Frid...*ahem* Saturday with Means - June 13th, 2009 - I'm not dead.

[Vhax]

okay, I'm trying tofollow along on the sidelines here.

So with this raytracing, you could have a high number of polygons without a performance hit, but having say 2048 x 2048 resolution textures as compared to 1024 x 1024 textures would be the bog down factor? Or did you mean another type of resolution? I'm a software n00b, so forgive me if I seem dumb here.

Ray tracing has a much higher initial performance hit based which is on the rendering resolution (size of window or screen resolution if fullscreen).
A highly optimized ray tracer will get slower with more polygons, but it scales pretty well where 1 mil vs 2 mil doesn't make a really huge difference.
However, ray tracers do heavily slow down on reflective/refractive surfaces.
Also overlapping lights also impact performance quite a bit as more shadow rays need to be cast.
Texture size doesn't matter much, it just takes more memory.

[PhoenixOfAges]

so we had to wait a whole day last week, we getting thursday with means this week to even things out?

Even though it was called Friday with Means we actually had a Thursday with Means not too long ago. In fact it may have been so early in some places it might have just barely been Wednesday with Means.

[Solstiare]

Ray tracing has a much higher initial performance hit based which is on the rendering resolution (size of window or screen resolution if fullscreen).
A highly optimized ray tracer will get slower with more polygons, but it scales pretty well where 1 mil vs 2 mil doesn't make a really huge difference.
However, ray tracers do heavily slow down on reflective/refractive surfaces.
Also overlapping lights also impact performance quite a bit as more shadow rays need to be cast.
Texture size doesn't matter much, it just takes more memory.

Which is my point, to get full raytracing to work in a live game environment (especially with reflect/fractive whatever), all those lit particle effects from buffs, nano effects, gunfire, and in pand the moving shadows cast by everything, in fact ANY extra light sources all create a performance penalty. You can get around some of it with "dart casting" from the light sources, vs full per-pixel raytrace, but then you have a whole other set of issues with accuracy and blending all of which get worse if the light sources are in motion.

And this is not even including caustics (light distortion from transparent solids), even stationary ones. Bottom line is until we can harness human stupidity to power floating point calcs, someone is going to have to dial up the horsepower of graphics chipsets and cpus.

But it would be pure eye candy if someone can figure out how to make it work in the here and now.

[Mastablasta]

How about a handful of teraflops? raytracing scales extremely well across multiple processing units, so it's not that far off to do stuff in realtime when it's worked towards

[Bugzybug]

right and where's my beowulf cluster.


TBH with the dreamworld2 engine maybe we can get support for dx11 hw tessellation,

but real raytracing is something we won't have for a while.

Dynamic lighting on the other hand, is much cheaper on processing power.

[Solstiare]

I will be content to have w/e attack nano I'm using atm to light the front of my character and the surrounding environment, esp when its dark.

raytracing scales extremely well across multiple processing units, so it's not that far off to do stuff in realtime when it's worked towards

No its not that far off, just need 2-3 GPU's on one card to handle the workload, and the software threading to take advantage of it.

[Mastablasta]

GPU pipelines are too short to be efficient for raytracing?

[Vhax]

Which is my point, to get full raytracing to work in a live game environment (especially with reflect/fractive whatever), all those lit particle effects from buffs, nano effects, gunfire, and in pand the moving shadows cast by everything, in fact ANY extra light sources all create a performance penalty. You can get around some of it with "dart casting" from the light sources, vs full per-pixel raytrace, but then you have a whole other set of issues with accuracy and blending all of which get worse if the light sources are in motion.

And this is not even including caustics (light distortion from transparent solids), even stationary ones. Bottom line is until we can harness human stupidity to power floating point calcs, someone is going to have to dial up the horsepower of graphics chipsets and cpus.

But it would be pure eye candy if someone can figure out how to make it work in the here and now.

It's not as slow as you think it is though.
Ofcourse don't start putting in stuff like caustics, a GPU can't do that either anyways. (That's closer to path tracing than ray tracing anyways).

It's not too different from a GPU in terms of light btw, GPU's also take quite some impact from multiple lights (depending on their implementation for lights).
Just don't make sure too many lights affect the same pixel, the same goes for GPU's. (Unless you setup a whole deferred shading pipeline and stuff).

I'm certain a scene of AO's complexity can run in real-time in Arauna on a high end machine on a low resolution, with Larrabee it would run in real-time on a high resolution even.
Rasterizing has had dedicated hardware to their single purpose for years, it's their edge. A CPU isn't particularly optimized for ray tracing and yet it's what we have to settle with for now. However, when Larrabee hits we finally do have hardware that is very suited for ray tracing.

Personally I don't believe ray tracing to be great for games since it doesn't outperform GPU's and there's plenty of visual hacks for the GPU that are completely incorrect but do look pretty.
But it's certainly possible.

No its not that far off, just need 2-3 GPU's on one card to handle the workload, and the software threading to take advantage of it.

GPU's suck at global data. Yet it's the biggest thing ray tracing needs.
If you look at images they claimed to have been ray traced with a GPU there's always some quite obvious restrictions they don't tell you about.
For example, often they have a very simple material, no proper reflections (reflecting a skybox only for example) or other sillyness like that.
Intel agrees on this an it's one of the reasons Larrabee is more like a CPU than a GPU.

[Troxxor]

Bungerman does.

Indeed! His pants are without shine, glint or power QQ

[Eruslavuro]

Before people fill this thread with meaningless replies...

Serious question for you Means.

Since you are planning on using AoC graphics engine, are you planning on using other features from that game as well (not necessarily bounded to graphics only)? CC tools - the way they are handled, pets - their role and behavior comes to my mind.

First of, i just came back from 3 years break so i´m a bit out of the news. The AoC graphic Engine is true? Content is really important in the game and is what make most people want to play and improve their characters but the graphics is also important. I hope this is in fact true and not planed for 2030..

[Troxxor]

It's true

With additional resources from the larger technical team we hope to bring the Dreamworld engine to AO in 2009. I truly believe that Age of Conan is a visually stunning world and I look forward to seeing the same technology available to us in the near future.

[Nosgoth]

Has anyone noticed that some FC employees have forced us to discuss about pants in the last two FWMs? First Macrosun with his post about pants being the key to space travel and now Means/Enno with their shiny pants. Hope we're not victims for an insider joke or something.

[shadowgod]

Has anyone noticed that some FC employees have forced us to discuss about pants in the last two FWMs? First Macrosun with his post about pants being the key to space travel and now Means/Enno with their shiny pants. Hope we're not victims for an insider joke or something.

no joke. pants are they key to space travel.

[Vhax]

Has anyone noticed that some FC employees have forced us to discuss about pants in the last two FWMs? First Macrosun with his post about pants being the key to space travel and now Means/Enno with their shiny pants. Hope we're not victims for an insider joke or something.

Pants has been a subject of conversation for quite a longer period
Fearsome pants!

[Solstiare]

MAN PANTS.

[Solstiare]

anyway, back to the running discussion.

I never said ray tracing was slow, that was you! I said (in synopsis) that it would take a beast of a system to push through as a slide show given the current gpu/cpu available to your average end user right now.

The generation 2 larrabee and/or Nvidia's or ATI's response* to Intell will most likely handle the previously mentioned live raid environment without any real issues, at a decent resolution and texture depth. Limiting the amount of light sources affecting a surface,** selecting surface over pixel,*** and seting the maximum number of times a ray can be reflected to 1 or 2, may not even be necessary at that point. As for time line, Intel had Q1 or Q2 of 2010 as a release date for the first generation larrabee. So no, its not that far off, and it is not too slow now, just too slow atm to be useful in a live environment. Which was my (overstated) point. Couple years from now? Oh my yes.

Rasterizing has had dedicated hardware to their single purpose for years, it's their edge. A CPU isn't particularly optimized for ray tracing and yet it's what we have to settle with for now. However, when Larrabee hits we finally do have hardware that is very suited for ray tracing.

Right. However, it's still catching up to raster based rendering, a software hack running on a hack chip that's on generation 10. Leading me to restate

Originally Posted by Solstiare
No its not that far off, just need 2-3 GPU's on one card to handle the workload, and the software threading to take advantage of it.

as: No its not that far off, just need 2-3 Hybrid chips on one card to handle the workload, and the software threading to take advantage of it.


Now this is all academic, since its Mature Tech vs. In Development. This is the biggest hurdle imho, more so then the technical side of RTRT. If you are producing a game and you want it working, shiny, and on the market NOW, (AO?) its raster-based. We won't see real time ray tracing creeping into live game environments until a developer takes a leap and the title is a financial success. I'll have functional dedicated hardware supported RTRT in BLENDER and MAYA/3dsMAX long before it's in any games (duh), but how long before it bleeds into live graphics engines is anyones guess. It will probably coincide with a larrabee-esq card droping under 600$.


*nothing over here about nvidia working on anything like Larrabee. They have been su****iously quite about this.

**pick the brightest two, the highest contrast, and discard similar shaded lights that affect a surface

***dart casting with a limiting radius based on light source type (point light, spot light, self illuminated surface, ect) and edge diffusion

[Vhax]

My point kinda was: it's not that far off.

It won't take a second generation Larrabee, the initial generation will already push out real-time ray tracing at a speed suitable for games.
It doesn't mean alot of developers will jump the wagon, but they could.
Also, scenes of AO's complexity can already be done in real-time (on low resolutions though) on current day hardware.
And to add to the fun, there's been some good progress on adding real-time global illumination to ray tracers as well.

Don't compare it to the stuff Maya/Max produce with ray tracing, it's quite a different approach.

[Solstiare]

1st gen for it to function, 2nd (1.1?) gen for it to work. Pentium floating point error anyone?

Yes, RTRT available now in low res, but who wants to play @ 640x480.
Higher Res (1280x1024+) is not far off, just not now.

The ray tracing in production 3dR is NOT real time. You are right, it's a completely different approach, I'm not comparing them. But dedicated, purpose-built, hardware supported RTRT will be available there before it's in a live engine.

[Axtilmor]

We'll probably never see specialized hardware for raycasting. The general trend has been for the GPU to become more powerful and versatile, and less specialized. Raytracing on the GPU wasn't practical at all before loops and breaks were available in the shader. Even, GPUs still have a lot of vestiges of the rasterization pipeline that aren't really relevant to raytracing.

As I understand it, the big advantages of Larrabee will be that it has a fully configurable pipeline, and that the processors it uses are just stripped-down Pentiums. Thus, it should be much more suitable to raytracing.

I haven't read much about it, but I think some approaches to raytracing rasterize the first hit of the ray, so the rasterization pipeline may not disappear all that soon even if raytracing is adopted universally.

[Mastablasta]

We'll probably never see specialized hardware for raycasting.

Intel is working on a piece and there's another upstart that made something that also does realtime raytracing on a dedicated piece of hardware. Just a matter of time really.