Okok thanks, I did read that, good reminder.
We've been able to do that since my iBook G4, if the marketing schlock was to be believed.
The statement was made during an earnings call. Even a lie of omission can have serious consequences.
The products you sell a lot of tend to have more predictable sales patterns and you worry less about being stuck with unused product, so you tend to build and warehouse more of them. That’s particularly true of chips because spinning up a run of a chip is not something you do lightly - it’s much more economical to do fewer large runs than multiple small runs. So even though Apple is just-in-time manufacturing the whole widget, they’re likely prestocking chips for months worth of computers on each fab run. That means Apple almost certainly had more low RAM package chips on hand when the demand both accelerated overall and skewed towards higher RAM packages. Put another way, “demand surges tend to hit low volume products harder.”
https://www.bbc.com/news/articles/c232z4yk437o
Quote pulled for emphasis.
There's also the issue that the base M4 is used for multiple current-generation products beyond the Mini (iPad Air, iMac), but the Pro and Max variants are now only used in the high end Mini and low end Studio. So, Apple presumably has regular production runs of the M4 scheduled on an ongoing basis and will keep doing so until updates for the iPad and iMac are near at hand. Balancing that supply with increased demand for base model Minis is a manageable problem. Scheduling an entirely new fab run for a new batch of M4 Pros and Maxes because high end Minis and base Studios have unexpected heavier demand is a much harder proposition. And the M3 Ultra would be even worse.
I hope so. I am just extremely intimidated by how many thousands it might be. My money put aside is suddenly feeling insufficient.
It's certainly interesting, and one can imagine some uses that primarily cater to user convenience. Key is recognizing that Nvidia's Blackwell Pro line – their newest and most well-spec'd GPUs sold as add-in cards – top out at 96GB of RAM. That's a quantity which Apple's current desktops and laptops meet or exceed.
Do you have any real benchmarks showing the Mac beating NVIDIA cards? The only stuff I can find running actual benchmarks all show NVIDIA handily beating the Mac like this one.
It's not about outperforming Nvidia cards (some results I've seen show Nvidia far ahead on prefill, but falling behind Apple on token generation), it's whether you can load and run very large models at all.
This is what I don't get. Historically, computers did workloads in one of two camps: Instant, and batch. And yeah, I see a lot of numberwanking in the LLM subreddit about guys with personal servers and 4 6000 gpus saying they just aren't enough.
Thank you for your sacrifice. This guarantees M5 Studios, coming soon!!
Qwen3.6 35B-A3B mostly for local, sonnet and haiku for the stuff that needs reasoning.
