Posted: Jon Peddie 11.09.17
After posting a good-bye letter to the people of AMD about his 40-day alleged sabbatical, Raja Koduri popped up as an employee at Intel. You don’t hire someone at that level in one day, Intel doesn’t vet new employees in a week, ask some of the people who work there. Which means the conversations between Koduri and Intel have been going on for some time, and probably what prompted Koduri’s sabbatical.
Item two is the shakeup in Intel and who will be Koduri’s boss— Venkata “Murthy” Renduchintala, the former Co-CEO at Qualcomm, that Intel hired in February this year with a compensation package estimated at more than $25 million. In just a few months, Renduchintala established himself as the number-two person at Intel. CEO Brian Krzanich gave him responsibility for making some of the most important decisions in Intel's $1.2 billion reorganization plan. One of those changes was Renduchintala's creating three-person leadership teams for each of the new products under development, while constantly looking for outside talent to join the company.
In April Renduchintala wrote an internal memo to Intel's senior management that got leaked, that said, "Over the last three months I have conducted numerous project reviews with our execution teams, and there is a clear trend that has emerged in these reviews — a lack of product/customer focus in execution that is creating schedule and competitiveness gaps in our products."
Item three, which may be un-related, but probably isn’t, is the deal Intel struck with AMD to use AMD’s GPU chips in a new multi-chip product for high-performance thin and light notebooks, which we think was driven by a demand from Apple. Koduri used to work at Apple prior to going (back) to AMD.
And so, Intel made the following statement on 8 November, 2017:
"Intel to Expand Strategy to Deliver High-End, Discrete Graphics Solutions"
Intel today announced the appointment of Raja Koduri as Intel chief architect, senior vice president of the newly formed Core and Visual Computing Group, and general manager of a new initiative to drive edge computing solutions. In this position, Koduri will expand Intel's leading position in integrated graphics for the PC market with high-end discrete graphics solutions for a broad range of computing segments.
Billions of users today enjoy computing experiences powered by Intel's leading cores and visual computing IP. Going forward under Koduri's leadership, the company will unify and expand differentiated IP across computing, graphics, media, imaging and machine intelligence capabilities for the client and data center segments, artificial intelligence, and emerging opportunities like edge computing.
Koduri will officially start in his new role at Intel in early December.”
This high-profile move can be read a few ways:
First, it is an open declaration that Intel is once again going to attempt the development of a discrete GPU. The sorting out of Koduri’s NDAs, AMD’s patents, and potential Intel cross licensing will be another matter, but be assured, Intel will get it worked out.
Second, that means the AMD-Intel chip deal is a spot product and the next gen will likely have an Intel discrete GPU in the mix. Intel’s discrete will likely be HBM3, and manufactured in 7 nm, which means the GPU will show up in 2020, or 2021.
The third point is Intel already has a good GPU design, it’s just in the wrong place. Koduri’s job will be to pull it out of the CPU, and add a HBM controller to it. He’ll probably add a few more shaders to it too. All that should be doable in three years, and I wouldn’t be surprised if such a project wasn’t already in progress. In fact, I’d bet it was, and running late, and Apple got impatient and forced Intel into the AMD deal. (ref – Renduchintala: “schedule and competitiveness gaps.”)
The fourth point is Intel’s desire to remove the sting of GPU-compute. Larrabee/Phi didn’t do it, FPGAs didn’t do it, and the Neuron chip won’t either. They will all help, but when HPC/supercomputers are built, the GPU is the preferred accelerator and data-cruncher. That fact will be compounded with the explosion in deep-learning, and AI training systems. I think AI is going to be the biggest thing in computing since the transistor, and Intel is now positioned to have a solution for every segment of it from the minimalistic IoT front end devices to the massive big-data learning systems. It will no longer be IA at intel but AI-IA.
In late August Intel made a quiet statement that the Xeon Phi 7200 Coprocessors AIBs, descendants of Intel's canceled Larrabee GPU project, would be retired.
“Intel continually evaluates the markets for our products in order to provide the best possible solutions to our customer’s challenges. As part of this on-going evaluation process Intel has decided to not offer Intel Xeon Phi 7200 Coprocessor (codenamed Knights Landing Coprocessor) products to the market.
Given the rapid adoption of Intel Xeon Phi 7200 processors, Intel has decided to not deploy the Knights Landing Coprocessor to the general market.
Intel Xeon Phi Processors remain a key element of our solution portfolio for providing customers the most compelling and competitive solutions possible.”
Intel has stopped offering the Xeon Phi co-processor AIBs, but will continue offering Xeon Phi processors—for now.