Do you know how we were literally talking about Nvidia's Gpus significantly outperforming NPU's AI performance on upcoming Cpus (that's enough well, whaddya knows, Nvidia also found it).

According to the Chinese site Benchlife (via Tom's Hardware), Nvidia is making a big AI processing chop on Gpus, dismissing AI accelerated Npu for upcoming AMD, Intel, Apple and Qualcomm chips as "basic".

More specifically, Nvidia classifies AI performance in the range of 10〜45TOPS as "Basic AI", with 100TOPS〜high AI performance as "Basic AI". We characterized 1300+ TOPS as "Premium AI". The Npu of AMD and Intel's current chips has achieved a top of 10-16. When updated later this year, it will increase to the top of 45. 

Of course, it's a coincidence that Nvidia's GPUs fall into more than 100 top categories of the latter. As pointed out earlier today, if you compare Apu with the upcoming laptop Cpu that will deliver 45 top performance from a dedicated AI Npu or neural processing unit, even the RTX4060, Nvidia's entry-level laptop GPU, is claimed to be 194 top. RTX4090 is suitable for 686 top. And what Nvidia is referring to in the figure of 1,300+ TOPS is the desktop RTX4090. All these important points mean that Microsoft's new "AI PC" definition requires 45 tops, not just the old 45 tops, but 45 tops from dedicated NPU, which means Nvidia's GPUs won't qualify, no matter how high the top of those NPU is. 

Nvidia offers a list of AI-related benefits whose "premium" Gpus are better than those terrible "basic" Npu. Nvidia, for example, thinks it can't do AI video generation or 3D denoising, but its Gpus do. Or, in a game, the NPU cannot perform AI frame generation, ray reconstruction, or AI in-game characters. That Gpu will be. Nvidia also had some benchmarks to bolster its claims, comparing the rtx4090 mobile GPU, the RTX4050 mobile GPU, and the Apple Macbook Pro M3Max with a dedicated NPU. The application included Stable Diffusion, Arnold, Blender, Chaos V-ray, Octane, Adobe Premier Pro Enhance Speech, DaVinci Resolve and ON1Resize AI. Nvidia figures pegged the RTX4090 with 7x performance than the Apple NPU in some tests, and the 4050 was about 2x faster with many results.

So, the question is whether Nvidia's beef is fair. Undoubtedly, part of Microsoft's motivation is to ensure that AI is added to PCs in an efficient and cost-effective way. It is undeniable that Nvidia GPUs do not do much for the efficiency of laptops in everyday use and are very expensive.

It certainly makes sense to encourage incorporating power efficient and cost-effective NPU into every CPU or APU. If ubiquitous AI is the purpose, Nvidia's GPUs will be priced down in terms of both power and cost efficiency. On the other HAND, it's a bit silly to certify a laptop with 45TOPS NPU as an "AI PC," but another laptop with 686TOPS worth of AI processing doesn't count.

On the other hand, it is unclear exactly when and how the various AI platforms are supported. For example, Microsoft's AI co-pilot currently assumes it will not run on Nvidia GPUs. But surely it could? Whose fault is it? Nvidia? Microsoft's "Both?

Anyhoo, you can guarantee that this is going to be executed, executed and executed. And then run. So, the AI grudge match is just beginning, so you'd better enjoy it.