Will AI chips find their way into desktop PC's in 2025?

Lsbcal

Give me a museum and I'll fill it. (Picasso) Give me a forum ...
Joined
May 28, 2006
Messages
8,977
Location
west coast, hi there!
Some Intel chips out now with AI features (has an NPU) are finding their way into laptops now. But I've not seen this for a small desktop pc yet. Maybe in 2025? Should I wait?

My motivation is that Windows 10 security will phase out in late 2025. My pc chip does not run Windows 11. Hence I have to dump a perfectly useful system. My desktop is a small built in area so I just want a 27 inch hi res monitor and a small pc box sitting next to it. Dell has a nice little Optiplex for $650 or so. But it does not have an NPU chip in it. I currently use Gemini and ChatGPT to answer common questions. Not sure yet what a local NPU will buy me but the apps may be coming so want to future proof.

I have a chromebook and don't need an expensive laptop. So the portability with also an iPhone is covered.
 
You could run Windows 10 IoT Enterprise LTSC 2021 (version 21H2), which has extended support until 13 January 2032.
 
One Dell sales guy suggested I just buy a laptop with a vertical stand. Then I can buy one with Copilot which has the NPU chip functionality. I would have good connectivity to peripherals through the premium 27 inch 4k Dell monitor. I just want my small walled in 2ft x 4ft space to be neat and very functional. So need to educate myself which is kind of fun. What else to do in retirement? :unsure: :dance:

I don't really need a laptop but maybe it's the way to go because of all the competitive forces in that market versus business oriented desktops. The money to buy this new stuff is not really an issue for me.

Thanks for replies so far.
 
How urgent is your need to get a new computer?

I ask because if you want a small box, a really small box you should definitely look into mini-computers, like Beelink or GMKtec. I'm not sure they have dedicated AI chips as yet. If you could wait 6 months you might get the best of both worlds--super compact size and AI chips.

I got DW a Beelink mini-PC and it's been great. Physically, it's about 6" X 4" X 2" in size.

I would not go with a laptop, unless you have room for an outboard keyboard and mouse. I can't stand typing on a laptop keyboard and fighting with the trackpad.
 
Yes, age and maturity count for something
:)

In the past I was also thinking like Qs Laptop. But none of us know where AI is going and that apps what might intrigue us. In my case I use Gemini and ChatGPT to just answer all the little questions that pop into my head. I can do that on just my current pc and iPhone 15. My computing needs are small, just some moderate size Excel spreadsheets of maybe 12Mb size.

I don't do gaming and don't anticipate doing video editing and such. But maybe something will pop up in the next few years that is fun and even useful.

That pcworld article is interesting. But it cannot tell me what the next few years hold in the AI application race.

Probably I will wait before pulling the trigger. But will buy before late 2025.
 
My understanding is that most mini pc's use laptop CPUs. Mini PC's don't need to worry as much about low power consumption since there's no battery.
 
Probably I will wait before pulling the trigger. But will buy before late 2025.
This is also my plan. I bought a new desktop PC in 2020, so it'll be only 5 years old next year, but I don't think I have the patience to wait much longer than that to start getting the full, "deluxe" NPU-powered AI experiences. I can certainly wait till around Black Friday 2025, though.
 
That pcworld article is interesting. But it cannot tell me what the next few years hold in the AI application race.
True, but it analyses the power of today's GPUs & NPUs. If you buy today, that's what you're getting. In a desktop PC, you'll have the option of some day using a discrete AI board just like today's discrete GPU boards. I don't know if those exist yet, but I'm sure they will when the apps appear.
 
I wouldn't be in a hurry to buy first/early gen AI anything, everything is still in flux, and the term "AI" is being thrown around where it really doesn't belong. And some apps are cloud based, so your local hardware may not matter much.

But I've always believed you start with the software you intend to use today and in the foreseeable future and then buy hardware/OS to enable same. Buying hardware and then seeing what it can run is a mistake IME.
 
I wouldn't be in a hurry to buy first/early gen AI anything, everything is still in flux, and the term "AI" is being thrown around where it really doesn't belong. And some apps are cloud based, so your local hardware may not matter much.

But I've always believed you start with the software you intend to use today and in the foreseeable future and then buy hardware/OS to enable same. Buying hardware and then seeing what it can run is a mistake IME.
As I mentioned in the OP, I am going to be forced into a hardware upgrade. Not in my case such a bad thing. I'd really like a 27 inch 4k monitor and why not. I might as well update everything and money is not the issue, just to make the right hardware decision.
 
The Macbooks and other Apple devices with M1 chips and 8GB of memory will run Apple Intelligence.
 
Progress in computer power has been mind-boggling to me for a long time. The first one I worked with was the world's biggest (physically) computer, the AN/FSQ-7, which weighed 250 tons, used 60,000 vacuum tubes, and had a crew of around 60 technicians, many of whom spent much of their time walking around replacing those tubes as they burned out. It was capable of about 75K instructions per second, using as much as 3 megawatts of electricity.

Today's computer speed is measured in FLOPS (floating point operations per second), and current supercomputers operate in the range of exaFLOPS (10^18 FLOPS).

Now the Japanese government has approved a program to build the world's first zetaFLOPS computer (10^21 FLOPS), so 1,000 times faster. Construction starts next year.
 
I thought Copilot was some Microsoft feature that just required Win 11 and a MS 365 license. New wrinkle: you have to pay monthly for the use of Copilot. I guess this is going to be a business thing at least to start with.

So yes, I will try to get the next computer with the latest cpu chip and npu. But not easy to see what it is good for me as just a retired personal use case.
 
Intel has just come out with a new chipset and maybe better NPUs:
Intel’s Core Ultra 200S Desktop CPUs Have Faster Speeds and NPUs

The article says these might be in some desktop pc's by the end of 2024.

Another thing, apparently to use the Copilot features from Microsoft an individual has to sign up to pay $20/month. I doubt many users will want to do that. Certainly not me.

Here is one article on how Copilot might help a MS Word users write better:
Microsoft Word Can Now Coach You on Writing
 
It is all marketing. NPUs can process AI tasks faster (and with less power consumption) than traditional CPU's, but unless you need a response faster than 1 second or so (like ChatGPT or Gemini) then you don't have to worry about it. "Apple Intelligence" is all marketing. They are using OpenAI for their backbone and the "new" apps can run on any phone - they just limit the access to the latest hardware to compel you to upgrade. It is all software based for the end user. Unless you are running enterprise-level AI (think corporate models based on proprietary terabytes of info) you don't need to worry about processing power or NPUs.
If you are in the Apple universe of course, they may require their "intelligence" hardware to run their apps. Great marketing!
 
It is all marketing. NPUs can process AI tasks faster (and with less power consumption) than traditional CPU's, but unless you need a response faster than 1 second or so (like ChatGPT or Gemini) then you don't have to worry about it. "Apple Intelligence" is all marketing. They are using OpenAI for their backbone and the "new" apps can run on any phone - they just limit the access to the latest hardware to compel you to upgrade. It is all software based for the end user. Unless you are running enterprise-level AI (think corporate models based on proprietary terabytes of info) you don't need to worry about processing power or NPUs.
If you are in the Apple universe of course, they may require their "intelligence" hardware to run their apps. Great marketing!
Apple touts that sensitive information like photos, voice, health etc data is processed locally.

They don't have to pay the cost of a data center or OpenAI to process that data. You pay to do that and for the hardware. Better deal for them?
 
Last edited:
Do you need it? If you don't know you may not benefit.
 
Apple touts that sensitive information like photos, voice, health etc data is processed locally.

They don't have to pay the cost of a data center or OpenAI to process that data. You pay to do that and for the hardware. Better deal for them?
Makes sense, but once again any phone from the last 5 years can handle that workload - you don't need a special NPU to process personal/local data.
 
I get a new laptop and phone every year. It gives me a productivity edge that pays itself back in short order.
 
Back
Top Bottom