AI's Memory Crunch Is Coming For Android Apps (techcrunch.com) 29
Google is imposing new memory-use limits on Android apps as the AI data center boom contributes to a broader memory chip shortage that could leave lower-cost phones with less RAM. Developers will have until February 2027 to meet new thresholds for memory and bitmap usage. To aid developers, Google is adding tools to flag apps that exceed the limits and help prevent slowdowns and crashes. TechCrunch reports: The company explains that the mobile industry is now facing "significant hardware supply constraints that are altering device memory availability," which can then, in turn, affect the consumer's experience with their devices. To address this, Google is now establishing new performance thresholds across several areas, like dynamic memory usage and bitmap usage. In addition, Google is adding code optimization requirements designed to prevent things like app slowdowns and crashes related to performance.
Honestly (Score:2, Insightful)
This sounds like a good requirement.
Re: Honestly (Score:1)
Probably. But it would mean no more java for angle of sphere, and possibly no more linkedlists for him either. You'd have to pull both from his cold dead hands. Is that something you're prepared to do?
Re: (Score:1)
Re: Honestly (Score:2)
Except you are accepting their stated reasoning as gospel. They don't want to free up RAM because of a RAM shortage. They want to free it up to run local AI models on the phone, mostly for data harvesting, behavior recognition, and surveillance.
All these years... (Score:4, Insightful)
So Google could have provided this service, this whole time?
ALPHABET's slogan: do evil, until forced to do otherwise!
Re: (Score:3)
No one wants this. AI on your phone is one of the worst ideas I can imagine. You have a 4K screen-- and now you're trying to do pixel edits on animation?
You want to find your way through Manhattan traffic at 4pm on a Friday in your Jeep? Imagine wickedly insane requests, seemingly unlimited by token costs, barraging your phone. Remember metered minutes and texts?
These models aren't compact. 5G traffic, which is sort of tenable in most areas now, is going to MCP'ing fifty requests a minute to answer simple q
Re: (Score:2)
Um, no, the whole model isn't on the phone. I didn't suggest that. Instead, each app will grow into blimp-sized monstrosities, load phones down with MCP traffic, and drag phone performance to a crawl, all while increasing net traffic meteorically.
Prove me wrong.
Re: (Score:2)
There's nothing to prove wrong. You're talking complete irrelevant and off topic rubbish. Nothing about this story has anything to do with running AI on your phone. Not one thing. This story is about app optimisation on Android.
But even if it were, let's unpack your verbal diharrea:
- Why would an app grow to a blimp size monstrosity? Just because Google's premier LLMs are massive, doesn't mean all AI models are. There's many AI models that are small. One person even got an AI model to run on an ESP32 with 1
Re: (Score:1)
- Why would your phone slow down? AI is used when you use it. When not, it doesn't do anything.
Apparently, OpenAI is already addressing this.
https://features.slashdot.org/... [slashdot.org]
Re: (Score:2)
While it would have been nice, it's not exactly Google's job to force developers to be less sucky. Some may argue "evil" is placing restrictions on developers, but to do those people I say this would be a necessary evil.
What's old is new again (Score:3, Insightful)
Re: (Score:2)
I don't know why more developers don't start with that in mind.
Because "get it coded fast, we'll worry about resource-related problems in a future release" is a winning strategy if you expect to never hit your resource limits.
Re: (Score:3)
To be honest, I don't know why we stopped, because it's not as if modern frameworks take less time to learn or are significantly easier to visualize HTML in.
And many of the sites that migrated to these absurdly heavy web frameworks seem to be just as slow as they ever were. T-Mobile's website comes to mind. Clumsy, falls apart, and makes the fans whir in every laptop I've ever seen - and it's been like this for at least 10 years.
What are devs trying to achieve here when they use them? 'cos it sure as hell i
Re: What's old is new again (Score:3)
Well, remember how every Unix app, server, and utility in the 80s and 90s was written in C? Because there was only one refrigerator-sized departmental server with a only few megabytes of memory and two dozen users connected through serial lines, and writing in C was the only way to squeeze out every bit of performance out of it. There was a disk usage quota system allowing each user only a few megabytes of storage. The developer mentality today is that the resources are infinite and they can write code in t
Re: What's old is new again (Score:2)
"Premature optimization is the root of all evil" - Donald Knuth. It's also my sig, in optimized form.
The point is worrying about optimization when it's not necessary can make it hard to get the app working at all. But now is a good time to worry about optimization.
Re: What's old is new again (Score:2)
ObMeme (Score:2)
we're-all-trying-to-find-the-guy-that-did-this.gif
Ooh, easy! (Score:2)
Re: (Score:2)
Eliminate all the hooks into AI.
Atari 2600 had 128 bytes of RAM (Score:4, Insightful)
Let that sink in a moment. The graphics were not great but the games worked, were playable, were responsive and used 1/500 000 000 the power of the PC I'm writing this post on. Modern programs are bloated in every possible way, graphics, computations, messaging
Re: (Score:2)
My favorite version of this is the original Super Mario Bros was only 40KB, with 32KB of that being code and 8KB of that being graphics. That's it. This means that most screenshots of SMB are bigger than the entire game, sometimes by orders of magnitude depending on specifics.
Re: (Score:1)
And the ROM only had 4K of addressable memory.
I hear what you're saying but let's be honest, Atari 2600 games had terrible, terrible graphics. Even back in the day, if your friends had something else, like a Mattel or a Colecovision, let alone a C64, you'd rather play them than your 2600. If you made Atari 2600 games today, nobody would play them. Aside from old retro geeks, the same guys who also insist on playing Mayfield's Star Trek.
But yes, the actual games were quite good and fun to play. More fun than many today.
Well, f gloogloo! (Score:2)
How about not shoving generative AI everywhere? (Score:2)
Generative AI is what's causing the memory crunch, so stop forcefully putting it everywhere. Most people do not want LLMs in every app.
Ironic (Score:2)
Memory limits from the company that brought us Chrome!