Google says the AI-focused Pixel 8 can't run its latest smartphone AI models | Hacker News Reader

Google says the AI-focused Pixel 8 can't run its latest smartphone AI models

arstechnica.com

33 pointsduxup26 comments

Not surprising. AI models are rapidly evolving and evolve more quickly than HW can keep up.

For instance, I'm guessing Tesla is wishing they had more capability in the latest FSD HW(HW3?), which is now something like 4-5 years old.

The point is excluding the ram difference, the hardware on the devices is the same. They launched at the exact same time.

We’re talking about LLMs. In that context it makes no sense to “exclude the ram difference.”

The same model utilizing the same amount of ram the Pixel 8 has runs on Samsung's latest phone.

> excluding the ram difference

Well, you can't really exclude the RAM difference, not for an LLM.

They are on HW4 at the moment: https://www.teslarati.com/tesla-hardware-4-hw4-deliveries-mo...

I barely follow Pixel phone news, and even I knew it was only the Pro model that could do this.

What kind of reporting is this?

This is....nothing

The press release over 3 months ago:

> Pixel 8 Pro — the first smartphone with AI built in — is now running Gemini Nano

> https://blog.google/products/pixel/pixel-feature-drop-decemb...

The Pixel 8 does not have Gemini Nano, the Pixel 8 Pro does.

Apparently, one of the things that make it "Pro."

It's probably the RAM - 8GB on Pixel 8, 12GB on the Pro. But also, companies do software locks all the time, eg various software features not available on the cheaper ones.

TFA:

> The main differences between the two phones are screen size (6.7 inches versus 6.2), battery size, a different camera loadout, and 8GB of RAM versus 12GB. RAM is the only known difference you can point to that could create a processing limitation, but Gemini Nano also runs on the Galaxy S24 series, where the base model has 8GB of RAM. RAM being the issue would mean Samsung phones are somehow more RAM efficient than Pixel phones, which is hard to believe.

> RAM being the issue would mean Samsung phones are somehow more RAM efficient than Pixel phones, which is hard to believe.

Not necessarily. It’s at least possible that the overall experience is degraded in the low RAM system, and Samsung is willing to overlook this… which isn’t hard to believe.

There could be other secondary factors. Perhaps it's something to do with e.g. computational photography, where the Google Camera app needs to quickly allocate a ton of memory for all the HDR frames it stacks. Or the fancy call screening. Samsung might not have those kind of constraints.

It's probably a safe bet to say that the phone itself is perfectly capable of running the raw model, but not necessarily while remaining usable or without disrupting other features.

Or maybe Gemini Nano takes 6GB of RAM and Samsung is okay with the performance of leaving 2GB for everything else.

Something that "wasn't clear" to Ron Amadeo is always the fault of Google and never the fault of Ron Amadeo not being qualified to report on the topic.

Tech news in general seems kinda anti-tech these days.

The Verge, Arstechnica, etc - their tone is at best highly skeptical about tech, and often downright opposed to it (especially if it's AI).

Are you suggesting in this specific case that taking Google's announcement at face value is somehow more technical than asking the question "What is the hardware limitation on the base Pixel 8?" Given they use the same CPU, and the Samsung phones support the model with 8GB of RAM, what is the hardware limitation on this 5 month old flagship phone that was sold as having this great AI chip?

We used to have tech reporting done by tech people. "Shannon Knows DEC" was written by an NSA Spook. Wired used to employ ex-phreakers like Steve Steinberg.

Wired in particular has gone downhill these days.

Now they write articles like "my balls-out quest to achieve the perfect scrotum": https://www.wired.com/story/my-balls-out-quest-to-achieve-th...

...you weren't kidding

You’re right, they should just unquestioningly report on every latest shiny thing without talking about the negatives. That’s journalism, right?

Also from a press release 5 months ago:

https://blog.google/products/pixel/google-pixel-8-pro/

> Meet Pixel 8 and Pixel 8 Pro, engineered by Google and built with AI at the center for a more helpful and personal experience. These phones are packed with first-of-their-kind features, all powered by Google Tensor G3.

Obviously the answer is "Google wants to differentiate these phone price structures with features" to make the Pro more appealing, but they use the same "AI" hardware, and it's not because they never marketed AI on the base Pixel 8...

Yet they literally said this time around

> "[Gemini] Nano will not be coming to the Pixel 8 because of some hardware limitations."

Some of us just want a clearer explanation of what the hardware limitations are, and they they don't show up in the spec comparisons between Pixel 8 and 8 Pro.

> they don't show up in the spec comparisons between Pixel 8 and 8 Pro

There's a 4GB RAM difference between the two.

With what I know of LLMs...that'd make sense. But sure I'd love a technical walkthrough of everything.

Furthermore, both have AI, e.g. creating photos, but only one of those runs Gemini Nano.

Which, to be clear, launched after both phones were released anyway.

The phones were released in October, while Gemini Nano's announcement happened in December. I, like other developers and consumers reaching for the smaller version, might've bought the device for the ability to run the ML features advertised in their keynote/based on the research they released the week prior to that (in the case of the former.)

During Gemini's initial release the language surrounding nano was that it was only the Pro initially, and I was happy to wait. The complete inability to run it, when the new Samsung phones can (including the model with 8GB as reported above) feels not only like a bait-and-switch/false-advertising, but a constraint based solely on driving sales. It does demand a clear explanation.

I care less about another potential Pixel class action, and more that I have to get another phone to test and deploy my apps to a smaller audience to.

The two phones might have similar or even identical memory chips, but that doesn't mean the carving of the address space is the same. A more meaningful comparison would be looking at how much of those 8GB are pinned by the various components (kernel, graphic buffers, sound, camera, telephony, radios, etc.), how much is left to user and system apps, how the system is tuned for active/background processes. 8GB is just a single data point, too simplistic to draw any even remotely plausible conclusions.

I for one am shocked, shocked, to see Google fail at basic product and messaging strategy.

Reply on news.ycombinator.com