Wednesday, November 8th 2017
Intel Hires Raja Koduri, to Develop Discrete GPUs, This Time for Real
Intel hired Raja Koduri, who resigned as head of AMD's Radeon Technologies Group (RTG), earlier this week. Koduri has been made Senior Vice President and Chief Architect of Intel's future discrete GPUs. That's right, Intel has renewed its dreams to power high-end graphics cards that compete with AMD and NVIDIA. Intel's last attempt at a discrete GPU was "Larrabee," which evolved into a super-scalar multi-core processor for HPC applications under the Xeon Phi line.
This development heralds two major theories. One, that Intel's collaboration with AMD RTG on graphics IP could only go further from here, and what is a multi-chip module of Intel and AMD IP now, could in the future become a true heterogeneous die of Intel's and AMD's IP. Two, that the consolidation of AMD's graphics assets and IP into a monolithic entity as RTG, could make it easier to sell it lock, stock, and barrel, possibly to Intel.Intel places great faith in Koduri's ability to either develop a major product line from scratch, or to effect tectonic shifts in the industry, such as Apple's transition to the x86 architecture for its Mac product line.
Intel could have one of three approaches to build a discrete GPU from scratch. The first and most obvious one would be to scale up its current gen 9.5 architecture. The trouble is, that Intel's SIMD parallelism is more transistor-heavy than even NVIDIA. It takes roughly 400-500 million transistors (coarse estimate) on the "Kaby Lake" die to build a GPU with 24 execution units. With 10 billion transistors, we're looking at around 480 execution units, if not more.
The second approach would be to build a new graphics architecture from scratch. Something like this, even with Intel's deep R&D pockets, could take a team led by Koduri 3-4 years. The resulting architecture has to relevant to the market of the time, or end up missing the bus like Larrabee. The third approach would be to either license or acquire GPU IP from AMD. Koduri has the reputation of a tech business strategist as much as an IP guru to effect such a change.
These are strange times in the vale, as silicon giants Broadcom and Qualcomm look to coalesce into the world's third largest chipmaker, and Marvell with Cavium follow on. Stranger things are currently happening between past industry rivals, than the possibility of Intel acquiring RTG from AMD in exchange for cash, and allowing AMD's merger with another chipmaker without affecting its x86 license. This is just a really audacious theory.
This development heralds two major theories. One, that Intel's collaboration with AMD RTG on graphics IP could only go further from here, and what is a multi-chip module of Intel and AMD IP now, could in the future become a true heterogeneous die of Intel's and AMD's IP. Two, that the consolidation of AMD's graphics assets and IP into a monolithic entity as RTG, could make it easier to sell it lock, stock, and barrel, possibly to Intel.Intel places great faith in Koduri's ability to either develop a major product line from scratch, or to effect tectonic shifts in the industry, such as Apple's transition to the x86 architecture for its Mac product line.
Intel could have one of three approaches to build a discrete GPU from scratch. The first and most obvious one would be to scale up its current gen 9.5 architecture. The trouble is, that Intel's SIMD parallelism is more transistor-heavy than even NVIDIA. It takes roughly 400-500 million transistors (coarse estimate) on the "Kaby Lake" die to build a GPU with 24 execution units. With 10 billion transistors, we're looking at around 480 execution units, if not more.
The second approach would be to build a new graphics architecture from scratch. Something like this, even with Intel's deep R&D pockets, could take a team led by Koduri 3-4 years. The resulting architecture has to relevant to the market of the time, or end up missing the bus like Larrabee. The third approach would be to either license or acquire GPU IP from AMD. Koduri has the reputation of a tech business strategist as much as an IP guru to effect such a change.
These are strange times in the vale, as silicon giants Broadcom and Qualcomm look to coalesce into the world's third largest chipmaker, and Marvell with Cavium follow on. Stranger things are currently happening between past industry rivals, than the possibility of Intel acquiring RTG from AMD in exchange for cash, and allowing AMD's merger with another chipmaker without affecting its x86 license. This is just a really audacious theory.
71 Comments on Intel Hires Raja Koduri, to Develop Discrete GPUs, This Time for Real
The prognostications and speculations just keep getting wilder and wilder. It's like a contest to see who has the most fertile imagination. Intel buy RTG outright from AMD ? Hey btarunr pass that Waterpipe my way !
BTW something else for people to consider. The state of the art in Silicon Semiconductors will be pretty close to the end in 3 or 4 years. There will be few if any node shrinks to put extra transistors. GloFo will be at 7nm (maybe 5nm) and Intel will be at 5nm. By 2025 Samsung will be at 4nm and TSMC at 3nm. Who says AMD is out of the GPU Game ? Vega and Navi are almost finished and there are plenty of other people at RTG capable of finishing the job.
It is quite typical to see a non-compete.. I am very surprised he didn't have one...
Turns out... its california... ;)
IF he even stays that long at Intel...
Working for a company, all IP belongs to the company. The same goes in the academic sector except you just get to have your name on the projects (usually research) in a clearer fashion than in small print or "thanks for so-and-so for their part". If the IP is created/made/whatever during the course of employment for an employer, it belongs to the employer. Else (being that IP is created/made/whatever other than in the course of employment), it is owned by the employee. That is the general rule-of-thumb.
You NEVER share your best ideas with your coworkers, bosses/supervisors, or project/research/academic mentors. Trust me, been there and done that. Some people are scum and I'll leave it at that. Sorry, I got semi-triggered.
Plus, it ain't going to stop one company from acquiring IP of another through corporate espionage, borderline sniping of employees from other companies, or just straight-up theft disguised in such a way as to hide the source of the IP to begin with (IP laundering, in-a-way; though, I have most likely used that term incorrectly from a legal sense).
See post 38
(merge your double posts. ;))
EDIT: Or thank me, and leave your double posts... :p