whats ai computing

[apoyusiken]

whats ai computing? i see ai laptops and mini pcs. is it only for ai training? 

[jaslion]

Its investor pleasing wording put on processors or gpus. Made to force as much ai stuff as possible down our throaths depseratly trying to find a way for ai to integrate in some form so they can start making money from it.

 

Basically they're just the regular cpu's and gpu's of the next generarion or current gen but with a name that has ai now and some have a npu added.

[apoyusiken]

Just now, jaslion said:

Its investor pleasing wording put on processors or gpus. Made to force as much ai stuff as possible down our throaths depseratly trying to find a way for ai to integrate in some form so they can start making money from it.

 

Basically they're just the regular cpu's and gpu's of the next generarion or current gen but with a name that has ai now and some have a npu added.

what does the npus actually do though?

[jaslion]

Just now, apoyusiken said:

what does the npus actually do though?

Nothing unless you have some specific local ai model rhat can use it and the npu isnt too weak.

 

For most people the npu will just be never used and be an added production cost you pay for.

 

 

[05032-Mendicant-Bias]

13 minutes ago, apoyusiken said:

whats ai computing?

Buzzword.

 

That said, GenANI performance is now a major factor in compute I buy because I use LLM and Diffusion a lot.

 

If you just game, you don't really need to care. The manufacturers will make hardware that play games and you really do not need to care how they do it, only the price/performance at various settins.

 

5 minutes ago, apoyusiken said:

what does the npus actually do though?

NPUs are basically APUs that can do MAC operations on 2D/3D matricies at low resolution.

 

Inside phones? For years they have been enchanting images. it's a big reason why phones are getting better

 

Inside GPUs? For years they have been running things like DLSS.

 

Inside desktop processors? Nothing useful for now, pretty much like RTX core in the RTX2000 series.

 

In the future, it's plausible, even likely OS will leverage NPUs to run locally GenANI assisted features of the operating systems.

[Mark Kaine]

It's just a buzzword/more options to spy on you (see Microsoft "co-pilot"...)

[05032-Mendicant-Bias]

25 minutes ago, Mark Kaine said:

It's just a buzzword/more options to spy on you (see Microsoft "co-pilot"...

Local models that run on GPU/NPU are the ones that don't spy on you and are private.

 

Copilot/Gemini/ChatGPT are all closed source, run on corpo computers and those do spy you.

 

Windows has lots of telemetry/spyware that has nothing to do with AI.

 

 

[Mark Kaine]

38 minutes ago, 05032-Mendicant-Bias said:

Copilot/Gemini/ChatGPT are all closed source, run on corpo computers and those do spy you.

Why does copilot "require" an AI chip then? "minimum TOPS" etc? It's all just marketing fluff of no benefit to average user (oh and thanks for all your data!) 

[Quackers101]

On 3/13/2025 at 1:07 PM, apoyusiken said:

whats ai computing? i see ai laptops and mini pcs. is it only for ai training? 

as others said buzzwords and its very broad and doesn't define "good AI systems", which makes it even more useless. mostly to push copilot etc.

 

there is so much future potential with "AI" content, not just generating spam, but have big local content dumps for what you need an algorithm to solve or having the right chips that can do the right functions you want from it. want your own local movement security? want an translator? want a full library for medical content saved on your phone and hopefully being accurate/helpful and one might be able to update it, while using sensors on your phone like camera to recognize injury?

most CPU's + GPU's can do their level of AI compute, but it really depends on what level of AI and what you need to compute. as some get more dedicated to just push AI data and maybe use less energy, or faster but less accuracy. but some of it can take more space from other more useful resources besides "AI", and that more GPUs will try to find more "AI/ML" to be run on them to stuff like upscaling etc, be it DLSS, XeSS, FSR.

 

future dedicated ai chips can get interesting though, with a promise for locally run and much faster or use less energy.

On 3/13/2025 at 3:06 PM, 05032-Mendicant-Bias said:

Local models that run on GPU/NPU are the ones that don't spy on you and are private.

 

Windows has lots of telemetry/spyware that has nothing to do with AI.

windows and nvidia be pinging them data collection to a new high.

 

local is mixed, various things that can go wrong, but like said, it can be nicer to have it run locally in some form.

[cmndr]

On 3/13/2025 at 5:07 AM, apoyusiken said:

whats ai computing? i see ai laptops and mini pcs. is it only for ai training? 

NPUs on consumer devices are mainly for inferencing. 
Inferencing is taking an already trained model, (which can take FOREVER to create and at great expense), and using it for something. 

AI, machine learning and statistical learning can do a bunch of things. 
1. Generate content (text blobs, images, video, music, etc. )
2. Modify content (photoshop style filters, language translation, etc.)
3. Make predictions and categorizations (what is in an image, what does the sound say, what comes next in a pattern, etc.)
4. Group like things together (often used in things like recomendation systems)
4. Estimate the change caused by some "treatment" (often used in things like recommendation systems)