Gallium3D's LLVMpipe Software Rasterizer Is Kicking

[suokko]

// totally offtopic

I would like to know that too.

It compares to the Intel Q6600...

I get single core 30fps raycasting, 15fps raytracing (shadow one bounce) and 3-5fps shadow (3 bounces) + global illumination + Reflection + volumetric lights + depth of field +... well everything lol @ 640x480 res. Benched with Realstorm Global Illumination bench 2006 on Windows XP.
Link: http://www.realtimeraytrace.de/

Conclusion: CPU's are getting powerful enough to render 3D scenes, yay!

Wolfenstein 3D? I know it was cheating but still

CPUs have been long able to render real time 3D graphics. But special purpose hardware in GPUs just can render more complex scenes in real time.

[whizse]

25 fps at 320x240 or at which res?

640x480 apparently:

http://zrusin.blogspot.com/2010/03/software-renderer.html?showComment=1268358433118#c2254273600518124890

[V!NCENT]

Wolfenstein 3D? I know it was cheating but still

Rofl No that same raytracing global illumination benchmark but without bounces xD

And pffffffffffffff.... tsssssssssssssssss... loser!! Mu computer can handle Final Doom! You honestly thought that it was so weak that I could only play Wolf3D? Get a life! xD

CPUs have been long able to render real time 3D graphics. But special purpose hardware in GPUs just can render more complex scenes in real time.

And by complex you are reffering to geometry?

Not long ago the fixed graphic pipeline was only able to render a limited set of triangles...

[drag]

If you want to get really technical we are all doing software rendering nowadays pretty much no matter what. The idea that you are rendering OpenGL on hardware is as dead fixed function hardware.

The GPU is just a graphics processor. The drivers and opengl stacks are mostly just software designed to be executed on the GPU.

A big issue, I suppose between using the GPU and CPU together to render (which would get you the best performance as the CPU is faster at some stuff and the GPU is faster at others) is the memory bandwidth and latency involved with moving large amounts of rendering information/textures from the video memory to main memory and back again.

Just the act of shoving a instruction down the PCI Express bus probably burns through thousands of wasted CPU and GPU cycles on both your CPU and your video card.

As you can imagine then shoving the GPU into the processor should theoretically yield the highest performance possible while reducing costs... Instead of spending money on sticking extremely fast memory on the far end of a PCI Express buss you should be able to spend the same amount of money on making more memory bandwidth available to the on-die memory controller in your CPU.

Intel is already shipping processors with GPUs integrated. They are budget items, but as you start to see things like AMD Fusion hitting the market these sort of 'software rendering' advances should end up being a huge huge deal. And it should yeild big performance increases for all sorts of applications, hopefully, and not just games.

Well, Hopefully.

[DavidNielsen]

Wow, did I write that? I need a nap.

Welcome to the club, 4 hours of sleep. Working hard on new idea.. sleep is for the weak (and the old like me.. nap time!)

[unix_epoch]

Ehm... Launch UT2K4 and in the options menu switch it to software rendering?

UT2K4 uses the Pixomatic 3 lib. The creater is now hired by Intel to work on software graphics. From Wikipedia: http://en.wikipedia.org/wiki/Michael...ame_programmer

Abrash's Black Book is my favorite programming book. I read a good chunk of it straight through. If anyone could design a lightning fast software renderer, it's him, and I'm not surprised he was/is one of the creators of Larrabee. Someone needs to win him over to Linux and Mac instead of Microsoft-specific stuff, though .