In a surprising announcement that finds two long-time rivals working together, Nvidia and Intel announced today that the companies would jointly develop multiple new generations of x86 products together. The products include x86 Intel CPUs tightly fused with an Nvidia RTX graphics chiplet for the consumer gaming PC market, named the ‘Intel x86 RTX SOCs.’ Nvidia will also have Intel build custom x86 data center CPUs for its AI products for hyperscale and enterprise customers. Additionally, Nvidia announced that it will buy $5 billion in Intel common stock at $23.28 per share, representing a roughly 5% ownership stake in Intel. (Intel stock is now up 33% in premarket trading.) We spoke with Nvidia representatives to learn more details about the company’s plans.
Nvidia says that the partnership between the two companies is in the very early stages, so the timeline for product releases, along with any product specifications, will be disclosed at a later, unspecified date. (Given the traditionally long lead-times for new processors, it is rational to expect these products will take at least a year, and likely longer, to come to market.)
Nvidia emphasized that the companies are committed to multi-generation roadmaps for the co-developed products, which represents a strong investment in the x86 ecosystem. However, Nvidia tells us it also remains fully committed to its other announced product roadmaps and architectures, including for its Arm-based GB10 Grace Blackwell processors for workstations and the Nvidia Grace CPUs for data centers and the next-gen Vera CPUs. Nvidia says it also remains committed to products on its internal roadmaps that haven’t been publicly disclosed yet, indicating that the new roadmap with Intel will merely be additive to its existing initiatives.
Nvidia hasn’t disclosed whether it will use Intel Foundry to produce any of the products yet. However, while Intel has used TSMC to manufacture some of its recent products, its goal is to bring production of most of its high-performance products back into its own foundries, and some of its products never left. For instance, Intel’s existing Granite Rapids data center processors use the ‘Intel 3’ node, and the upcoming Clearwater Forest Xeons will use Intel’s own 18A process node for compute. This suggests that at least some of the Nvidia-custom x86 silicon, particularly for the data center, could be fabbed on Intel nodes. However, Intel also uses TSMC to fabricate many of its client x86 processors now, so we won’t know for sure until official announcements are made — particularly for the RTX GPU chiplet.
While the two companies have engaged in heated competition in some market segments, Intel and Nvidia have partnered for decades, ensuring interoperability between their hardware and software for products spanning both the client and data center markets. However, these products have long used the PCIe interface to connect Intel CPUs and Nvidia’s GPUs. The new partnership will find tighter integration using the NVLink interface for CPU-to-GPU communication, which affords up to 14 times more bandwidth along with lower latency than PCIe, thus granting the new x86 products access to the highest performance possible when paired with GPUs. Let’s dive into the details we’ve learned so far.
Intel x86 RTX SOCs for the PC gaming market
For the PC market, the Intel x86 RTX SoC chips will come with an x86 CPU chiplet tightly connected with an Nvidia RTX GPU chiplet via the NVLink interface. This type of processor will have both the CPU and GPU units merged into one compact chip package that externally looks much like a standard CPU, rivaling AMD’s competing APU products.
This type of tight integration packs all the gaming prowess into one package without an external discrete GPU, providing power and footprint advantages. As such, these chips will be heavily focused on thin-and-light gaming laptops and small form-factor PCs, much like today’s APUs from AMD. However, it’s possible the new Nvidia/Intel chips could come in multiple flavors and permeate further into the Intel stack over time.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Contact me with news and offers from other Future brands Receive email from us on behalf of our trusted partners or sponsors
Intel has worked on a similar type of chip before with AMD; however, there is at least one significant technical difference between these initiatives. Intel launched its Kaby Lake-G chip in 2017 with an Intel processor fused into the same package with an AMD Radeon GPU chiplet, much the same as the description of the new Nvidia/Intel chips. You can see an image of the Intel/AMD chip below.
... continue reading