This week started with a milestone announcement of Japanese SoftBank moving in to acquire ARM Holdings for a record $32 billion. Many analysts lamented that the takeover was a result of Brexit, even though everyone in their right minds should know that negotiations of this magnitude take years and typically are made in complete silence for 6-18 months (due dilligence and all that jazz).
However, this might not be the only acquisition in the modest pool of British semiconductor players. As we reported on multiple occasions, Intel is reorganizing, i.e. restructuring the company. As a part of that reorganization, the company performed no less than three rounds of layoffs, with over 12,000 people being removed from the company. According to conversations with the recently departed and people still with the company, no department got as culled as the Intel’s mobile processor and Graphics department (IGP), with Intel focusing the culling to its aging staff – as reported on Oregon Live, if you were above 40 years of age, your chances of being laid off were 2.5 times higher than if you were under 40, a trend also known as ageism.
This cull was part of Brian Krzanich’s drive to build talent inside the company by acquisition. Expanding the licensing agreements isn’t something which Intel would like to do, but the company does not want to bleed out significant amount of cash having paid through the nose for Altera. Paying $16.7 billion for a company that doesn’t break $2 billion in annual revenue (yet alone net income) means a long pay back cycle. However, the product cycle does not stop, no matter how you slow it down.
Thus, 2018/19 10nm Ice Lake processors do not have a GPU at the moment. They can use the Cannonlake design, but the question is will that be enough as the graphics now occupy more than 50% of processor die, making Intel CPUs… APUs (Accelerated Processing Unit). Or HPUs (Hybrid Processing Unit). The picture above shows Skylake die and the percentage off it dedicated for the GPU. If you look at picture below, which shows same Skylake-R processor with embeddd 128MB of memory, that percentage changes to over 60% of the total on-package die dedicated to accelerating graphics.
In 2017, the cross-licensing agreement between Nvidia and Intel will expire. Contrary to the popular belief, Intel has no other choice but to continue the cross-licensing agreement with Nvidia, since Intel does not use GeForce hardware in its line-up, but was rather found to infringe on Nvidia’s patents. This was solved with a $1.5 billion settlement which instituted in $50 million-per-quarter payments which will end on January 31st, 2018.
Intel’s cross-licensing agreement with AMD is safe for another four years (April 1, 2010 to March 31st, 2020), but that deal does not include integration of Radeon or FirePro IP.
The NVIDIA – Intel Deal
First Intel – Nvidia deal happened in 2000, when Intel gave the best offer to Microsoft and won the deal for the first Xbox console. Nvidia made its AMD-based nForce chipset to run on Intel, and both companies signed an agreement. In 2004, both companies went back to the negotiating table as Intel wanted to integrate Nvidia’s IP inside Larrabee and future integrated processor graphics, while Nvidia wanted to build chipsets for Intel processors. We know the rest of the story – Intel walked out after nForce started to sell like hotcakes (claiming QPI bus was not the part of the agreement), Nvidia went to the court and ultimately settled for a nice cash prize. While we do not have access to the current agreement, multiple conversations with Nvidia executives claimed that Intel cannot integrate GeForce into its lineup – without a new contract which would probably yield $10-15 fee per every “Intel CPU with GeForce” made.
Over the next 12 months, the companies will have to decide what to do – extend the original 2004/2011 deal or expand the deal to include licensing the GeForce (Core) and potentially Quadro IP (Xeon E5).
The AMD – Intel Deal
AMD and Intel have a cross-licensing deal which was originally created by IBM in order to allow for second-source of x86 processors. First AMD processors were indeed clones of Intel’s own 286 and 386 architecture, but that started to change as AMD ultimately ended in the lead after bringing Jim Keller, Fred Weber & Dirk Meyer – AMD K7 was the answer to the expiry of cross-licensing deal in 2000, and the K7 ultimately killed off the CISC architecture, which was the base for Intel’s x86. Today, both companies make x86 RISC processors, and Intel was even forced to adopt AMD’s x86-64, 64-bit architecture which is now present in almost every processor Intel makes.
Back in August 2005, AMD moved to acquire ATI after a fallout between Hector J. Ruiz and Jen-Hsun Huang over who would lead the new AMD, a result of SNAP (Strategic NVIDIA AMD Partnership) becoming a takeover bid for Nvidia. You can read more in my old analysis on Tom’s Hardware.
More recently, AMD spun off its engineering teams under the Radeon Technologies Group, which is lead by Raja Koduri. By focusing and bringing new and returning the old talents into the fold, RTG is now a non-incorporated company inside AMD holding that is building new products for AMD, Microsoft, Nintendo, Samsung, Sony and potentially, for Intel. That disclosure was a part of big blow out between RTG and HardOCP, or to be more precise – between Raja Koduri and Kyle Bennett. If we disregard the inflammatory tone of the article, there is a momentum between AMD/RTG and Intel at the moment.
In real world, chances that RTG would license anything but baseline Radeon and FirePro at a premium price – are close to zero. That is, unless ZEN-based products tank on the market – which is something that won’t happen (according to the initial leaked performance figures).
Acquiring Imagination Technologies
This brings us to the most probable solution for Intel’s conundrum. Imagination Technologies is a company close to Intel, and the Santa Clara giant already owned a 5% stake in the company. Imagination Technologies tried to sell themselves to Apple, who owns 8.5% of the company, while Tsinghua Unigroup International owns another 3%. Imagination is listed on London Stock Exchange, and carries a modest 595 million pound market cap i.e. under 800 million US dollars.
Intel already has an cross-licensing agreement with Imagination, and more importantly, uses Imagination PowerVR-based GPUs in some of its processors (read: low-power Atom and smartphone chips). Most importantly, besides its graphics IP, Imagination Technologies also owns Caustic Graphics, real-time ray-tracing technology firm with their own silicon, as well as MIPS processor architecture.
Acquiring the company could happen for approximately $1.0-1.5 billion range, which is a fairly cheap acquisition, which more importantly would give Intel access to Apple, who uses PowerVR SGX graphics in all of its products. There are also several silicon vendors from Far East who exclusively use PowerVR SGX and let’s not forget Apple’s largest competitor, Samsung – who switches between PowerVR and Mali (owned by ARM) GPUs.
The question is – who will power the upcoming Intel processors once the time Cannonlake / Ice Lake passes. And unless your name is Brian Krzanich, your guess is as good as mine.