50 TOP of your NPU

• With these chips, AMD beats the 45 TOPS of the Qualcomm Snapdragon processor.

• The company boasts better performance in all areas, but does not talk about efficiency.

• More than 100 devices from major manufacturers based on these chips will be available in July.

AMD debuted at Computex 2024 with two major presentations. Firstly, it’s a new twist on “traditional” processors, which are updated with the promising AMD Ryzen 9000.

The second, even more striking, is new AMD Ryzen AI 300, which represent AMD’s first big bet in the new era of AI-powered PCs. And pay attention, because here the company has become more active and promises to knock all its competitors off their feet.

AMD Ryzen AI 300 consists of three main components:

1. CPU. With Zen 5 architecture, the main processor of these processors will consist of up to 12 cores (4 Zen 5, 8 Zen 5C efficiency) and 24 threads, clock speeds up to 5.1 GHz and cache memory up to 36 MB. It’s true that the power in this section isn’t what the best Ryzen 9000 series models offer – they’re more geared towards gamers – but there’ll still be more than enough wiggle room for most users.
2. GPU. With RDNA 3.5 architecture, the integrated graphics can be either Radeon 880M (12 CU) or Radeon 890M (16 CU), this is an evolution of a component that will allow you to enjoy video games with more than decent performance, even if we are not talking about gaming hardware as such .
3. NPU. The real protagonists of these chips will be the neural processors with the XDNA 2 architecture. We are faced with a huge leap since the first generation of these components: if XDNA reached 10 TOPS performance, these chips will achieve 50 TOPS.

Thus, we are faced with proposals that will directly compete with Qualcomm Snapdragon.

Ready for the Computer Age Copilot+

These promised NPU specs make it clear that these chips are specially prepared for the new batch of Copilot+ PCs, where AMD will offer the full power of Windows 11’s new AI capabilities.

This way, you can take full advantage of features such as calling, translation or real-time subtitling, as well as the creative capabilities of Cocreator. The transition from the first and second generations of AMD NPU chips is noticeable and leads to the use of new artificial intelligence capabilities, and, above all, those capabilities that work directly locally, without connecting to the cloud.

AMD emphasized that the power of its XDNA2 NPU allows for much higher responsiveness in artificial intelligence applications than previous chips. For example, when running Llama 2 7B, the time to generate the first token on the new processors is five times faster than on the AMD Ryzen 8940HS, which had a first-generation NPU.

According to the company, it also far exceeds what the Intel Core Ultra 9 185H and Meteor Lake processors can offer in other areas, but not aimed at artificial intelligence. When Intel unveils its upcoming Lunar Lake, yes, the comparison will be much more interesting.

AMD also emphasized that accuracy is critical in artificial intelligence applications. 8-bit integer operations have better performance but lower precision, while 16-bit floating-point operations have much higher precision but lower performance.

It is with these chips that AMD introduces the first “block” NPU FP16, which achieves a combination of both characteristics. Considering that most AI applications use 16-bit operations, AMD offers this technology which makes model quantization unnecessary.

Quantization is a technique used to reduce the precision of numerical values ​​(weights and activations) in an AI model. The main purpose of quantization is to reduce the memory and compute requirements of the model, allowing for faster inference and deployment on slightly more resource-constrained devices, such as our PCs, laptops, or mobile phones.

AMD boasts that it outperforms its competitors. We will see the effectiveness

AMD’s internal tests show its chips outperform Qualcomm’s Snapdragon X Elite in everything from performance (10% faster) to graphics (60% faster performance).

They also outperform the Apple M3 with 9% more performance, 70% more multitasking, and 98% (almost double) more 3D rendering performance. History repeats itself when comparing them to the Intel Core Ultra 185H: AMD boasts up to a 36% performance boost for gamers who choose these chips.

These internal tests are certainly promising, but at the moment there is no data on the consumption and energy efficiency of these chips, the section in which Both Apple and Qualcomm are likely to overtake AMD. We will have to wait until we have this equipment to confirm both of these internal tests and be able to talk about the consumption and effectiveness of each.

Starting from July 2024, we will see more than 100 computers based on Ryzen AI 300. Acer, Dell, HP or Lenovo are among these protagonists, but also some of them, such as ASUS or MSI, will present their news at Computex based on these promising chips.

Thus, they will join the numerous proposals based on Qualcomm Snapdragon.

In Hatak | Qualcomm is just the first step towards Windows 11 with Copilot+. Intel and AMD chips are waiting in the rear