Nvidia not only invests five billion dollars in Intel, CEO Jensen Huang also opens the doors to a new category of chips for which AMD has enjoyed exclusivity until now.
The enemy of my enemy is… my investor. In a surprising move, Nvidia announced yesterday that it would pump five billion dollars into Intel. Nvidia and Intel are somewhat competitors, but not really.
Nvidia is making waves with the world’s most powerful GPUs, both for workstations and gaming laptops as well as in data centers for AI training and inference. Intel is still the market leader in the CPU segment. The manufacturer’s x86 chips power most PCs worldwide and are still the standard in servers.
Both components are complementary. Nvidia’s GPUs don’t function without a CPU. Often that’s an x86 chip from Intel, although Nvidia also makes combined components with ARM processors. In practice, you often find Intel CPUs and Nvidia GPUs together in one PC or server enclosure.
The APU: a (not so) New Idea
What Nvidia and Intel can achieve together was already clear before Nvidia’s CEO Jensen Huang literally explained it. Together, they can develop components where a CPU chiplet and a GPU chiplet are closely connected with shared memory and work together, all on one chip.
Such a component is called an APU (accelerated processing unit) in the PC world, and although Huang casually claims that this creates a new class of integrated graphics power, APUs have existed for much longer. One company is the master of it: AMD.
AMD is Intel’s only competitor and makes x86 processors that can easily stand alongside or even above Intel’s alternatives today. The manufacturer also makes GPUs and accelerators, thus also being a direct competitor to Nvidia.
In the world’s most powerful supercomputers, you even find AMD Epyc processors and AMD Instinct accelerators more often than Intel CPUs and Nvidia GPUs. AMD has more synergy possibilities between CPU and GPU because they come from the same stable.
The Siren Call of Strix Halo
AMD competes directly with both Nvidia and Intel. The company has been tinkering for years with the combination of CPU and GPU with its own interconnect on one powerful chip. The most recent incarnation of this is the Strix Halo component that we find in, among others, the HP Zbook Ultra G1a 14.
This Strix Halo chip, an AMD Ryzen Max+ Pro 395 for friends, combines Zen 5 compute cores with a Radeon 8060S CPU chiplet and up to 128 GB of memory, glued together with AMD’s own Infinity Fabric into one whole. As a result, the thin Zbook can run AI models locally. Don’t think of small gimmick models, but a full version of ChatGPT.
read also
AMD surprises with Ryzen AI Max Pro: workstation on a chip
However, no other company in the world today can make an x86-compatible component that challenges Strix Halo on its own. Apple, based on the ARM architecture, did develop a successful chip that combines CPU and GPU, thus challenging Intel from a different angle. The fact that Apple computers no longer have x86 and Intel on board has certainly been felt by the processor specialist.
Even in the more affordable segment, the combinations of AMD’s CPU chiplets and GPU chiplets deliver solid performance. And don’t forget the game consoles: the PS5 and Xbox are popular computers, and both are built around an AMD APU.
The Right Ingredients
This brings us back to the marriage between Nvidia and Intel. Look beyond the money to what’s on the table: Intel is the x86 market leader and, despite some setbacks, still builds very powerful processors in recent years. Nvidia’s RTX GPUs are certainly as powerful as AMD’s Radeon GPUs.
Moreover, Nvidia is used to connecting its GPUs with other GPUs and memory. For this, it has its own Nvlink interconnect. Put CPU, GPU, and interconnect in a cooking pot with a pinch of memory and a little freshly ground pepper, and you get a credible challenger to AMD’s Strix Halo and other APUs.
Put CPU, GPU, and interconnect in a cooking pot with a pinch of memory and a little freshly ground pepper, and you get a challenger to AMD’s Strix Halo
That’s exactly what Intel and Nvidia plan to do. The result of this cooking will be a new x86 RTX SOC. It should power a wide range of PCs. Nvidia and Intel are not only targeting the top premium segment but want to widely roll out the power of the RTX-x86 tandem.
Sights Set on AMD
The two parties tell the story as if they’ve reinvented the wheel. The reality is that Intel, besides a much-needed cash injection, also gets the technical support to compete fiercely with AMD in the segment where that company traditionally stands strongest.
Nvidia also doesn’t like competition. The less AMD’s graphics solutions are present in the market, the smaller the chance that the dominance of the CUDA software ecosystem will ever be broken.
Who Still Believes in Arc?
With this collaboration, another competitor for Nvidia practically disappears: Intel itself. Intel has been tinkering with its own Xe architecture for GPUs and accelerators for several years. The long-term goal was to be able to supplement its CPU offering with a GPU, both in the data center and in the PC market. For example, the Intel Arc B580 GPU is a direct alternative to the Nvidia RTX 4060. In the data center, Intel did achieve some successes with Gaudi, where competition with Nvidia was mainly on price.
Intel now claims that the Arc GPUs will survive Nvidia’s investment. The wording is vague, but Intel emphasizes that the collaboration with Nvidia is complementary to Intel’s roadmap with its own GPUs. That’s very hard to believe. There’s little reason to continue tinkering with a mediocre proprietary GPU architecture when a collaboration with the GPU market leader provides access to much more powerful GPU chiplets.
read also
Nvidia and Intel become friends: Joint Chips and Billion-Dollar Investment
Additionally, Intel is slimming down heavily under CEO Lip-Bu Tan. Restructurings follow one another, and everything that doesn’t belong to the core business has to go. Those who believe Intel and think that Xe and Arc have a future in the medium term do so at their own risk. The future of Gaudi also seems uncertain in that respect, although the question is whether Intel really dares to put all its eggs in the Nvidia basket.
Investment with Potential
In any case, Nvidia’s investment in Intel is not just a financial story. The collaboration will lead to very interesting hardware with potential that we’re excited about in the short term.
By merging CPU, GPU, and ideally memory, chips with unique capabilities emerge. Apple demonstrates this with its M-series. In the x86 world, AMD is the only one innovating in this segment. Components like Strix Halo offer an answer to Apple’s M-chips and demonstrate what’s possible with a good implementation. It could very well be that Jensen Huang’s pat on the back is exactly what Lip-Bu Tan needs to turn the Intel ship around.