Right now that station’s price is at 25% discount (until?). That workstation was necessary to bring that Star Wars film showcased, it doesnt mean you would need that much power to produce a game or that consumer would need that much power to play games. As I am involved in film production atm, I am quite pleased that not only the technology will be part of Unreal, but that also the engine finally will support multi-GPU as common feature at 4.20.
While you might see some games “support” DXR in 2018, consumer GPUs are not likely to exist. This will be the pro market for 2018. This will run on normal GPUs, but without the denoising tech from Nvidia, I doubt it will be very fast. I believe everything is done with Direct Compute, so your GPU performance will scale based on raw fp performance. Compute is also resolution dependent and scene dependent.
Here is an interesting comment from one of our devs on the project:
I worked most of the time on a pretty standard 1070 GTX and got about 8-10 fps. at half of the resolution showed in the demo. Do not trust my numbers too much because I worked most of the time with debug code and experimental drivers, but in summary this technology is not meant to run just in super premium hardware.
I did a quick math based on numbers:
1080 is 26% faster than 1070
1080ti is 20% faster than 1080
TitanV is 55% faster than 1080ti
new Quadro V100 is ?? faster than TitanV
on will result in 18-23 fps which can run actually faster because it does not take into account speed per feature. I still want to see the numbers for the new graphics cards to be launched, but I will consider just purchase that DGX station, I just can’t afford it soon.
A question which seemed not being commented on all those videos from GDC:
- total resolution for that Star Wars realtime rendering was UHD, 2K or 4K?
- was it used FP16 or FP32 to achieve that?
We are working on a blog to better describe what you saw, so post questions here. I don’t know the answer to these questions.
any idea if we will have the ability to generate lightmaps with this from a developer point of view?
I think I’m allowed to say we have people working on Lightmass. They are planning much more than incremental efforts. Timeframe is not well defined. Just stay tuned for details to emerge.