Nvidia RTX Spark 评测：Arm CPU 加 Blackwell GPU 的本地 AI 平台为何令人失望

I was so excited for Nvidia RTX Spark and its unique combination of Arm CPUs and Blackwell GPUs, but it is crushingly disappointing.

❌ RTX Spark laptops are heavy, hot, loud, and wrong for AI agents

❌ RAM scalability and bandwidth fall short of Apple Silicon

❌ Manually configured shared memory, not unified memory

❌ Windows 11 offers no unique RTX Spark functionality or platform capabilities

❌ Local AI is complex, slow, and unlikely to meet anyone's needs

John F. Kennedy once observed that "Success has many fathers, but failure is an orphan." But with the Nvidia RTX Spark platform, the failures are many and the blame can be assigned to multiple parties. So many things had to fail all at once for this platform to land the way it has.

The only parties involved that are blameless in this fiasco are the PC makers. I've not used all the initial RTX Spark laptops, but I have used three--the ASUS ProArt P16, the HP OmniBook Ultra 16, and the Lenovo Pro 9n--for over two weeks. Each is a premium, handsome, and well-made device that communicates its makers' design expertise. ASUS, HP, and Lenovo, at least, have upheld their ends of the bargain.

The problems with RTX Spark are many, but they can be broadly organized into two top-level issues: Form factor and readiness. And in both cases, the PC makers can only do what they always do and follow the industry titans on which they are reliant. In this case, Microsoft, which supplies the OS their customers demand, and Nvidia, which makes the RTX Spark chipset that both Microsoft and Nvidia tell them their customers will want.

So let's take a look at both.

Form factor

Compared to conventional PC chipsets, which these days consist of a system on a chip (SoC) with microprocessor, graphics, and NPU chips, RTX Spark is unusual, an outlier. I discuss the architecture below, but the simplest way to describe how and why it's different from traditional AMD, Intel, and Qualcomm designs for PCs is that RTX Spark is engineered specifically for local AI.

AMD, Intel, and Qualcomm make families of PC chipsets. AMD and Intel have offerings that specifically target laptops and desktop PCs, and Qualcomm has offerings that specifically target laptops (and, semi-coincidentally, NUC-type mini-form factor PCs that are basically laptops without screens). Nvidia, in contrast, makes a single chipset, the N1X. That there are two nearly identical models--one with an 18-core CPU and a 5120-core CPU and the other with a 20-core CPU and a 6144-core GPU--is mostly a coincidence or byproduct of the chip manufacturing process, not a strategy.

That said, while there are those two N1X models for laptops, there's also an N1X variant--again, with a 20-core CPU and a 6144-core GPU--for desktop PCs that only varies in the amount of heat--as expressed by TDP, or Thermal Power Design--it can expend in normal operation. Where the laptop...