Google says running AI locally on phones requires tons of RAM

Shawn Knight

Posts: 15,297   +192
Staff member
In a nutshell: Google's Gemini AI model landed on select mobile devices earlier this year via Android app, but some were surprised to learn that one of the company's latest flagship smartphones didn't make the list. During Mobile World Congress, it was revealed that Gemini Nano couldn't run on Google's Pixel 8 due to unspecified hardware limitations. Now, we've got a better idea of what was holding the phone back from on-device AI compatibility.

On a recent episode of the Made by Google podcast, VP of devices and services software, Seang Chau, said the Pixel 8 Pro with 12 GB of RAM was the perfect platform to load Gemini Nano onto and see what was possible. The entry-level Pixel 8, however, ships with 4 GB less memory, and it seems Google didn't want to "degrade the experience" by forcing the AI model to run on just 8 GB.

Google has since had a change of heart – well, sort of. In a recent announcement on the Pixel Phone Help site, the tech titan said it will be rolling out Gemini Nano to Pixel 8 users as a developer option in the next software drop. That's good news for devs that know how to enable it, but mostly a nothing burger for the average Pixel 8 user – at least, for now.

Jumping back to Chau's podcast appearance, the Googler explained that the company wants some of its AI-enabled features, like smart reply, to be "RAM resident." This means they "permanently" occupy a chuck of memory, which will allow them to be ready to go at a moment's notice.

It's worth mentioning that Google told Ars Technica that neither the Pixel 8 nor the Pixel 8 Pro are keeping Gemini in memory currently, and that you have to turn on a the developer flag to do so.

Armed with this new information, some will no doubt question whether or not they want Gemini Nano running on their phone at all. An unrestricted install will result in "permanently" losing a chunk of system memory, which could slow down the performance of other apps and services on your smartphone. Is that trade-off going to be worth it?

Permalink to story.

 
Then don't run AI on phones? Seriously, nobody is asking for this and it's very typical of the mobile market: this is a solution in search of both a problem and the necessary hardware to run it, mostly because Google and it's partners must sell more phones even when there's been literally no reason to own a new phone when even a midrange phone from like 10 years ago (Outside of batteries dying...Which they used to be replaceable btw and there's no good reason for them not to be other than buy more phones) is probably identical in performance for you know, phone tasks: taking pictures, checking social media and wasting time on 80s era level of gameplay if not 'games' that basically play themselves for you to just watch something in case you run out of social media posts to view while waiting in line somewhere or taking a poop.

That's all mobile phones need to be for 99.99% of the user base but of course since Google wants to pester us with an AI phone they will demand that their next version of Android will be AI 'Enabled' and in other words mandatory AI on by default and that means of course 'You need more ram' and that in turn means 'Yeah all phones are now 30 to 50% more expensive since they need minimum 16gb ram now...Because of AI of course'
 
For a tech site the comments sure are anti-tech.

Local AI sounds pretty good to me instead of this laggy cloud crap. If prices for phones with it are going up because a lot of RAM is needed so be it. Dont need it? Get a cheaper phone! It's not like suddenly there won't be any non-AI phones and the price floor suddenly goes up by $200.

Unless you're someone who buys the Super Ultra Hyper Pro (and those kinds of adjectives) version of whatever phone because you have to have the best for no good reason then why would you care. You get to spend even more to distance yourself from us poor plebs.
If you're not that person then just keep buying a phone that offers what you need.

I'm still perfectly happy with my 6 (7?) year old phone. If I do AI things it'll be on the PC where the memory doesn't come bundled with the rest and can be upgraded independently.
 
If you build me a phone that can run Borderlands 2 or 3 locally and without lag, I’ll buy it. But I have no interest in this AI crap. Why is this being forced upon us on literally every platform?
 
Just because some AI is stupid or randomly tacked on doesn't make all of it useless.

Machine learning (”AI”) can do some cool things and running locally is both more reliable and less creepy.
 
It's coming whether you like it or not.
I get at moment use case to cost/battery probably not there for many.

But it's a tool like the calculator , stop watch, compass etc on your phone . Will be nice to have it on your phone as a pretty good expert in crutch situations.

It's like saying decades ago , I don't want the WWW/internet or AF on cameras. Remember "real" photographers speaking bad of focus on peoples eyes as a consumer gimmick

It's how much interference you want in your life "hey human you forgot to put yeast in your dough "

Anyway on a PC , I waste time that a AI agent could do and speed up -"go look at all sites for new science fiction books to read and that I might like".

People write tools to monitor for stuff they like. Now normal people can "write " these tools
 
Really Google..?
AI on phone, right now..?
Indeed, the development of AI cannot be stopped, but wouldn't it be better to prepare better (and more affordable) hardware first rather than forcing the use of technology that is currently not needed by many people..?
regular phones without AI are very useful for many people..
 
It's coming whether you like it or not.
I get at moment use case to cost/battery probably not there for many.

But it's a tool like the calculator , stop watch, compass etc on your phone . Will be nice to have it on your phone as a pretty good expert in crutch situations.

It's like saying decades ago , I don't want the WWW/internet or AF on cameras. Remember "real" photographers speaking bad of focus on peoples eyes as a consumer gimmick

It's how much interference you want in your life "hey human you forgot to put yeast in your dough "

Anyway on a PC , I waste time that a AI agent could do and speed up -"go look at all sites for new science fiction books to read and that I might like".

People write tools to monitor for stuff they like. Now normal people can "write " these tools
18 months ago, the EXACT same thing was being said of crypto/NFTs.
For a tech site the comments sure are anti-tech.

Local AI sounds pretty good to me instead of this laggy cloud crap. If prices for phones with it are going up because a lot of RAM is needed so be it. Dont need it? Get a cheaper phone! It's not like suddenly there won't be any non-AI phones and the price floor suddenly goes up by $200.

Unless you're someone who buys the Super Ultra Hyper Pro (and those kinds of adjectives) version of whatever phone because you have to have the best for no good reason then why would you care. You get to spend even more to distance yourself from us poor plebs.
If you're not that person then just keep buying a phone that offers what you need.

I'm still perfectly happy with my 6 (7?) year old phone. If I do AI things it'll be on the PC where the memory doesn't come bundled with the rest and can be upgraded independently.
Being a tech site doesnt mean you uncritically accept all tech being shoved down your throat with no criticism.
 
If AI needs more ram and more ram is added, wouldn't that also increase the price of the phone when memory prices go up? There are just some things cell phones should not have.
 
The way people use it, it will be a goldmine for Google's Telemetry. Think about it.
AI as a [in this case] corporate plaything is the outcome of perhaps the biggest privacy and intellectual property scandal in scale thus far. It is the result of these vague privacy policies intentionally designed for us to ignore, and more.
 
This push for AI in computers, phones, toasters make me think it is a much better way than the conventional to steal data from people. And cost effective. You will pay the cost of hardware you don`t need and they will get results of your traits and behavior already analyzed and ready for sale. Ofc, you`re also going to massively benefit from it, like putting bunny ears on people and other crazy fun stuff!
 
The fact that AI requires a lot of ram is known for some time, so I wonder why google did not release the pixel 8 with 12 - 16 GB ram given their focus on AI. If motorola can put 12 GB ram in a $250 phone, I believe it was not the price that stopped google.
 
The way people use it, it will be a goldmine for Google's Telemetry. Think about it.

Agreed. This is exactly the bottom line and what they are counting on to recover their investments. I think most people don't realize how they are getting bent over by their connected devices until they consider the telemetry they send out daily. Especially phones and cars.
 
Then don't run AI on phones? Seriously, nobody is asking for this and it's very typical of the mobile market: this is a solution in search of both a problem and the necessary hardware to run it.

This is the same tactic that nVidia used to push raytracing on the world. Most gamers were just fine without raytracing and the effects it added to most games are "meh" at best. But it was an excuse to launch new hardware at exorbitant prices.
 
Agreed. This is exactly the bottom line and what they are counting on to recover their investments. I think most people don't realize how they are getting bent over by their connected devices until they consider the telemetry they send out daily. Especially phones and cars.

I'm 100% sure of it that it will some how end up that way. Why do you think MS is pushing the Co-pilot AI thing so hard?

If you have millions of users feeding it data every day, well....
 
What's the ultimate goal here? Running an AI model on your phone is going to be incredibly energy-intensive, chew through the battery life and offer minimal actual utility to the end user, so to what end is this being pushed as the new hotness?

My theory is that Google (and Apple) know that AI workloads are extremely expensive and demanding to run on their servers. So, rather than eat the cost of running this stuff in-house, instead they might create smartphones with particularly good specs--for a price--that just so happen to be really good at hosting LLM models, that they can then offload workloads onto. They pass the cost of running AI onto the consumer while selling a bunch of premium products in the process, which makes their business expenses go down and their stock go up. After all, who doesn't want the latest and greatest stuff?

But that's just a Theory...A TechSpot Theory! Thanks for reading!
 
Nah.

Microsoft, Google and such are both betting "huge" on AI. And for what reason? What do both company's have in common that made them initially, great to begin with?

Data. Harvesting of data of over a million devices. Think of it like Tesla's fully automated driving. It can harvest data from live video feeds over a million of sold vehicles on a daily basis. Maps works exactly the same, by combining all that data into one.

If people use Ai to their benefit it will "feed" the bigger model at their end and likely open up new opportunities which can lay off staff to begin with in the first place. This AI is dangerous in various ways, one thing for sure: it's that it's going to lose lots of jobs all over the world.
 
Back