In late April 2024, AMD released the Ryzen 8000 series professional desktop product line, which is the 8000 series family of Ryzen PRO. Compared with the well-known "normal version" Ryzen 8000 series, the "professional line" Ryzen PRO 8000 series has not changed much, mainly adding support for professional features such as Microsoft Pluton, ECC memory, secure virtualization, and remote management.
As for the CPU, built-in GPU, and NPU computing power specifications, there is almost no difference between the Ryzen PRO 8000 series and the Ryzen 8000.
It is precisely because we are aware of this that when we see in the relevant report of a well-known technology media, they claim that "the new Ryzen Pro 8000 series provides up to 39TOPS of AI processing power on desktops", and we immediately felt very confused.
The original text of the related report, in which the description of the 14th generation Core is even more confusing
As we all know, the Ryzen 8000 uses AMD second-generation NPU units, and the core architecture of both the "desktop version" and the "mobile version" is consistent. Therefore, theoretically speaking, the desktop version of the Ryzen PRO 8000 series is of course also equipped with the NPU with 16TOPS computing power. Unless AMD overclocks it by more than 100%, where will the 39TOPS computing power come from?
However, the truth of this matter may be far beyond everyone's imagination, and there are even some strange and funny elements.
First of all, the media that claim that AMD's new Ryzen processors "deliver 39TOPS of AI computing power on desktops" are wrong?
On the one hand, as we mentioned earlier, the new desktop version of Ryzen PRO has a built-in NPU that is not overclocked. In other words, its NPU AI computing power is actually the same as the well-known mobile version and desktop version of Ryzen 8000, which is 16TOPS error-free.
But on this basis, the so-called "39TOPS" is actually not made up randomly, because you can find the relevant instructions on AMD's official website. According to AMD, the Ryzen 8000 series processors have a built-in NPU computing power of 16TOPS, but at the same time, some of these models will have a "Total Processor Performance" of 39TOPS.
What is this concept? The so-called "Total Processor Performance" actually refers to the AI computing power of the CPU, GPU, and NPU of the entire processor combined.
In other words, in this report, the relevant media mistakenly regarded this "sum AI computing power" as the NPU computing power of the new product. While the numbers aren't wrong, the actual meaning is clearly very different.
Although everyone knows that today's CPUs, GPUs, and NPUs can generally perform AI computing, the problem is that their respective architectures, energy efficiency ratios, and AI computing formats are generally different. As a result, although they both support AI computing, it is basically impossible to use them in the same program and in the same task at the same time.
From this perspective, the "overall AI performance" metric of a processor, or SoC, is not completely meaningless, but it is clearly not a measure of how fast or slow it is when running AI tasks in most cases, and it does not have the value of side-by-side comparisons between different products.
If you follow this line of thinking, the server CPU can be claimed to have amazing AI computing power, although it does not support instant AI response at all (no built-in NPU)
Therefore, from another point of view, in fact, under the current technical system, as long as the operating system and AI program itself cannot achieve "100% efficient heterogeneous computing", then the propaganda of the so-called "total AI performance" by relevant hardware manufacturers will inevitably be more or less misleading.
[The picture in this article comes from the Internet]