Intel reasoned it could do what AMD and NVIDIA did by using a single, scalable GPU core (Xe) design and service multiple market segments such as gaming, notebooks (Xe-HPG), data center, and AI acceleration (Xe-HPC)—and they can, and are. But, those segments in Intel represent different fiefdoms with different ambitions, executive bonuses, and resource allocations.
All GPU market segments had been placed under Raja Koduri’s leadership, and he pulled in Intel’s one-size-fits-all API—oneAPI. All of that can be accomplished at the device level, but it’s simply too much for any one person to manage three to five business centers that also involve companion CPU and network processors.
Graphics Division Gets Divided
In the second half of December 2022, Intel announced it was splitting up its graphics group to separately address the gaming and data-center markets by placing it under two other business units. Koduri, who was the executive vice president of the AXG business unit, will go back to his role as Intel’s chief architect for GPUs. A setup not unlike that of AMD and NVIDIA.
The company said in a statement that those who reported to Koduri will move to either Intel’s PC or server chip business units. “Discrete graphics and accelerated computing are critical growth engines for Intel. With our flagship products now in production, we are evolving our structure to accelerate and scale their impact and drive go-to-market strategies with a unified voice to customers. This includes our consumer graphics teams joining our client computing group, and our accelerated computing teams joining our data-center and AI group. In addition, Raja Koduri will return to the Intel chief architect role to focus on our growing efforts across CPU, GPU, and AI, and accelerating high-priority technical programs.”
Koduri’s business unit operated independently and labored to get its products to market on a schedule it had announced. It now has GPUs for consumer PCs and for accelerator chips used by data centers and supercomputers, as well as the Flex series for intelligent visual cloud/edge computing.
Intel said it remains fully committed to its existing roadmap of Arc consumer discrete GPUs, and that the recently launched Alchemist series will be joining by its second-generation Battlemage and third-gen Celestial gaming GPUs as planned.
Koduri will work on high-performance technical programs and lead the integration of GPU, CPU, and AI architectures that will show up in products like Falcon Shores and Intel’s zettascale ambitions.
Koduri joined Intel in 2017 after having worked at AMD, Apple, and ATI. He took on the position of leader of AXG last year. Just before the announcement, Koduri posted on Twitter that he was recovering from an unplanned back surgery in India and would remain there for a month before returning to work.
Reorganization
Intel’s new organizational structure of GPUs puts the graphics business unit under Intel’s Client Computing Group (CCG), headed by Lisa Pearce, who has led the GPU software and driver divisions. Pearce reports to Michelle Johnston Holthaus, executive vice president of CCG since January 2022. Intel’s CCG handles consumer computing platforms around the company’s CPU products.
The data center and supercomputing GPUs such as the Ponte Vecchio and Rialto Bridge devices, as well as the GPU SoC and IP design teams, will be managed by the Datacenter and AI (DCAI) business unit. Jeff McVeigh, the VP and GM of the Super Compute Group, will be the interim leader of this AXG team while Intel looks for a permanent leader. McVeigh will report to Sandra Rivera, executive vice president and general manager of the DCAI Group.
Speculation rose earlier this year as to whether Intel, which was cutting costs and selling off marginal business, would continue with its foray into the GPU market; its third or fourth attempt. At the time, CEO Pat Gelsinger assured investors, employees, and partners that the company was committed to the long haul and saw it as a 10-year investment to establish a strong position in the market (see figure).
Gelsinger’s reassurances were not helped by Intel’s marketing department issuing announcements without consulting the product group on the actual status. As a result, several explanation videos were produced and various visits and phone calls were made to journalists and analysts. No one who has been in the GPU business for more than five years doubted Intel could do it. The company has been shipping respectable integrated GPUs for almost 20 years.
However, when the first generation of AIBs arrived, the Arc A750 and A770, there was some misinterpretation outside the company with the marketing messaging that the AIBs were more than they actually were. What they are, however, are damn fine mid-range products with good to great price-performance characteristics—nothing to be ashamed of.
Intel says it's getting the Battlemage products finished and prepared for release. Considering the missteps in announcements and pronouncements of the Alchemist launches, hopefully Intel has learned some lessons and the Arc Battlemage (B-series) release in 2023 will be better coordinated.
Also, in Q1 of this year, Intel reorganized its financial reporting into six business units. AXG will report its revenue as one of the six Intel business units as it has, at least until the January Q4’22 reporting.
What Do We Think?
Intel is here to stay. The new divisional organization makes sense and is logical. The risk is the GPU products will be in competition with the CPU products for resources, marketing airtime, and Intel’s senior managements’ enlightenment about goals and bonuses.
Based on Intel’s previous parochial propensity toward CPUs over GPUs, and even with Gelsinger’s sympathetic attitude toward GPUs, we don’t see the split GPU teams doing very well under the new organizations. Intel has over 50 years of hating GPUs and a mantra of “If it isn’t x86, it’s just wasting silicon.” That’s going to be tough to overcome, and now there’s no fortifications around the GPU. And with Koduri out for a month, who knows what he’ll find when he returns.