Ketan Patel, head of HP’s PC and notebook division, is clear. He is sitting in front of me and two other journalists from other countries and He doesn’t seem too concerned..
Yes. The PC segment is going through a tough time. After spectacular sales growth during the pandemic – restrictions changed habits and customs – the last three quarters (Q4 2022, Q1 2023, Q2 2023) have been terrible for the industry.
Patel freely admits this. Things have not been going well and it will take some time to get better, he says. And yet his speech is not necessarily pessimistic, and in fact the conversation soon changes tone. They do so with these two words, which are the promised next great revolution in our world.
Artificial Intelligence.
Patel has been with HP all his life. He was responsible for that business unit in India before making the leap to run it globally, so he knows what he’s talking about. Of course: play with advantage. He knows something we don’t, and even though he seems dying to tell us what it is, he can’t. Not yet. But it does reveal something.
In fact, HP talked about it all day during the HP Imagine 2023 event. Alex Cho, another of the division’s top executives, has done this before and similarly nodded along the same lines during the launch conference. “I have something here,” he explained, showing part of a notebook in his bag, “but I can’t say more for a few months.”
Enrique Lores, HP’s CEO, laughs, perhaps a little sheepishly. This Spanish engineer has risen to the top of the company’s hierarchy after more than 30 years working there. He also knows what he’s talking about, and he also knows something (a lot) that we don’t. Alex Cho looks at him and says, “You didn’t expect that either, did you?” We don’t know if he expected it, but he laughs, perhaps admitting that he liked the wink.
We asked Ketan Patel about this revelation. “What does it mean?”. And then the debate begins about the role of AI and how They see it as a vital part of the future of the PC. Cho and his colleagues talked about possible changes in interaction – from mouse and keyboard to voice? Who knows, but above all in how to use artificial intelligence models.
From cloud to PC
Today, these models are in the cloud and are expensive. At least if one wants to have access to the latest developments. This happens with ChatGPT Plus or Midjourney in their latest versions, but the fact that they want to charge us for using these models is normal: the costs associated with these requests are very high in computing capacity and therefore in energy consumption and even water.
Those responsible for ChatGPT and Midjourney use huge and very expensive infrastructures. Tens, hundreds, and even thousands of GPUs are responsible for this chatbot answering any question with apparent solvency, and for Midjourney’s generative artificial intelligence to create stunning images in seconds.
Doing something like this on PC is possible, but it’s not cheap at all. If you have the latest generation graphics card – RTX 4090 better than 4080 and 4080 better than 4070 – you can install one of the existing open source models. Llama 2 is the reference when it comes to alternatives to ChatGPT, and Stable Diffusion is the reference if you want to generate images at home. The problem is that the performance of these models is not currently comparable to those maintained by companies like OpenAI.
If you don’t have the aforementioned equipment with the best graphics of the last generation, the reactions of those “home ChatGPTs” will be annoying if you want to use the most competent models (which often handle the most parameters).
For RTX 4090 and Llama 2 model 30B, according to some benchmarks, one is around 30 tokens per second, which is a more than respectable number, but in which the energy costs must be taken into account: the consumption of these graphics is very high, making Today a practical solution for the general public it’s easy to use cloud-based options like ChatGPT or Bard.
For Patel and other HP executives, however The future of these chatbots and AI tools is not the cloud, but the PC and laptop. Although they don’t give details, they make it clear that just as Microsoft just announced its big update to Windows 11 23 H2 with Windows Copilot, in a few months we will see some kind of “HP Copilot” integrated into its PC and that will go much further than Microsoft’s proposal .
“How is that possible?” I ask. “You can get by with a very powerful graphics card these days, but it’s not a feasible or practical solution for the vast majority of users,” I add. Ketan Patel smiles confidently again. Apparently he knows (or thinks he knows) something we don’t.
Alex Cho, President of HP’s Personal Systems Division.
He tells us how we will see it in a few months. Within a year for sure. At HP, they are developing a solution that is not detailed, but that It seems to be either ChatGPT or Copilot itself. One that will be separate from Microsoft and that will be a fundamental element of that transformation of the PC concept that Alex Cho and Enrique Lores and the rest of Microsoft’s leadership mentioned at the launch conference. It is also clear to them.
For them, the computer changes from a “personal computer” to a “personal companion” (personal assistant).
The difference is important because these devices offer the use of all the data we keep on our computers to help us work with them and facilitate all types of tasks. Do you have data on your finances? This assistant will be able to create alerts, summaries and even recommendations. Do you have a full calendar? This artificial intelligence will help you find gaps or organize everything better. Have you made a list of places you want to visit in the future? No problem, the assistant will ask you if you want to organize a trip to one of these destinations and book your flight, hotel and restaurants.
These are just some examples of this new “vertical” philosophy that Patel and his colleagues talked about, which adds to the traditional horizontal and general vision of computers: all for browsing the Internet or working with documents, but if you want a specific, “vertical” approach , the AI assistant will provide it to you. It will also do so in a completely private and secure way.: This data does not go to any cloud and the processing is local, on our PC or laptop, so everything stays at home.
The question, of course, is how he will do it when the energy and process demands are so high right now. For this manager, the answer lies in the partner chips they work with. The secret is NPUs (Neural Processing Units).
Before that, CPU and GPU were the main heroes. Make way: NPU will arrive
This is where doubts arise. Apple with its A17 Bionic and Google with its Tensor G3 have been promoting precisely this type of processor units in their mobile SoCs for a long time. Meta and Qualcomm are also working on these advances and announced an alliance in July 2023 to bring Llama 2 AI to mobile phones, but although the capacity of these chips has increased, their current use is limited (at least in mobile phones). field of computational photography.
The new Intel Meteor Lake NPUs are designed to be a pillar of the future “Personal Companion” that HP has been talking about.
The question is whether a processor aimed at laptops and PCs could achieve exactly that. Intel certainly seems to be suggesting this., which announced a promising leap in quality this year with Meteor Lake and its NPU. These “neural processing units” don’t necessarily work on their own, and in fact Intel revealed in September how they’ll be able to combine with CPUs and GPUs in all sorts of scenarios. We tried to analyze its performance and capacity, although without clear references it is difficult to decide whether the jump will be as significant as Intel promises.
In fact, in internal tests, Stable Diffusion v1.5 saw how it took 20.7 seconds to run 20 iterations of this generative AI engine using just the NPU and consumed 10W. When they did it with GPU help, the NPU time dropped to 11.3 seconds , but 30 W was consumed. It would be ideal to be able to choose whether we want to complete the task in less time or consume less power, but at this point it is not clear how they will decide on this.
Be that as it may, it doesn’t seem possible in the short term that a future iteration of Google Tensor (G4, G5?) on a mobile phone or Meteor Lake integrated into a laptop can offer the processing capacity offered by the RTX 4090. in AI tasks .
Or can it? Patel could say no more. And yet he looked confident.
“We’ll see in a few months“, he explains. To be sure, he warns, “this technology will be first generation and will improve in successive iterations.” And yet there he remains, apparently convinced that they have a winning card.
One that will obviously make our computers much more than they are now. And if that’s the case, maybe the PC will reappear. And then – we assume – a smartphone will come, which could also give us local access to ChatGPT or Stable Diffusion.
One that’s private, secure, and uses everything we do on mobile to make our lives easier. If this promise is fulfilled, We will certainly have a very strong argument for changing the PC.
In Xataka | How to have AI on your computer as ChatGPT local with GPT4All