With more and more hype around "AI PCs", it's important to learn about the benefits they offer for those of us who often need to use computers to get things done. Recently, AMD's Ryzen 7040 (Phoenix) and Intel's Core Ultra series (Meteor Lake) have brought new surprises to a new generation of chips by adding a neural processing component (NPUS). However, the test results can be more complicated than you think. Although we see the advantages of the new generation of chips, most of them do not seem to come from their own NPUs (neural processing units).
The processing devices we often use in desktops and laptops contain many different components, such as CPU cores that do traditional work, GPU cores that process graphics, and connections to memory, storage, and external devices. Thin and light laptops typically rely on the processor's integrated graphics, and AMD has been leading the way in this regard thanks to Intel's leading process technology.
Last year, when we tested a variety of laptops, we found that laptops with AMD Ryzen 7000 series performed faster in graphics-intensive tests, while laptops with 13th Gen Intel processors performed better in traditional applications. This year, we've seen some laptops with processors from both companies add an NPU to the base processor. For example, AMD's Ryzen 7840 series (Phoenix) uses a ZEN 4 CPU, which can perform 10 trillion operations (TOPS) per NPU, and the entire system (including CPU and GPU) is up to 33 TOPS. The processor was launched in May last year, and the first thin and light laptop to carry it has already arrived.
Intel's recently launched Core Ultra series (Meteor Lake) features a very different design from the previous Raptor Lake series, based on various "chiplets" or "tiles". I've seen a few machines with Core Ultra 7 155H that have an NPU as well as Intel's Arc graphics, eight XE graphics cores. Users say that the Meteor Lake chip can handle up to 34 TOPs, with the NPU on board contributing 11 of them. Both the Ryzen 7840U and the Core Ultra 7 155H scored significantly better than their predecessors in traditional PC benchmarks, but overall, Ryzen is still slightly faster.
However, when we went through more rigorous testing, the results became interesting. In some tests, Ryzen performed better, while in others, Intel's chips performed better. This situation shows that whether the NPU can make a significant difference lies in the design optimization of the program application, and more data needs more practical application testing to speak. In general, these new generation chips have indeed improved their performance, but whether they are different from traditional CPUs in specific applications still need to be further observed and evaluated by users, and the differences in the software of the application will be different. And we will continue to pay attention to and use more NPU applications to make more accurate judgments and evaluations for everyone.