Nvidia unveiled its own RTX Spark laptop CPUs at Computex, with considerable support from PC manufacturers and Microsoft, who clearly expect Nvidia processors to capture a significant share of the PC market. Whether that’s achievable for the first generation, which has been delayed by more than a year, remains to be seen. Nvidia, however, already has a full roadmap of successors in development and revealed it at Computex.
Nvidia plans to release the first generation this year, 2026, (there were some leaks suggesting that it’s been delayed at least a whole year, nevertheless this is the officially announced release date, which is fall this year) and named it Grace Blackwell Spark. It’s a processor also known as GB10.
- Read more: Nvidia has a CPU for Windows desktops and laptops: RTX Spark
- Read more: GB10 or N1X? Nvidia’s Arm Superchip Packs RTX 5070–class GPU
Nvidia described its CPU component as featuring “20 Grace cores,” which is somewhat peculiar terminology. The Grace designation appears to be primarily motivated by brand building needs. The Grace processors Nvidia shipsfor servers are entirely unrelated and use significantly older Arm-licensed cores—Neoverse V2, a server derivative of Cortex-X3. In contrast, the first-generation RTX Spark processors use ten Arm Cortex-X925 cores (two generations newer than Cortex-X3) and ten Cortex-A725 cores, again licensed from Arm by Nvidia—or more precisely, by MediaTek, which developed the CPU portion of the SoC. The GPU does genuinely use the Blackwell architecture.
- Read more: ARM introduces Cortex-X925 „Black Hawk“: Core expected to outperform Apple
- Read more: Blackwell: GeForce RTX 5000 architecture and innovations [Analysis]
Second generation
According to Nvidia, the next generation is called Vera Rubin Spark and is scheduled to arrive in 2028. It could therefore compete against products such as AMD’s Zen 7. Earlier reports suggesting that the second generation could launch just one year after the first, during Q4 2027, have therefore not been confirmed.
It remains unclear whether the processor’s name means it will feature Nvidia-developed Olympus cores, which the company introduced with its Vera server processor. We do not know whether this processor will already be a purely Nvidia-developed product or whether it will still be co-developed with MediaTek—in which case the CPU cores could continue to be licensed from Arm.
What is likely true is that the GPU will already be based on the Rubin architecture. It is also expected to use LPDDR6 memory, but that is the last detail Nvidia has officially confirmed. The manufacturing process, core counts and GPU specifications remain unknown.
Third generation
It appears Nvidia is planning two-year intervals between products. According to the company, the third generation will be called Rosa Feynman Spark and is scheduled for 2030. Rosa refers to the CPU portion, and the name is once again shared with the company’s server processors, which are also expected to carry the Rosa name in 2030. However, it is unclear whether this indicates an identical architecture, and we do not know whether it will be a purely Nvidia-developed product or one created in cooperation with MediaTek.
By contrast, for the GPU portion it can once again be assumed that it will be based on the same architecture (Feynman) as the company’s other GPUs. LPDDR6 memory will likely still be used, although it cannot be ruled out that an enhanced version such as LPDDR6X may emerge by then and be used. Nvidia has not disclosed any additional details.
Nvidia to also offer more affordable CPU models
As you may notice, the roadmap shown at Computex depicts two Grace Blackwell Spark chips. The upper one appears to be the processor Nvidia has just unveiled—the version with 20 CPU cores, a 256-bit memory interface and 6144 GPU shaders. This is indicated by the theoretical AI performance listed as 1 PFLOPS, which Nvidia claims for the fully enabled RTX Spark configuration.
However, below it there is a depiction of a different chip—apparently smaller and less expensive—also labeled Grace Blackwell Spark. This suggests Nvidia plans to introduce a smaller, lower-cost version aimed at more affordable laptops. Earlier reports stated that Nvidia was preparing N1X and N1 processors, with the 20-core RTX Spark likely corresponding to the N1X.
The smaller version is therefore likely to be the non-X “vanilla” N1. According to leaked information, it could feature a 12-core configuration (4 + 8 cores, although it is not clear which number corresponds to the big cores and which to the little ones). The GPU is reportedly expected to have 2560 shaders (20 SM units)—which matches Nvidia’s official specification indicating theoretical AI performance of just 40% of the higher-end model (0.4 PFLOPS).
This processor will therefore likely use a 128-bit memory interface. According to leaked information, the chip could provide eight PCIe 5.0 lanes for up to two SSDs and three PCIe 4.0 lanes for additional peripherals. TDP was reportedly configurable between 18–45 W, although these figures originate from information dating back to 2024 and power consumption characteristics may have changed since then.
Laptops equipped with the 20-core “N1X” version of RTX Spark could be very expensive, partly due to their larger memory capacities. Reports have suggested pricing could start at 2,899 USD. By contrast, the non-X version with the smaller chip could reportedly start at 1,799 USD.
According to the roadmap, these smaller variants of Nvidia’s high-performance flagship processors are also planned for subsequent generations—the Vera Rubin and Rosa Feynman processors. This likely gives Nvidia a better chance of securing a meaningful share of the Windows PC market with these products than if the company were to sell only the much more expensive higher-end versions.
Sources: Nvidia, VideoCardz
English translation and edit by Jozef Dudáš
⠀
