NVIDIA Comments On Ray Tracing and Rasterization Debate

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
NVIDIA's Chief Scientist David Kirk sat down with the crew from PC Perspective and weighed in on the current ray tracing and rasterization debate. Very interesting stuff, definitely worth clicking on over and checking out.

"Reality intrudes into the most fantastic ideas and plans. Virtually all games and professional applications make use of the modern APIs for graphics: OpenGL(tm) and DirectX(tm). These APIs use rasterization, not ray tracing. So, the present environment is almost entirely rasterization-based. We would be foolish not to build hardware that runs current applications well.
 
Ooh, maybe we'll have a "render war" in the coming years! May the best graphics win.
 
I agree. I hope they arent going to turn this into another "Format War":rolleyes:

This could be really good for the people who are complaining that a lack of competition is stagnating the video card market.
 
Lame interview.

Nvidia has everything to lose and nothing to gain with ray tracing unless they're the ones pushing it. Right now they rule the roost with high end rasterization and they don't have to shovel R&D into a wholly new unknown (for them).

Intel on the other hand has nothing to lose and everything to gain. They have absolutely no high end rasterization product.

Ray tracing models physical reality. It requires less silicon, less bandwidth and less fill rate to pull it off. And it's extremely parallelizable.

The saarcor folks proved that anti aliasing with ray tracing is as simple as running an edge detect filter on a scene and adaptively anti aliasing it. Damn fast and efficient.

All it would take would be for a next gen console to decide to use ray tracing as their primary engine and likely rasterization would then quickly fall out of favor.

Or for intel to manufacture ray tracing specific hardware with some basic legacy rasterization hardware included for backwards compatibility.

The future for graphics looks good. OpenRT is out now, I wonder what ugly hacked up beast of an API microsoft will come up with to try to be contrarian and lock in the game companies?
 
All it would take would be for a next gen console to decide to use ray tracing as their primary engine and likely rasterization would then quickly fall out of favor.

PlayStation 4 with an 80-core ray-tracing cpu/gpu ftw. You heard it here first.
 
Ray Tracing is more efficient when creating true 3D images in space (IE: Holography)

Rasterizing another thousand pixels into a Z-axis would require 1000x the processing power of todays graphics cards.
 
Lame interview.

Nvidia has everything to lose and nothing to gain with ray tracing unless they're the ones pushing it. Right now they rule the roost with high end rasterization and they don't have to shovel R&D into a wholly new unknown (for them).

I understand the thought behind this up, until you claimed Nvidia is clueless about ray tracing... Leaving aside the fact they own Mental Ray, and there own render Gelato can do ray tracing, why do you think your cards matters when rendering in OpenGL if ray tracing can only be done on the CPU?

There was too much spin in that article, considering that the future is in mia_materials, and more realistic looking cg, there are something you simply can not fake with raster. But Intel's had way more brown stuff in it, and ignores the simple fact that ray tracing is very much camera dependent meaning to render in real time nothing can be pre calutated which will kill render time(The players view is the camera). To make things worse there will be no artist in the loop to decide if 10 rays looks good or if a 100 rays are needed, so either you living with crappy looking areas, or you push up the rays every where slowing things down to crawl. The reason films look so good is because someone went through and tweaked all the setting of every frame to make sure it looked good. Kinda why they take 3 years to make and two hours to watch. Games on the other hand are built for real time where everything around the player must be built to be seen with parallax in mind. Texures can be bakes and compressed to load fast lights can be tweaked so they do not need to be computed in real time, so that when the player looks around all the math is done already. To raytrace that in real time you would have to be shooting out rays from a moving camera which the rays have to go out hit an object scatter and return to the camera telling the render what can be seen. If a render could be built to take into account that motion and cal based on what is scattered that would give better lighting than films but you would need something like the eve cluster per player just to figure out what one player was seeing.
Even with mores law trucking along, were a still about a decade out from that.
 
I wonder why didn't David mention Gelato which is basically a gpu based raytracing renderer.
 
I hope someone remembers what happened to 3Dfx when they tried to go their own "format".
 
You mean Glide? Glide is not some alternate means of rendering 3D worlds -- it's just an API. Ray tracing and rasterization are not APIs.
 
I hope someone remembers what happened to 3Dfx when they tried to go their own "format".

They ruled the 3D graphics world? For years? And people used their API in pretty much anything that was worth considering when it came to 3D games at the time? Yeah, as it happens, I do remember. Quite fondly.
 
I remember 3Dfx fondly too and I wish I kept my Voodoo2 pairs too. The comparision I was trying to make was some of us thinks one of the video card company should take the leap and switch to the superior raytracing method, and nVidia should be that company because of it's strength in the market.

I was just remembering how powerful 3Dfx was and that didn't stop them from going under because of proprietary API when everyone else was using OpenGL at the time.

But you may be right. 3Dfx was just ahead of its time and nobody was really looking into the 3D market back then.
 
I understand the thought behind this up, until you claimed Nvidia is clueless about ray tracing... Leaving aside the fact they own Mental Ray, and there own render Gelato can do ray tracing, why do you think your cards matters when rendering in OpenGL if ray tracing can only be done on the CPU?

There was too much spin in that article, considering that the future is in mia_materials, and more realistic looking cg, there are something you simply can not fake with raster. But Intel's had way more brown stuff in it, and ignores the simple fact that ray tracing is very much camera dependent meaning to render in real time nothing can be pre calutated which will kill render time(The players view is the camera). To make things worse there will be no artist in the loop to decide if 10 rays looks good or if a 100 rays are needed, so either you living with crappy looking areas, or you push up the rays every where slowing things down to crawl. The reason films look so good is because someone went through and tweaked all the setting of every frame to make sure it looked good. Kinda why they take 3 years to make and two hours to watch. Games on the other hand are built for real time where everything around the player must be built to be seen with parallax in mind. Texures can be bakes and compressed to load fast lights can be tweaked so they do not need to be computed in real time, so that when the player looks around all the math is done already. To raytrace that in real time you would have to be shooting out rays from a moving camera which the rays have to go out hit an object scatter and return to the camera telling the render what can be seen. If a render could be built to take into account that motion and cal based on what is scattered that would give better lighting than films but you would need something like the eve cluster per player just to figure out what one player was seeing.
Even with mores law trucking along, were a still about a decade out from that.

Hmmm...we have a commercial product out now that does ray casting with an indepent camera position & orientation which needs to be calcuated for every single scanline. It's a pretty dumb implementation but it's fully threaded. We're getting 36MP/second with 8 core 1.6 clovertown with 94% efficiency, no SSE. Same stuff running on a 2.33 harpertown showed linear speed increase. IO throughput is ~4MB/second for pushing the pixels onto the scene. We even demo with an external usb powered 20GB notebook drive.

I realize there is some algorithmic work still needed with real time ray tracing, including efficient scene organization with moving objects, but those problems don't seem overwhelming to me.
 
I remember 3Dfx fondly too and I wish I kept my Voodoo2 pairs too. The comparision I was trying to make was some of us thinks one of the video card company should take the leap and switch to the superior raytracing method, and nVidia should be that company because of it's strength in the market.

I was just remembering how powerful 3Dfx was and that didn't stop them from going under because of proprietary API when everyone else was using OpenGL at the time.

But you may be right. 3Dfx was just ahead of its time and nobody was really looking into the 3D market back then.

You've got your order of events out of whack.

Glide came along and almost every dev studio making 3D games used it. 3DFX was untouchable for a while there. Then there was a shift from 16bit to 32bit color rendering, 3DFX's architecture wasn't as good at this and they began to fall behind. They made some headway and it looks as though their next major architecture would've made the necessary improvements to 32bit performance to keep them in the market, but bad management meant they ran low on funds and were bought out by Nvidia before they could bring the next card to market.

Glide didn't kill 3DFX, and it's not as if their cards couldn't run OpenGL applications well (or Direct3D for that matter). They just made some really poor monetary decisions.
 
Back
Top