Slashdot Mirror

This is realtime speed for a virtual intel CPU with about 36 GHz (to be more precise: a cluster with 20 AMD XP1800 was used). Alternativly one slow PC (1 GHz) with a hardware raytrace GPU that is 3 times more powerful then an actual prototpye could be used.

SaarCOR is a hardware architecture for generating highly realistic images of 3D environments in realtime. It implements the well-known ray-tracing algorithm on a single chip and reaches performance comparable to todays graphics technology but uses less hardware resources and requires less memory bandwidth.
[...]
A first prototype of this graphic board purely based on ray tracing was presented summer 2004. This summer on Siggraph 2005 the second prototype featuring fully programmable shading, geometry and lighting will be presented.

real time ray tracing isn't too far away

It's already here. There was this article about real time ray tracing hardware a while back. And there was this one about physics processing units. It will be interesting to see these things implemented together.

FWIW, this is not a new idea. FPGAs (i.e. dynamically reconfigurable processors) have been around for about 20 years now, and have allowed hardware developers to produce custom hardware in many situations. The key, you see, is that hardware designed for a specific task is almost always going to perform that task better than a general purpose processor. That's why the SaarCore can outperform a P4, and why your computer has a custom built GPU.

As a result, the idea of runtime-dynamic hardware sounds great. Unfortunately, the issue that developers run into in developing a runtime-dynamic processor is the matter of knowing how to configure the chip. One tack is to allow programs to load chip designs themselves, thus creating specific hardware for that individual program. The down side to this tack is that someone must go through the time consuming task of manually writing the chip in a Hardware Design Language such as VHDL or Verilog. Most programmers aren't going to do this when they can get the program out faster with a general purpose CPU.

This has led to another tack of using software to analyze a program and automatically create a machine to optimize it. This is conceptually similar to the Java JIT method, but is more complex by far. A lot of research is being done into this area (as this story shows), but I wouldn't hold my breath for now.

Another design that makes a lot of sense is the concept of "hardware on demand". i.e. Imagine if you had a library of accelerator chip designs. Whenever a program needs a particular form of common hardware acceleration (e.g. GPU, Sound, DSP, etc.), the onboard FPGAs could be reconfigured to meet the demand. This wouldn't have the same punch as task-specific hardware, but it would provide an inexpensive method for obtaining a bundle of hardware that would otherwise be extremely expensive and use up a lot of bus space.

Read this response of mine... I am aware of FPGA chips and I also understand that they are limited as to what their top performance capabilities are.

The performance gap between FPGAs and ASICs have dwindled in recent years, with FPGAs taking advantage of smaller fab processes than ASICs currently have readily available. That's why Xilinx preaches their "Make Spartan your ASIC" line and gets away with it. :-)

That's why I see this as being a difficult to get running kind of venture. Unless they can perform some kind of miracle, very few regular PC users will be interested in this sort of thing.

I wouldn't be entirely sure about that. This sort of thing should be able to be competitively priced (granted, with a poorer price/performace ratio) to where it could easily be the standard choice for OEM Linux machines.

As another poster suggested, this might work great for embedded or tiny form factor devices, like PDAs, Cell phones and similar. It will be quite a feat to see these graphics cards come close to competing with current mid to low range offerings from ATI and Nvidia.

Nonsense. You can get a lot more out of the FPGAs than most people believe. A lot of individuals still think of FPGAs as those electronic oddities from the 80's. Projects like SaarCor are demonstrating how FPGAs can outperform even a Pentium IV. Do not underestimate these guys.

To your original question, if the ASIC route is taken, the chips can either be delivered by the fab to the board manufacturer on a tape and reel, or sockets can be used to plug in the chip during packaging.

It is highly unlikely that you'll be able to
program an FPGA to do something faster than a
modern computer can do.

Now that's just nonsense. This is the thinking of "More MHz is better". The truth is that a custom chip design targetted at a specific task can easily out-perform a more generic chip. For example, the SaarCor can render a raytraced scene many times faster than a Pentium IV, using nothing more than off-the-shelf FPGA hardware running at 1/300th the MHz.

That being said, it's doubtful that the OGP will outperform someone like NVidia or ATI who already build custom chips. But it might be able to give them a decent run for their money.

Why not use ray-tracing? Not on cpu but on dedicated hardware. After all it's been covered before on slashdot and most games already use the needed data structures for levels.

If you haven't taken a gander at it yet, you may want to take a look at OpenRT and projects using OpenRT such as Quake3 Raytraced. Also take a look at the hardware architecture as well.

Ray-tracing presents a much more detailed rendering of a scene, but was always considerably slower than rasterization. If hardware-accelerated ray-tracing architecture grows in the market, you may see your skyline beautifully rendered in real-time .. with traffic, crowds, etc.

From what I remember from my neural networks days the human brain/neocortex works so well because of its massively parallel nature (not because of the processing power of any one neuron), and that computers simply aren't able to exploit this as they aren't designed to work like this

Computers aren't *normally* designed like this. They can be however, and in recent years have been moving in that direction. When neural networks were first being researched, a Cray supercomputer was about the closest you could get to that sort of parallelism. Fast forward to today and we find that Intel (Pentium), AMD (AMD64), Sun (Sparc), and Sony (Emotion Chip) are all building machines that are highly parallel in nature.

Even more interesting is that today you can build yourself a custom, massively parallel computer on a shoestring budget. All you need is a handful of FPGAs, a PCB layout service like Pad2Pad, a few other parts, and reasonable VHDL or Verilog skills. That's more or less what OpenRT did to build their SaarCORE architecture. :-)

also view: http://science.slashdot.org/comments.pl?sid=142507 &cid=11941863

it's a virtex II 6000-4
from some pdf at http://www.saarcor.de/pubs.html

Right. I attended a talk at EuroGraphics '04 by nVidia and ATi engineers (and a Microsoft guy who talked a bit about DirectX in Longhorn), and they outlined what they have in mind for the next few years. The bottom line: more of the same, but faster and more programable. They explicitly stated that they weren't going to move away from traditional rendering methods any time soon. I would be very interested in a card which did basic 2D and traditional 3D `well-enough' but also included something like SaarCOR for real 3D, especially if it could be used for volume rendering. Oh, and don't underestimate the requirements of Quartz Extreme. Exposé is not quite smooth on a GeForce 2 MX.

I was actually thinking more along the lines of http://www.saarcor.de/, i.e. for use with games.

nVidia Gelato
Mental Ray taking advantage of OTS hardware
Dedicated raytracing hardware
Raytraced Quake 3
OpenRT realtime raytracing framework

The whole scene doesn't need to reside in the RAM of the GPU. Have a look at the virtual memory architecture for ray tracing.

I'm not sure that's gonna happen. The fact of the matter is that current graphics hardware is fast approaching the point where raytracing will be irrelevant.

Actually, AFAIK the opposite is true.
Raytracers scale very nicely with geometric complexity: O(log n). So as the virtual environments continue to grow, raytracing should gain popularity over scan conversion. Have a look at this - that's 50 million triangles raytraced at 4-5 fps!

Most of the current interactive raytracing is still done on parallel computers or PC clusters, but there are a lot of optimizations that can be combined to achieve interactivity even on a single CPU. And hardware architectures are underway as well...