Could Nvidia be hedging their bets against Intel's larrabee for their reason in looking at raytracing?
I think of this as kind of a circular question, and a lot of it is conjecture on my part(not saying I'm right, but it's the impression I get from what info I can scrounge up).
Around the time that Intel and AMD were starting to push the idea that GPUs were just going to end up as functional units on CPUs nVidia was starting to find the idea of GPGPU very appealing as it could allow them a significant increase in revenue without having to enter the CPU market directly. They start having their R&D team working on GPGPU designs and how to evolve them in architectures that won't hurt them against ATi until they reached the level they wanted to be at(ATi was still a seperate entity at this point).
With the G80 we started seeing this idea pay dividends albeit to a very small market segment to start with, mainly folders and video encoders with some moderate uses for HPC available. GT2x0 pushed this quite a bit further on the HPC side adding close to complete IEEE DP which opened up a great deal of possibilities only this happened right in time for a global economic meltdown. Because they still had themselves positioned to be competitive on the graphics side they obviously didn't pay the price in lost marketshare, but their margins did take a bit of a hit.
At some point during that evolutionary process Intel started to make noise about Larrabee. Intel clearly saw the potential issue with GPGPU, their highest margin market is HPC and one that they aren't likely to want to surrender. To make matters worse, the possibility of nV getting enough general purpose computing power out of their GPUs could put Intel in the position of their higher end CPUs being marginalized even for desktop useage. How warranted that concern is long term remains to be seen, but it is certainly the direction that nVidia is taking at the moment. How would the market look if a Via $19 CPU paired with a powerful GPGPU offered end users the same experience as a $500 Intel CPU? Obviously we are a long ways off from that being a reality, but the parts we see shipped today are ones that hit the design phase years ago obviously.
So Intel decides they need to come up with a part to try and counteract nV's attempt at pushing into their market. In a realistic sense nV's main strength is that they pretty much own the high end graphics segment from high end gamer and up(close to 70% marketshare) and on a realistic basis Intel stands almost no chance of catching nV at their own game(just as nV has no chance, even if they secured a license, of catching Intel's performance on x86). So Intel comes up with the idea of making a GPGPU part which is flexible enough to enable a ray trace render engine which would totally negate nV's extensive lead in R&D(not to mention IP) in rasterizing. Given the level of flexibility that nV has with its' upcoming designs, it takes little effort to make a very competitive ray tracing solution of their own to combat Intel. I honestly wouldn't be shocked if nV offered significantly better performance then Intel while ray tracing out of the gate. I don't see a lot of industry support for the new rendering technique either way, but it is possible that Intel will try and use considerable financial backing to try and push developers into using their new rendering technique so they can at least take over the portion of the market that nV doesn't already have. I can see them on a realistic basis having Abrash work up an engine and then license it for next to nothing(or even nothing) to help promote their new rendering techniques and then use pricing to battle nV.
nV's specialty is graphics, and they aren't dealing with the type of legacy architecture for general computing Intel is, so I wouldn't be shocked in the least if they end up faster as a ray tracer then Larrabee from day one, and likely by a decent amount. On the flip side of that, I think Larrabee is going to prove more flexible as a GPGPU from day one and by a reasonable amount too, and it's likely to hold that advantage for quite some time. While for the people in this forum, myself included, gaming performance is going to be the most important factor in either of these new parts, I think that both nV and Intel are going to be far more concerned with the level of GPGPU performance they can extract from their parts as long terms that is likely to be the larger factor in long term success in this segment. Yes, gamers are going to be more interested in gaming performance and that will not change at any point really, but the idea of nV being able to push at least a $150 part into almost
every new PC sold is far too lucrative for them to ignore.
For the other party in this race, it really seems to me that AMD stayed focus on their original idea of getting GPUs small enough to put on die with their CPUs. I can honestly see this as something that could work out well for them as a business perspective. It seems that if they can manage to pull this off moving forward the note book and netbook market segments would be very interested in making use of these parts and that is a very rapidly growing segment. May not be the highest margin market, but it also would be a bit safer in terms of amounts of R&D required.
I could be wrong on all of this, but is seems to me to be where everyone is heading.
Facebook
Twitter