Page 1 of 1
Posted: Fri Feb 13, 2004 12:26 am
not so much a request as a question... yafray has some killer results and therfore logically uses amazing amounts of computation.
would it be possible (theoretically speaking) to make some sort of rendering card for it? the renderdrive card http://www.art-render.com
seems like a great idea, it is said to accelerate raytracing by up to 400% but is priced around 3000$.
the cost is of course validated by the development costs of the company and the limited choices of similair products.
the manufacturing cost of the card is somewhere around 20$, see where i'm getting?
3 years development + 50 coders salaries for 3 years + buying propriety code for 3dsmax, maya, lightwave = 3000$ a pop
but with free coding and development... that's a whole nother story.
my idea would be to make some sort of open source rendercard - open enough to write code for almost any raytracer. development costs would be kept low by having the enourmous codemonkey resource of the oss society working on it, and manufacturing costs could probably be funded by the oncheering crowd.
we've seen it done before, with blender being bailed out and the new 2.3 manual funding. the revenue for the cards would give the blender foundation all the funding it needs, while having the increased performance of the yafray-blender combo make it a serious contender to practically anything out there!
Could it be done?
you tell me...
Posted: Fri Feb 13, 2004 2:29 am
Well don't think of this as a money making scheme for the blender community. I honestly think it's doubtfull you'd sell enough cards to really bring in enough.
But it is definitely a good idea. Making a piece of hardware that has such an "open structure" that all sorts of applications can tap into it's power.
I don't know enough about it, but it seems plausible.
I think it comes down to... are you yourself nuts enough to do it? If you believe in the idea, go for it. Find one or two more people... and who knows... maybe you'll make some money off off it in the end but don't take that into account right off the bat. Make it... and if it works... hell I'll buy 1... or 2
Posted: Fri Feb 13, 2004 2:43 am
for open source chip designs. Some can be implemented on a FPGA.
real time global illumination on GPU
Posted: Fri Feb 13, 2004 3:07 am
Here's a link to a master's thesis detailing a real time global illumination algorithm on an ATI Radeon 9800.
It would be worth it to take a look and see if a basic framework could be built which others could then expand upon.
Posted: Fri Feb 13, 2004 9:27 am
On a Funny note.
I made(in3d) a Blender all-in-one Console called the blender box!
Sell a few thousand of these puppies with your render chip
http://www.free-webspace.biz/Blenderpro ... derbox.htm
Posted: Fri Feb 13, 2004 9:42 am
i wasn't thinking that i alone would create it. i was more thinking along the lines of us as a community work out the theory if it could be done, what would be the best way to do it and if it all works out help fund the initial manufacture of it. if the cards boost preformance enough then they would definetley sell them selves. just think about giving the artists a tool that multiplies their computational power by 400% or more, the PURE card was made in 1999, one would think we could do better now...
the blender community is the only one oss community with enough manpower, cash and the need to do it. (unless you want to spend 20.000$ on generic computers that aren't geared towards rendering)
i was actually thinking of this card starting the open source hardware movement. i mean c'mon it's bound to happen someday, with microsofts "trusted computing" and whatnot just around the corner. most of the hardware out there is amazingly cheap to manufacture, it's the thinktank work behind it that makes it expensive.
Posted: Fri Feb 13, 2004 10:27 am
the blender box looks smashing! if the card works out okay, then maybe the blender box could be some kind of overkill version of it with another 200% capacity increase and built in clustering capability over firewire or something...
one can only hope... hang on i'll think i'll have ton take at least a look at my idea...
Posted: Fri Feb 13, 2004 10:30 am
It's a nice idea! Sadly, I think the development and manufacturing costs would make it almost impossible to realise - since this would have to be paid up front... Don't forget you need to sell this to people who don't often have cash to spend on nice toys.....
But you gave me another idea: maybe this is a bit easier to do. It would be a "plug n play" renderfarm!
I was thinking that there are plenty of oldish pc's lying around doing nothing, and you can pick these up really cheaply (especially if you don't need a monitor, keyboard etc).
So you'd have to make and sell a little "box" which could communicate to just about any old PC (don't ask me how though!), and a bootable CD (linux?) with the necessary software you'd put in each computer connected to the box.
Your normal desktop PC would be the controller for all this.... Anyone think this is possible?
Posted: Fri Feb 13, 2004 10:39 am
the development cost for it would be close to 0, the manufacturing cost for say 10.000 units would be in the range of 8-25$ a piece (otherwise it isn't possible, remember what i said about working the idea out before development?) and say that we initially sell these at somewhere around 240$ initially, that makes 1.100 cards pay for all of the expenses suffered at first hand, and the rest of them would probably sell them selves with the first purchases being used for amazing renders.
i still think it's a good idea and should be atleast broken down an analysed to see if it can be done by someone with a bit more technical know how than little old me...
Posted: Fri Feb 13, 2004 10:43 am
leon, what you are talking about is a general purpouse cluster. the little box you are talking about is an ethernet switch and can be picked up at around 20$ (dependent on size), and what is needed for the computers to communicate are *tada* network cards. you can use these computers together as a beowulf or say open mosix cluster for practically nothing, but you have to get amazing amounts of them to replace, at least preformance wise, a more or less top of the line pc...
Use existing programmable GPU cards
Posted: Fri Feb 13, 2004 11:28 am
Using the existing reprogrammable graphics cards from Nvidia and ATI would get around the hardware cost problem. People are already using them to create global illumination rendering systems and publishing their results on how they did it. The bridging step for the Blender community would be to create an open source implementation for widely available cards, and to create links to Blender.
The easiest way would perhaps be to modify one of the existing implementations to use a existing rendering standard for its inputs, either YafRay or Renderman would probably be fine. The paper and demonstration video at http://graphics.stanford.edu/papers/photongfx/
are quite impressive.
Posted: Fri Feb 13, 2004 12:20 pm
WOW! i was told that GI and radiosity simulations could not be done in gpu, but evidently this is not true. and the speed of it all... on a downside though, they claim that it can only be done in direct x but believe it to be possible with open gl in the near future.
how could i contribute to this feature being included in blender besides pestering the authors of it or pleading to the yafray coders?
Posted: Fri Feb 13, 2004 8:12 pm
I can see it now - a system similar to FilePlanet where you "wait in line" for a turn to use the large render farms via verse. The mesh/object/material data is sent and the calculated stuff is returned back.
Would that work? It would probably cost tons of money for the bandwidth and all but it would be cool! There could be "premium" members who pay a certain fee a month and get to bypass the waiting lines.
But hey it's just a dream [echo] dream dream dream...