NVIDIA was without a doubt one of the most anticipated keynotes of Computex 2024and the green giant always tends to arrive in Taipei with pockets full of interesting novelties, which has greatly strengthened in recent years, in which it has been crowned as one of the most relevant companies in the artificial intelligence ecosystem, it has been working for years before its use became popular, as we have seen between 2022 and 2024.

So while not all of the announcements we’ve seen today are AI-related, they’re not in the slightest, as NVIDIA once again focused not only on the computing power of its integrated devices, but most importantly for many, in all solutions that rely on said technology deliver enhanced user experiences for games, app design, content creation, data analysis, and much more.

On that note, and before I go through all of today’s announcements, I think it’s interesting to remember NVIDIA responds to the AI ​​PC paradigm with the RTX AI PC, a proposal that focuses on the computing capacity of RTX GPUs, capable of providing much higher performance than the NPUs found in the chips that currently integrate them, as well as those that we expect to see integrated in the next generation. In the image below this paragraph you can find the performance in the TOP (trillions of operations per second) of all GPUs of the RTX 40 generation

Computing capacity for AI (maximum)

To put this data into context, let’s recall that the Copilot+ PC specification that Microsoft unveiled less than two weeks ago stipulates that systems must have NPUs with a computing capacity of 40 TOP. Therefore, the most modest model of the RTX generation, RTX 4050 Mobile is on the verge of multiplying the value set by Microsoft by five.

However, in this regard, it is important to remember that NVIDIA does not deny the use of NPUs in systems. On the contrary, until now it has offered remarkable computing capacity with its GPUs and with the advent of processors and SOCs with NPUs it is promoting solutions which use the most convenient technology depending on the needsas we saw a few days ago when they learned of their commitment to a hybrid approach, which is very well illustrated in this diagram:

It is important to point this out in order to better understand the announcements we have seen today because it is clear to us approach used by technology for development both the necessary hardware and AI-based solutions for different purposes, and which suggests taking advantage of the most suitable platform (NPU, GPU or cloud) in each case depending on the specific needs (efficiency, latency, performance, etc.).

From left to right: SUS ProArt XP13/P16, ASUS TUF A14/A16, ASUS Zephyrus G16 and MSI Stealth A16 AI.

RTX AI notebooks

As we could already deduce from the table that shows the computing capacity for AI tasks, based on the TOP GPU of the RTX 40 generation, any laptop equipped with the same very comfortably exceeds the set minimums so the system can easily perform large type of AI tasks. And this, as we mentioned earlier, made sense in the definition of the RTX AI PC specification, which already includes more than 200 notebook models, according to data provided by NVIDIA.

Today we learned that this already big family adds six new teamssigned by ASUS and MSI and which, given that they have NPUs in their integrated devices, will soon receive the necessary support through an update so that their owners can use the features announced by Microsoft for Copilot+ computers.

These are the models announced today:

• ASUS ProArt XP13 Copilot+
• ASUS ProArt P16 Copilot+
• ASUS TUF A14 Copilot+
• ASUS TUF A16 Copilot+
• ASUS Zephyrus G16 Copilot+
• MSI Stealth A16 AI Copilot+

Related to RTX AI PC capabilities and its capabilities in the context of Copilot, we also learned today that Microsoft and NVIDIA are working together to offer developers new generative artificial intelligence features that will be integrated into the Windows Copilot runtime. That way, once available, their integration into development can be done using a simple API that offers options for working with SLM (small language models) and optimizing with GPU-accelerated RAG (augmented recovery generation).

Development and optimization

The technology base provides the necessary tools to create solutions, but it is at this point that developers step in and play a vital role in making artificial intelligence no longer an abstract concept and becomes a good handful of features and tools for our daily life. So while they may sound a bit far-fetched at first glance, the truth is that developer-focused announcements mean a lot to everyone.

That makes him Announcing the NVIDIA RTX AI Toolkit It’s so interesting, and as you can see in the image above, this toolkit allows developers to perform a few basic tasks in an easier way to guarantee that the AI ​​will perform at its best, which is achieved. first with personalization based on a generic model and later with its optimization.

And how do these improvements translate to the end user? To illustrate, I will without a doubt stick to the example shown by NVIDIA, on which we can see provided output before the same challengefor a general model and another that has been optimized and adapted for a specific context.

The result, as you can see, is surprising in every way. First and foremost, because the response given by the “tuned” model (right) fits the context perfectlysomething that doesn’t happen with the generic model (left), which lacks the context to give us a more precise answer.

However, it is not only that the response is better in the case of the model with which the RTX AI Toolkit was used, but also with much more efficient operation. The following table lists the key points

So when we take the data from the table of the maximum performance of the current generation of RTX measured in TOP, we see that the most modest of the models in the family (mobile RTX 4050 with 194 TOP) is able to provide a model optimized, response almost eight times faster than the absolute top of the line (RTX 4090 desktop, 1,321 TOP) with a general purpose model.

Another very important news for developers, from which we users, especially the majority of gamers, will undoubtedly benefit, has to do with ACE, a technology that NVIDIA already surprised us at last year’s Computex, and which during these twelve months has developed and, moreover, aroused interest a growing number of developers who see in this tool a proposition that allows them to give much more life to both Player interaction with NPCs as well as more recently the cinematography integrated into games.

Well, we found out today Artificial Intelligence RTX PCs will be able to use NVIDIA ACE technology locally through NIM (microservices). In this way, by reducing the dependence on the cloud, developers could be even more inclined to integrate this technology into their titles, and users will be able to enjoy a much more personalized gaming experience in more circumstances.

Developing…

More information