At least once a week we (the Krakatoa developers) google the web to see if anything new has been posted related to our favorite piece of software. On good days, we find a new creation by Matthias Müller. (If you have been living under a rock, go check out his particle work on YouTube - http://www.youtube.com/user/MatthiasmVideos )
Some times I find comments related to such videos that are quite amusing. For one, people still consider one million particles a big deal. On a modern machine (I just bought one yesterday, but about that later), one million particles require about 0.6 seconds for lighting with one light and about as long to render, textured with a couple of procedural maps. Including the PFlow processing and getting the particles into memory, it still takes less than 2 seconds. The new Voxel mode is a bit slower, in the same benchmark it took slightly more than 3 seconds. Around the time Krakatoa was first released, we considered 1MP (MegaPoints!) a typical amount for quick testing. With the new Intel i7 CPUs, we are seeing ourselves forced to move to 10MP just to get rid of statistical errors.
But people still assume particle rendering requires renderfarms working on frames for days. If I may quote one comment, "It's amazing what you can do with brute force computation these days". The fact is hard disk and memory bandwidth still don't allow us to saturate 8 cores at 100% - one can get about 75% in Voxel Mode, and in Particle Mode Krakatoa is still single-threaded, drawing at nearly 2MP/second. Still, we have no plans to make it realtime, CUDA or not. The problem has never been the drawing performance, but the time it takes to get all the gigabytes of data into memory, and graphics memory is today where RAM was a couple of years ago - my card has "just" 1 GB...
So I got a new top of the line gaming machine the other day. 4 cores i7 920, shows up as 8 cores in Task Manager. My office workstation has Dual QuadCore Xeons and costs a fortune (I assume), but this new gaming box for under $2K totally kicks its behind! I had to start writing a public benchmark that could be used to compare the performance of all systems out there running Krakatoa to find out how the two compare. Before buying it, I studied all the hardware site reports about how Tri-channel memory does not bring any advantage in typical applications. Well, it is quite possible that it brings some to particle rendering - if you have not heard yet, Krakatoa is quite memory intensive! While people were telling stories about how 64 bit would not bring performance improvements or would actually cause slowdowns, Krakatoa was twice as fast in 64 bit since day one. I would love to compare my results to a similar system with dual-channel memory to see whether the CPUs have anything to do with the performance differences. Hopefully once the benchmark script is released, we will be able to provide the answer.
Apropos scripting... While reading through the Krakatoa-related talkbacks on one of the video sites, I found a comment that made me thinking. It said "Remember when Bobo was writing Maxscripts back in the day?"
I kind of remember those days. Yesterday was one of them. The big difference is now I write them as a job. MAXScript actually manages to put bread (and chocolate, and ketchup) on my table. VR Boboland really has one or two new scripts a year (although DIMaster, NextPlease and LightTouch are quite cool ones and a lot more polished than most of my earlier works). And I still try to go to CGTalk and The Area and help people out, but the days of me writing eveybody's tools for free are probably over. Still, I had not thought about this until reading that post.
Those days were good, but writing the Krakatoa GUI is quite an adventure, too...