Nvidia recently announced a product that could propel the company into the consumer CPU market – the RTX Spark – a supercomputer- grade processor that can run large AI models entirely locally.
This is a 20-core ARM-based Nvidia Grace processor developed with MediaTek, featuring a Blackwell RTX GPU (equivalent to an RTX 5070) with 6144 CUDA cores, and both the CPU and GPU are interconnected via NVLink, with support for up to 128GB of unified memory.
Agentic describes the capacity of an entity—especially an AI system—to act autonomously, make decisions, and use tools to achieve a specific goal without requiring step-by-step human intervention.
I’m not sure about describing this as a new era in computing, but what is true is that currently, when you use AI – Grok, Copilot, Gemini, etc.- to create an image, for example, it’s being created remotely. RTX Spark allows developers to create locally, which is a major step forward. But it’s very expensive, ranging from $ 5,000 to $10,000 depending on the model. Checking on Amazon, I see that over 500 units at $5000 have already been sold!
Enter AMD!
It’s no secret that AMD and Nvidia have been battling it out for decades in the GPU sector, with Nvidia winning hands down. But more recently, they have been racing against each other in AI, with many of us wondering when AMD or Intel were going to compete with RTX Spark. AMD Ryzen Halo takes the RTX Spark head-on with some impressive specifications to run local LLMs:
The big question is, who will win the local AI race?
—
