The all-new RTX Spark processor is inarguably the star of NVIDIA's ongoing GTC Taipei event, with the GPU giant managing to bring together ARM, MediaTek, Microsoft, and a host of hardware OEMs in a consequential collaboration that is essentially a shot across the bow of Intel, AMD, and Qualcomm.
Even so, on a granular level, NVIDIA appears to have gone on a shopping spree for its RTX Spark, essentially ripping apart the Dimensity 9400 and Dimensity 8500 chips for CPU-related architectural inspiration.
The CPU of NVIDIA's RTX Spark is essentially a mish-mash of MediaTek's Dimensity 9400 and Dimensity 8500 chips
For the benefit of those who might not be aware, NVIDIA's all-new RTX Spark platform leverages TSMC's 3nm node technology and sports:
- A 20-core Grace CPU.
- A Blackwell RTX GPU, with 6,144 CUDA cores, that delivers up to 1 PFLOP FP4 AI performance.
- Up to 128GB of LPDDR5X-based unified memory
- Around 600 GB/s NVLink-C2C bandwidth between CPU and GPU.
- Support for full NVIDIA software stack, including CUDA, TensorRT, DLSS, Reflex, G-SYNC, and RTX ray tracing.
It is the RTX Spark's CPU, however, that interests us most for the purpose of this post. NVIDIA says it developed the CPU in collaboration with MediaTek. What it should have said instead is that it simply copied a part of MediaTek's architecture for its Dimensity 9400 and Dimensity 8500 mobile chips.
For instance, the Dimensity 9400 chip uses ARM's Cortex-X925 as the sole prime core (ARM has since abandoned the Cortex branding for its cores). And voilà, the RTX Spark uses 10x Cortex-X925 cores.
Also, the Dimensity 8500 chip uses 8x Cortex-A725 cores. And, as such, the RTX Spark's Grace CPU uses 10x of these cores, rounding out its 20-core number.
While this pairing does not in any way take away from NVIDIA's overall vision, we find it interesting that it is premised on the humble mobile chips, and those too of the older-gen variety (Dimensity 9400).
Follow Wccftech on Google to get more of our news coverage in your feeds.
