

I looked around a bit more last night after reading this thread, this was the best example of real time ray tracing (actually path tracing) that I could find, running on 2 GTX TITANs: https://brigade.otoy.com/gtc/brigade...-gtc-2014.html

As I understand it the graininess comes from not being able to trace enough rays of light in real time to build up a clear picture, they limit the number of rays and/or number of reflections on each ray to improve performance at the cost of image quality.

[bledd]

But can it play Crysis ?

[Shoom]

And his photo has aliasing also, can you please use better GPU so we can get realistic graphics ? ;-b

On the side note, we are close to 10TFlops,in theory they could already test it with four cards, in two years we hope to double the performance so we will need two cards....so should be on the shelf for everybody in about 10 years?

All we need is great programmers to make it happen.

The next Titan X can reportably push 11, so we are getting there.

[blockedSounds]

So long as 'photo-realistic' isn't anything like current CGI in movies (I'm looking at you Batman Vs Superman..) and actually resembles the real world then that's cool.