Slashdot Mirror
NVIDIA To License Its GPU Tech
An anonymous reader writes "Today in a blog post, NVIDIA's General Counsel, David Shannon, announced that the company will begin licensing its GPU cores and patent portfolio to device makers. '[I]t's not practical to build silicon or systems to address every part of the expanding market. Adopting a new business approach will allow us to address the universe of devices.' He cites the 'explosion of Android devices' as one of the prime reasons for this decision. 'This opportunity simply didn't exist several years ago because there was really just one computing device – the PC. But the swirling universe of new computing devices provides new opportunities to license our GPU core or visual computing portfolio.' Shannon points out that NVIDIA did something similar with the CPU core used in the PlayStation 3, which was licensed to Sony. But mobile seems to be the big opportunity now: 'We'll start by licensing the GPU core based on the NVIDIA Kepler architecture, the world's most advanced, most efficient GPU. Its DX11, OpenGL 4.3, and GPGPU capabilities, along with vastly superior performance and efficiency, create a new class of licensable GPU cores. Through our efforts designing Tegra into mobile devices, we've gained valuable experience designing for the smallest power envelopes. As a result, Kepler can operate in a half-watt power envelope, making it scalable from smartphones to supercomputers.'"
We want to transition to an IP company.
Then we only have to employ lawyers and executives, and save ourselves the trouble of all that making stuff.
We want to transition to an IP company. Then we only have to employ lawyers and executives, and save ourselves the trouble of all that making stuff.
Nah, that's not why. They're following in the footsteps of Apple and Sega - license out your key strengths to strategic partners, and you're sure to succeed.
Right?!?
- Nec Impar Pluribus, or so I'm told.
If you're wondering about AMD, they also had a project doing graphics for ARM CPUs, but it was outright sold-off to Qualcomm.
Qualcomm's "Adreno" GPU? The name is an anagram of Radeon.
David Shannon says:
PC sales are declining with the rise of smartphones and tablets.
Uh oh, our traditional PC market is dying.
High-definition screens are proliferating, showing up on most every machine. Android is increasingly pervasive. Yesterday’s PC industry, which produced several hundred million units a year, will soon become a computing-devices industry that produces many billions of units a year. And visual computing is at the epicenter of it all.
But wait! The mobile market is hot hot hot!
For chip-makers like NVIDIA that invent fundamental advances, this disruption provides an opening to expand our business model.
We should go all in on mobile and get some of that delicious moolah.
But it’s not practical to build silicon or systems to address every part of the expanding market. Adopting a new business approach will allow us to address the universe of devices.
How can we like, totally dominate this market?
So, our next step is to license our GPU cores and visual computing patent portfolio to device manufacturers to serve the needs of a large piece of the market.
Lets licence out our IP! You saw how it like, totally worked for ARM, right?
The reality is that we’ve done this in the past. We licensed an earlier GPU core to Sony for the Playstation 3. And we receive more than $250 million a year from Intel as a license fee for our visual computing patents.
We tried it in baby steps, and the money was delicious.
Now, the explosion of Android devices presents an unprecedented opportunity to accelerate this effort.
More money is good.
The ONLY company on this planet with an interest in very high-end desktop class GPU technology for their own use is Intel. No-one else has the need (PowerVR fills the gap for most companies that license GPU designs) or the ability to build such a complex design into their own SoC.
Anyone else with an interest in Nvidia GPU capabilities would opt to buy discrete chips from Nvidia, or one of Nvidia's existing ARM SoC parts.
AMD is currently devastating Nvidia in the high end gaming market. Every one of the 3 new consoles uses AMD/ATI tech for the graphics. EA (the massive games developer) has announced their own games engines will be optimised ONLY on AMD CPU and GPUs (on Xbone, PS4 and PC). Nvidia is falling out of the game.
The x86 space is moving to APUs only. Chips that combine the CPU cluster with the GPU system. Intel's integrated GPU is pure garbage. However, Intel spends more on the R+D for its crap GPU than Nvidia and AMD combined. It would be insanely cheaper for Intel to simply license Nvidia's current and future designs. Doing so would give Intel parts that compete with AMD for the first time ever. Of course, it still wouldn't fix the problem that AMD tech is in the only hardware AAA games developers care about.
Next year AMD completes its project to take desktop x86 parts to full HSA and Huma (as seen in the PS4). Next year Intel begins the process to use this tech (and will be two years behind AMD at best). Both companies are moving to PC motherboards that solder memory and CPU on the board itself. Both are moving to a 256-bit memory interface, although again AMD will have a significant lead here.
Intel wants to copy AMD's GDDR5 memory interface (again, as seen in the PS4) but that requires a lot of tech Intel does not have, and cannot develop in-house (god only knows, they've tried). Nvidia also has massive expertise with GDDR5 memory interfaces, and the on-chip systems to exploit the incredible bandwidth this memory offers.
Everyone should know Intel wanted to buy Nvidia, but would not accept Nvidia's demand to have their people run the combined company. The top of Intel is BRAINDEAD, composed of the useless morons who claimed credit for the 'core' CPU design, when all core was in reality was a return to Pentium 3, after Netburst proved to be a horrible dead-end. This political power grab is responsible for all Intel's current problems, including this biggest disaster in semiconductor history- Larrabee. Intel's FinFET project has crashed twice (Ivybridge was much worse than Sandybridge, despite the shrink, and Haswell is worse again). Intel has no new desktop chips for 2014 as a consequence.
Now we can see it is likely Intel is readying Nvidia based parts for 2015 at the earliest. Intel has used licensed GPU tech before, notably the PowerVR architecture. However, Intel's utter inability to write or support drivers meant the PowerVR based chips wee a disaster for Intel. Intel's biggest problem with its current GPU design is NOT that it is a Larrabee scale failure, but that Intel is actually making headway. So why is this an issue?
Well companies like S3 also made successful headway with their own designs, but this didn't matter because they were way behind the competition at the time. It is NEVER a case of being better than you were before, but a question of being good enough to go up against the market leaders. Intel knows its progress means that internally its GPU team is being patted on the back and given more support, and yet this is a road to nowhere. Intel needs to bite the bullet, give up on its failed GPU projects, and buy in the best designs the market has to offer. Nvidia is this.
Unlike PowerVR, which is largely a take it or leave it design (which is why Intel got nowhere with PowerVR), Nvidia comes with software experts (for the Windows drivers) and chip making experts, to help integrate the Nvidia design with Intel's own CPU cores.
For super-computing type workloads, ARM does not have a CPU fast enough to deliver the Ethernet, Infiniband, SSD, and other communications traffic to keep a Tesla fed with data.
However, Nvidia's long-term strategy must be to sell low-power and high-power ARM chips with GPU accelerators. Within 2 to 3 years, Intel will have a Xeon product that merges the existing 12-core Xeon processors with the 60-core Xeon Phi accelerators. Similarly, AMD will be building equivalent APU units with their mixed x86, ARM and GPU technologies. To be even marginally useful, Nvidia needs something to compete.
Personally, I think AMD stands a decent chance of having the fastest APUs. I think attempting to maintain cache-coherency between massive numbers of cores reduces the performance/watt advantage of the Xeon Phi. Also, if you are going to have heterogenous cores where the CPUs cannot run standard x86 code (like the Xeon Phi), then why not go fully heterogenous to maximize APU performance? Currently, AMD has the fastest merged processing units.
nVidia's graphics drivers include proprietary and patented Microsoft technologies. They cannot open source them, ever. They made their deal with the devil and they have to live with it.
Help stamp out iliturcy.
Wouldn't it be great if they open sourced their fucking Linux drivers instead of playing the little game they play?
Why? What would be so good about that? AMD did it and it didn't do much for them.
Server side is where margins are at. AMD borked a server CPU generation and wound up cleaning house. If the world were different this would not be a recoverable error. Since Intel needs AMD to blunt monopoly supervision, Intel server tech will probably be delayed to give AMD a chance to catch up a little bit. Intel will keep inventing clever new stuff, but stuff it in a closet again as they have done many times before. This isn't a big deal since server tech is so way overpowered from what it needs to be that Intel could probably coast for 6 years before they had to start innovating again. Maybe they'll retask some engineers from servers (and God please, Itanic) to mobile. That would be nice.
Help stamp out iliturcy.
Consoles are focused on lowest possible cost of their hardware, since they sell to consumers at a loss, or at the best a slim profit. They need their suppliers to give them hardware for bottom dollar. That means you don't get much profit per unit.
Now that doesn't mean AMD is getting screwed, I'm sure they are making money per unit sold, but make no mistake: The reason they got the contracts is they could offer the lowest price and that means a thin profit. So 10 million chips sold in the console is less profit than 10 million sold in a desktop or server or the like.
It is not the grand prize of hardware contracts.
On another note I find it hilarious how fanboys relish in the concept of a competitor doing badly, as if we all wouldn't be more screwed if there was a single company. Personally, I like nVidia GPUs, they work better in my experience. However I'm real, real glad AMD is around. Why? Well if they weren't nVidia could, and would, charge more than they already do, and they wouldn't release new tech as fast.
So if you are an AMD fanboy wishing the death of Intel and nVidia, what you are really saying is "Gee I hope AMD will be able to overcharge me for lower end technology when they have nobody to push them!"