I think you are overstating the relevance of the IP licensing angle
Actually, I think he's probably understating it by a lot. If an appeals doesn't come through soon, Apple is going to be barred from selling computers in certain countries with AMD GPUs because they lack the proper S3TC license and the chips use the technology. Relatively speaking, it is a very simple patent, but is just a singular example of the mine field of IP that is involved in the graphics industry and is a major reason why Intel failed despite throwing billions at the problem while already having enormous resources in every area outside of IP.
Back in the early days of real time 3D technology(think mid 90s) we didn't even now what we were going to be using for primitives(quads, voxels, polys etc). We didn't know what sort of render structure we were going to be using(ray, scene graph, raster), we didn't know how we were going to apply images on these surfaces(procedural shading, per face color fill, texture mapping etc). None of the things we take for granted today were obvious- and there are many different books written in that time era that makes this obvious. If you want to build a GPU today you have a couple of ways of going about it. You can sell your soul to nVidia- and they *might* be willing to license some crucial IP, and then strike a deal with AMD(their IP is much weaker then nV's, but still strong enough to shut any startup down). To give you an idea of how much of an obstacle the IP is we are talking about- you can't apply a filtered texture map to a polygon using dedicated hardware without a licensing agreement.
The alternate route is to swap to a completely different rendering type. There are numerous options that you could take, ray tracing being the most likely candidate as of this point, but then you run into a massive issue. You need to have software support to run on your hardware. If you look back at the nV1 you will see that nVidia decided to bet on quads being the dominant primitive. Carmack went with polys, as did 3Dfx, and that was that. It took nV years to recover and eventually assert their dominance. They are honestly very lucky to have survived that era. As of right now, there are hundreds of millions of devices in the world that use polygon based rasterized images and every software developer can count on that.
Intel tried to rock this boat and failed losing billions in the process. That in itself should tell you exactly how obscenely high the price of entry into this market it. One of the wealthiest companies in the world that already employs thousands of the best engineers in the world and has the clear fabrication lead in the world spent billions of dollars and failed to even show up to the fight.
The largest chip had more transitors on it than a Pentium3.
This is a good example. A company trying to design something as simplistic as being comparable to a Pentium 3 and was burning ~ $100 Million a year. That is pocket calculator complexity by today's standards. A miniscule little simplistic 9.5 Million transistor device. Back in those days when designs were very simple, a relatively modest sum of money could compete. Today we are looking at 3 billion transistors. Here legacy plays an *enormous* role. AMD and nVidia don't have to design a TMU, they don't have to design a triangle setup engine, they don't have to design a shader pipeline- they already have done this.
What we see today when we look at new GPUs is over a decade's worth of $100Million + per quarter R&D- none of that vanishes. Neither AMD nor nVidia are reinventing the wheel. Even if you look at AMD's upcoming shift in shader structure, it's that one element of the chip with some refinements throughout.
If you were to start today you would be likely looking at somewhere in the 15-25Billion transistor range for the part you are going to launch. Think about that for a bit. If every engineer you hired were capable of designing a chip with the complexity of a Pentium 3 by themselves you would need ~1,578 engineers(using the low end of the estimate). Finding *one* engineer on the face of the Earth capable of such a task within a decade I would say is nigh impossible. Finding fifteen hundred of them willing to work for peanuts?
Facebook
Twitter