AMD is Changing the Naming of the Strix Point APUs Series Again

[Vya Domus]

None of companies would be wasting such area of silicon and time taken to design it if they thought that AI enterprise was just a momentary tech fart.

They absolutely would, history is chuck full of failed types of processors.

 

[john_]

As if you're constantly bumping your toe on something

I mostly had music in my mind, not pain

They absolutely would, history is chuck full of failed types of processors.

Especially AMD that spend 30+ billions for Xilinx, seen RX 7000 failing because they chose to not spend more die area for increasing RT performance and also sees Nvidia going fast for 3 trillions valuation(after their share split), I am expecting them to go full in AI.

 

[Tek-Check]

They absolutely would, history is chuck full of failed types of processors.

The difference this time is that entire industry is in it. The only difference is that marketing departments are shouting about it and we are sick of it.

 

[Vya Domus]

The difference this time is that entire industry is in it.

It doesn't matter, it can still mean that most of this newly made silicon wont ever amount to anything. Think how many semiconductor companies existed in the 80s or 90s and most of them are gone, they all often made the same type of products and they still went under.

What's funny about all this is that phones have been shipping with NPUs for like 5-6 years by now, so long before the recent AI craze, and the use cases for ML on these platforms are still laughably limited, such as making photos better (maybe?), improved translation/dictation (still sucks), small things like that. It's so strange how hundreds of millions of people already have devices with ML accelerators in their hands but this features has been so underutilized they don't even know about it and now everyone thinks that adding NPUs to PCs is some crazy next big step lol.

It's frustrating seeing people go nuts about AI but then you have basic problems that should have been solved with ML by now and yet years have passed and they haven't. Like if you record a video on your phone and there is even a little bit of wind/noise the audio quality will be in the gutter, you'd think they could use the NPU to clean up the audio but nope, still sounds like utter garbage and these company already talk about "AGI".

Don't worry GPT-8634 will be sentient and replace every job on the planet, trust us bro, just buy one more gazillion dollar accelerator from Nvidia, we're nearly there.

 

[AusWolf]

Yeah this AI hype is really taking off now, isn't it!

The funny thing is, it would have probably never taken off so quickly if people actually called it what it is (language learning model, or matrix calculations, or anything similar) instead of this dumb AI moniker.

 

[Vayra86]

Don't worry GPT-8634 will be sentient and replace every job on the planet, trust us bro, just buy one more gazillion dollar accelerator from Nvidia, we're nearly there.

'Don't worry, this (insert crypto coin) will truly decentralize and liberate all your finances and your money from the great claws of evul governmuntz, just add a few more ASIC's, we're nearly there!'

The similarities are too striking to ignore.

The difference this time is that entire industry is in it. The only difference is that marketing departments are shouting about it and we are sick of it.

Yeah, they added a few square mm of transistors to their chips so they can sell their old shit as new. Or in many cases, not even that, there's just the marketing and the chip is simply a next iteration.

 

[Tek-Check]

What's funny about all this is that phones have been shipping with NPUs for like 5-6 years by now, so long before the recent AI craze, and the use cases for ML on these platforms are still laughably limited, such as making photos better (maybe?), improved translation/dictation (still sucks), small things like that. It's so strange how hundreds of millions of people already have devices with ML accelerators in their hands but this features has been so underutilized they don't even know about it and now everyone thinks that adding NPUs to PCs is some crazy next big step lol.

I can definitively say that AI-assisted photo editor on Galaxy S24 Ultra is good, relatively fast and asks you to view and choose proposed enhancements. There's also a live translator for on-going calls and 4-5 other applications. It's early days as local hardware has just started to be better, so that software could take more advantage of it.

Like if you record a video on your phone and there is even a little bit of wind/noise the audio quality will be in the gutter, you'd think they could use the NPU to clean up the audio but nope, still sounds like utter garbage and these company already talk about "AGI".

One issue with NPUs is that they have been really weak in recent years, and so software solutions could not compensate for weak local hardware. It's a closed circle. Cleaning audio and video noise and colour fine-tuning of movie files in Hollywood studios require powerful Threadripper/Xeon + GPU systems, in conjunction with professional applications, user sitting there and fine-tuning the process multiple times. One such file can be 100GB and this post-production work can last several weeks. The task does not sound trivial at all.

Now, we already record 4K files private on mobile devices, which is a success on its own. Perhaps we expect too much from current AI-aided de-noiser in much smaller and less capable devices such as mobile phone or laptop? We want things to be done automatically and fast for us. That requires processing power, as you do not want to get a file without audio noise, but with messed trebles, bass or dialogue. This requires such AI algorithm to run multiple instances of checks across frequency ranges after removing some sounds. And fast, as you don't want to wait two hours. Such things cannot happen instantly without more capable hardware.