(Credit: Cole Kan/PCMag/Nvidia/Intel/Narumon Bowonkitwanchai via Getty)
Last week, Intel and Nvidia announced a stunning $5 billion investment deal that signals a new strategic alliance between two of the biggest chip makers on the planet. Longtime frenemies in the microprocessor world, both companies have had their respective CPU and GPU hardware in the same laptop and desktop products for decades. Now, Nvidia has bought a $5 billion stake in Intel, taking ownership of 4% to 5% of the company through common stock. But this isn't merely an investment deal.
The duo just announced that it will co-develop several new products that combine the companies' strengths. Intel will design and manufacture custom x86-based CPUs for data centers and AI infrastructure that can use Nvidia’s NVLink processor connection technology. At the same time, Nvidia will make GeForce RTX GPU chiplets to fit onto Intel system-on-chip (SoC) processors for consumer PCs.
Neither company will say when new products from this partnership will reach the market. But in a press conference on September 18, Nvidia CEO Jensen Huang revealed that Intel and Nvidia teams have already collaborated for a year on planning a shared architecture for these new SoCs. So silicon resulting from this partnership could be here sooner than expected.
Our news coverage of the deal, linked above, gets into more of the agreement details and the business impact of the partnership. However, my thoughts immediately went to consumer PCs and how this new team-up could shake up the world of laptops and desktops. If there's a new line of Intel-Nvidia hybrid chips in the offing, what does that mean? How might specific Nvidia technologies play into new Intel systems? And where will this kind of CPU fit into the familiar categories shoppers already know?
Nvidia’s $5 Billion Intel Investment Is Not Just About the Money: It’s Tech Synergies, Too
Nvidia has several technologies that Intel can benefit from in this new partnership. I’ll break them down one by one.
NVLink: Faster Internal Chip Connections
One mentioned explicitly in the announcement is NVLink, Nvidia's high-bandwidth, low-latency, proprietary interconnect that supports massive amounts of bandwidth for moving data between processor elements. (NVLink, on a bigger scale, is also a key enabling technology in Nvidia's data center strategy.) With current GPU implementations reaching up to 1.8 terabytes (TB) per second (albeit, across 18 100GB-per-second connections), NVLink dwarfs the bandwidth of PCI Express 4.0, which has a maximum aggregate bandwidth of 64GB per second, and PCI Express 5.0, which doubles that. Not knowing how many connections a hybrid chip would actually use, the difference could be pretty significant.
Here, NVLink would be implemented as a connection between the Nvidia RTX graphics chiplet on the SoC and the chiplet with the Intel CPU cores. That sort of stable point-to-point connectivity promises to reduce bottlenecks for data-intensive uses, like graphics workloads, and it could mean faster AI inference performance and multitasking when incorporated into an SoC.
CUDA and Tensor Cores: Supplanting the NPU?
Nvidia GPUs rely on CUDA processing cores, specialized parallel processors that efficiently handle the thousands of simultaneous operations needed for modern graphics rendering. But they can also handle floating-point and integer calculations, making them extremely useful for AI workloads.
Tensor cores comprise the rest of Nvidia’s silicon offering. These are specialized accelerators for the matrix math needed for AI. Where CUDA cores handle the bulk of computations, Tensor cores speed up deep learning and AI tasks, boosting speed for specific functions that need neural network training and inference. These specialized cores also serve as the AI engine behind Nvidia’s DLSS tech, which I’ll get to shortly.
(Credit: Nvidia)An Nvidia GPU chiplet that included CUDA and Tensor core technology would dramatically boost both graphics and AI PC capabilities in a consumer laptop, easily exceeding traditional integrated graphics, and possibly supersede the robust neural processing unit (NPU) hardware altogether that we are seeing in late-model AMD and Intel laptop processors like the Ryzen AI 300 and Core Ultra 200V series.
DLSS: Faster Gaming From Lesser GPUs
DLSS, or deep learning super sampling, is Nvidia's AI-powered image-upscaling tech, and the most recent DLSS 4 adds upgraded frame generation that massively boosts GPU performance. The technology uses AI pattern recognition and prediction to create new, additional frames beyond those rendered by the GPU's rendering cores for faster frame rates at higher resolutions. (For a deep dive into these technologies, see our article GeForce RTX 50 Series Demystified.)
While a chiplet-compatible version of this technology can’t match what a dedicated GPU can produce, the result could still be profound for mainstream systems. Not only would it unlock gaming for a segment of the population that doesn't want to invest in a gaming laptop or pay extra for a dedicated GPU, it could also boost the visual performance of pretty much everything else, particularly creative work.
Tegra Architecture: Carrying Over Learnings From Arm
Tegra is Nvidia's own existing SoC processor, an Arm-based chip that combines CPU, GPU, memory controller, and neural network accelerator into a single integrated chip. Tegra chips more often appear in mobile devices (phones and tablets), console handhelds like the Nintendo Switch, and non-consumer applications like automotive and robotics.
Based on Arm, the actual Tegra technology won't show up in Intel chips in any capacity. However, the design principles and lessons learned from Tegra will undoubtedly play a role in whatever designs Intel and Nvidia make together. Nvidia's experience with modularity, power management, and AI acceleration will definitely be part of what it brings to the table.
A New 'Middle Kind' of Processor: Putting It All Together
We traditionally talk about CPU and GPU technology in one of two ways. You have integrated graphics processors (IGPs), which use GPU cores built directly into the CPU, sharing the same chip. These GPUs work for basic visual tasks, like web browsing or video playback, using the same memory (RAM) allotted to the CPU.
On the other hand, discrete GPUs use dedicated graphics processing hardware, a separate chip with its own processing cores and specialized video RAM, independent of the CPU and system RAM. These specialized chips are substantially faster for gaming, video editing, and other graphically intensive uses. In recent years, as GPU hardware has been tapped to power AI applications, they’ve provided the necessary muscle for running local AI models quickly.
However, a new Intel-meets-Nvidia chip wouldn't really fall under either of these types, and it's not the first of its kind. Apple's M-series chips do something similar, pairing the CPU with graphical compute units in the same package, using shared memory pools and higher-bandwidth interconnects to deliver GPU-grade performance without a discrete graphics chip.
AMD has recently blended CPU and GPU integration in similar fashion, with the AMD Ryzen AI Max+ ("Strix Halo") chips that combine a capable CPU with Radeon graphics and up to 128GB of unified memory. We've seen this capable hybrid approach in all sorts of applications, from a gaming tablet to a small desktop, and even a workstation laptop, all without a traditional discrete GPU.
With this Intel and Nvidia collaboration, we could see these early examples become a mainstream option quickly. What will we call this new class of SoC's beefed-up graphics? "Hybrid graphics"? "Chiplet graphics"? "SoC graphics"? (Seriously, I don't actually know yet. If you have any great ideas for terminology, drop them in the comments below. We might just adopt your idea!)
Intel-Nvidia Silicon Could Uplift These Kinds of PCs
Here's where the train leaves Speculation Station and arrives in Imagination Land. While I can make informed logical leaps about how Intel might leverage Nvidia technology and how a new hybrid chip aligns with other emerging trends in PC technology, we haven't seen any word about where these new chips will fit into the laptop and desktop landscape.
Dedicated GPU hardware has traditionally been the domain of gaming machines and professional workstations. But what happens when a new option is more potent than an IGP, but not as beefy as a discrete GPU? I have ideas.
1. Boosted AI Laptops, Beefed-Up Mini Desktops
These new chips look like they could serve as a massive step up from the NPUs used in current AI PCs. While I have no hard numbers for future hybrid SoCs, we know that Intel's NPU hardware can generate 40 to 50 trillion operations per second (TOPS). Meanwhile, Nvidia's entry-level mobile RTX 50-series GPUs produce 440 TOPS.
(Credit: Joseph Maldonado)While that will be a significant boon for AI PC laptops, I also see a place for mini PCs and small form factor (SFF) desktops that can deliver substantially more powerful AI capabilities than Intel's NPU-equipped chips. While tiny, a mini PC provides enough room for more cooling hardware than what even a top laptop can fit.
That positions compact desktop designs to take advantage of this new hardware in ways that laptops might not be able to match without mimicking the thicker, heavier designs of gaming laptops, which would be better served by a "true" discrete GPU, anyway.
2. More Affordable Content-Creation Laptops
The emerging category of laptops that leverage low-end GPU hardware for GPU-accelerated content creation, rather than gaming, could use a lift. These new hybrid SoC processors could make those media-focused machines much more affordable.
(Credit: Joseph Maldonado)For tasks like video streaming, video and image editing, graphic design, and other functions like 3D rendering, Intel-Nvidia combo chips would open up a lot of capability for new designs that combine the slimness of ultraportables with the capability of entry-grade workstations and gaming machines.
3. Slimmer, Longer-Lasting Midrange Gaming Machines
Finally, gaming is the most obvious application of Nvidia chiplet technology on Intel CPUs. These new chips will provide efficiency and power for gaming performance in a thinner, lighter laptop, with longer battery life for day-to-day use.
Thin-and-light gaming laptops tend to be even more costly than some powerful gaming rigs, since the design constraints are more severe. A slimmer chassis challenges your cooling solution; the current CPU/GPU model uses slower PCIe interconnects, and these systems require separate memory for graphics. The additional hardware means that the thinner gaming laptops still have much higher power draw, making on-battery use less impressive and battery life dramatically shorter than the 20-plus hours that some midrange consumer systems can now deliver.
(Credit: Joseph Maldonado)But putting more capable graphics into a midrange consumer laptop? That could mean gaming capability in something closer to an ultraportable, as opposed to the chunky designs that dominate our current budget gaming laptop picks, while still presenting a more affordable option than current GPU-equipped laptops.
I’m just as excited as you to discover how these new chips will look and perform once they come to market. What kind of premium Intel and Nvidia will demand for this level of computational sophistication is anyone’s guess right now, but until we see hard performance numbers, I wouldn’t pay more than what a MacBook costs for the privilege.