Refractive should mod GTX 480's

[TyrellBlack]

I know that this is probably (monumentally) easier said than done, but why don't you guys (and gals)
at Refractive add more RAM to the GTX 480.
After all the New quadro 6000 is nothing but a 480 with an extra 4.5 gigs of RAM.
Unfortunately NVIDIA charges $5K(!!!!) for the priviledge of owning one.
If you could mod a 480 to have 6GB of RAM and price it at $1500-$1700, you would sell bajillions of them.
Literally,.. bajillions.......
And plus If you put in 2 "Refract"-GTX 480's in a GPU-X you have a defacto Quadroplex (?),
Except for $4500 instead of $15K that NVIDIA demands.

There seems to be no legal hurdle as to selling modded equipment, as long as you state that it's modified.
Sparkle adds ram to the 460 and they get away with it.
Porsche can't touch RUF,
and NVIDIA can't touch you.
NVIDIA can't actually stop you from buying a retail 480, modding it, and selling it as such.

So get to work, start taking the 480's apart, and put your heads into it.
(Or at least tell us why my idea is very stupid.)

[havensole]

There are some differences between the Geforce series and Quadro, not just vRam. There are a lot of bios tweeks and features that one has over the other an vise versus. It would be interesting to see if one could simply buy additional ddr5 ram, if you know of some place we can pick said ram up easily I am all ears, and begin the very difficult and laboring task of desoldering the existing vram and soldering in the new stuff. I doubt refractive would want to venture into the hardware world as that takes a huge step and a lot of resources that software development doesn't. I'd think one would try first on a lower end gtx460 so that if it did end up in complete failure that they wouldn't be out so much money as a gtx480.

[timbarnes]

It's theoretically a good idea: the problem is the innovator's dilemma: what do you have the resources to do? How do you prioritize? Octane is still maturing, and there's a lot to do. I think it would be distracting for Refractive to take on hardware while there's so much to do on the software. I would prefer to see Octane establish a really clear position of leadership as a software package first: that will help them sustain the company and provide more value for us as customers.

But maybe later, when the software is done (is software ever done?)...

I think there will be companies providing higher memory versions, because CUDA is growing and clearly offers important opportunities for specialized algorithms. Octane may not need to be the one to implement your idea.

There's another issue related to the business model: profit margins for software are high; for hardware low. A software company that gets involved in hardware typically looks less financially strong because their margins drop (which affects IPO and future financing opportunities).

tim

[TyrellBlack]

Wow guys, really fast response.
Unfortunately, your reasoning actually makes sense, which was kinda expected.
Although, I thought they were already in the hardware business with GPU-X.
So it wouldn't be that much of a stretch to add more RAM to a vid card,
then add on an extra $800 for their trouble.

Have any of you actually modded a modern vid card before?
There are kits to water-cool the GTX 480, so why not to add more memory.
Couldthe process actually be as easy as remove old RAM, solder on new RAM.
Or is it just like over-gearing your car to raise your top speed,...not as simple as it seems,
with a lot of tradeoffs.

[havensole]

Relating adding a water cooling system to a gpu is very different from adding more ram. Its not like system ram where it is readily available and easy to just add it to an open or existing slot. Most dedicated ram, like on gpu's, is surface soldered to the pcb surface and held in place with some sort of epoxy. This all poses a lot of problems. I'm sure it can be done, as shown by many of the hardware companies out there. I think the biggest thing is finding the vram as the manufactures of those chips generally only want to sell in big bulk quantities. Another issue might be whether or not the gpu would recognize the added ram without needing to really dive into the bios and such. This might be a better proposal for Cubix. Don't get me wrong, I've thought about taking apart one of my gpu's and doing just this, but haven't had the luxury of possible bricking one if it does go wrong, or know where to pick up the vram chips.