Microsoft Patents GPU-Accelerated Video Encoding | Hacker News Reader

Microsoft Patents GPU-Accelerated Video Encoding

conceivablytech.com

15 pointsadbge15 comments

How can this be worthy of a patent? This has been the example of GPGPU from day one. It is not even a technological invention, it is an application of existing technology. It is like IBM patenting "using a computer for business administration" or a kid patenting "swinging sideways on a child's swing" (that last one is actually not made up: see http://www.newscientist.com/article/dn2178-boy-takes-swing-a...)

update: I just noticed that the New Scientist article actually refers to a patent for "a system that issues reservations for using the toilet on an aeroplane", issued to four IBM developers. How ironic :P

It is not even a technological invention, it is an application of existing technology.

Can you cite an example of an invention which was not an application of existing technology?

Do you want to get into a philosophical discussion on what constitutes technology? :P

I believe patents should be used to protect inventions of new technology (e.g. things like light bulbs, telephones (ouch) or cars), not "methods for illuminating the workplace", "methods for notifying people of cancelled appointments" or "ways to transport oneself from home to work".

Do you want to get into a philosophical discussion on what constitutes technology?

No. My question was unambiguous.

Do you want to fight it?

Do your customers, shareholders, managers, investors want you to fight it?

Or would you agree to buy MSFT's product if it included a promise not to sue you?

I am not saying it does not make sense for Microsoft to pursue a patent like this; I am saying that the fact that it does illustrates how out-of-touch the patent system has become.

Patents were meant to protect Patents have by themselves become a source of revenue, used by big corporations to sit on their lazy ass. They no longer hold any value for the small inventors they were meant to protect in the first place.

Well apparently Intel is totally unaware of this fact, so it's news to somebody.

Just as a side comment, x264 using one of its faster speed presets is both faster and a better compressor (sometimes at the same time!) than any GPU-accelerated encoder I know of. Attempts to port part of it to GPGPU are welcome, if they're faster and not worse, but so far 5+ attempts at it seem to have failed.

Luckily, the concept of CPU-accelerated video encoding is not patented.

It will be interesting to compare x264 against Sandy Bridge's GPU encoder; they added a scoreboard to accelerate wavefront-parallel motion estimation (AFAIK this has been problematic for previous GPU encoders).

I'm not sure this patent is very strong. Anybody who did motion estimation on video frames prior to the 2004 filing and did it with a co-processor could probably claim to be prior art on the idea of GPU accelerated motion estimation. The independent claims are quite wordy. This makes me think they were probably made to put in specific claim details that weaken the patent by making it narrower.

I am not an expert at reading patents so correct me if I am wrong. From what I understood it only appears to cover encoding accelerated using graphics APIs. If you are not using the texture unit for performing loads and your data is not stored as textures on the GPU (and you dont really need it for CUDA programs), then the patent does not seem to apply.

As a sidenote, the article linked here gets some basic details wrong ("CUDA was launched in 2008", it was launched earlier than that).

Also they have a patent on 2 pass LZ compression. Insane.

I think at a bare minimum any patent that feels like it's just been a race to file a patent shouldn't be granted.

That's why you have a team of patent lawyers spend a week taking the engineer's specs and extending it into a 50+ page document.

This patent includes a bunch of specific optimizations that appear to be the result of some real work on GPU encoding.

Reply on news.ycombinator.com