Rumor has it that NVIDIA’s next-generation “AI PC” SOC will feature Arm-based Blackhawk CPUs, Blackwell RTX GPUs, and LPDDR6 memory.
NVIDIA’s next generation SOC could shake things up in the AI PC segment if these rumors are true, Blackhawk Arm CPU cores, Blackwell RTX GPU and LPDDR6 memory all on board the same package
Recently we’ve seen several rumors and reports that NVIDIA may be planning an entry into the client PC space with its own SOC. There are reports that these SOCs will also open the door to new consoles, primarily handhelds at first. Yesterday, CEOs from both NVIDIA and Dell teased this venture and it looks like we have a new rumor that could give us an early hint at what these SOCs would look like.
https://x.com/XpeaGPU/status/1793469879334687122
The details come from @XpeaGPU who we quoted in the previous article talking about a potential handheld and console SOC being designed by NVIDIA based on the same architecture. It seems like NVIDIA wants to integrate several next-generation IPs into one package, which seems like the way to go when it comes to modern PC designs. We’ve seen it with Apple and now with Intel’s Lunar Lake, which offers multiple IPs on the same chip with internal memory and it makes for a very efficient and powerful solution with a small footprint.
https://x.com/Kepler_L2/status/1793523940968820899
For its next-generation SOC, NVIDIA is said to be using a 3nm process node, reportedly Intel or TSMC, and an advanced packaging solution to bring together Arm-based Cortex X5 BlackHawk CPU cores and the Blackwell RTX GPU cores. It’s likely that NVIDIA would use BlackHawk’s CPU cores, as they are reportedly working with MediaTek to create these SOCs.
MediaTek would use the same Arm architecture for the upcoming Dimensity 9400 chip and would tackle Apple A17 Pro and Qualcomm’s Oryon architecture.
The use of Blackwell RTX GPUs isn’t that far-fetched either, as MediaTek previously confirmed in a press release that they were going to use the latest RTX & AI GPU IP within its Dimensity SOCs, so perhaps we’ll get a custom version of that further designed and optimized by NVIDIA . The NVIDIA “AI PC” SOCs are also said to include LPDDR6 memory as an on-package design. LPDDR6 will offer speeds of 10,667 MT/s from scratch and reach 14,400 MT/s over its life cycle, as recently revealed.
Another clue to this development comes directly from NVIDIA’s CEO, Jensen Huang, who states that Microsoft’s recent introduction of Copilot+ PCs has paved the way for opportunities within the customer segment. Jensen says Copilot+ PCs support ARM beautifully and Qualcomm Snapdragon
As you saw yesterday with the build, which I thought was really excellent, today Satya announced the next generation of PCs, Copilot+ PC, which runs fantastic on NVIDIA’s RTX GPUs that come in laptops. But it also supports ARM beautifully. And so it opens up possibilities for system innovation, even for PCs.
Jensen Huang – NVIDIA CEO (Q1 2024 earnings call)
Again, it’s too early to say what the final design and specs for NVIDIA’s “AI PC” SOC for client PCs would look like, but the company has industry-proven experience integrating high-end GPU architectures alongside data center ARM CPUs (Grace). we look forward to seeing NVIDIA bring the same combo to the client side.