The Register Home Page

back to article Bad news for OpenClaw stans: Apple’s Mac Mini now starts at $799

The Mac Mini is the latest victim of the AI-fueled RAM-pocalypse. Last week, Apple discontinued the 256 GB version of the system, which cost $599. To get in now, you'll need to drop at least $799 on a 512 GB version. The price hike makes the Mac Mini Apple's most affordable Mac no longer. That distinction now goes to the $599 …

  1. williamyf Silver badge

    IIRC, Mac minis do not come with 256 or 512GB of RAM. They come with 256 or 512GB NAND Flash.

    Since NAND Flash != DRAM AND NAND Flash and DRAM are made in different production lines (i.e. production lines are NOT interchangeable between them), blaming this price increase on the "RAM-pocalypse" in the first sentence of the article is plain wrong, when what changed was the storage.

    While most people conflate DRAM and NAND Flash as "Memory", this is "ElReg". Readers here know better, so, reporters should know better too.

    1. Hurn

      Was just about to post that:

      "RAM-pocalypse. Last week, Apple discontinued the 256 GB version..."

      RAM is not NAND Flash. While both are expensive these days, they're different.

      Thanks for getting there first.

    2. Spazturtle Silver badge

      Yeah the Mac Studio was the only Mac with 512GB of RAM, and they no longer sell that option since 512GB of RAM now costs over $10k which is more than Apple were charging for the whole Mac Studio.

    3. Dan 55 Silver badge

      It doesn't matter, this AI nonsense is affecting DRAM, NAND flash storage, and even spinning rust prices because all three are needed by AI datacentres. It all comes under the RAMpocalypse term.

    4. Rich 2 Silver badge

      While you are not wrong, flash prices have doubled this year - not quite the x4 increase of DRAM but still

    5. DS999 Silver badge

      NAND is also having an apocalpyse

      It just isn't making the headlines as much.

    6. Bill Stewart

      They didn't say it WAS RAM.

      As another person replied, El Reg is using the term RAMpocalypse to refer to this whole AI-driven mess, which hits RAM, SSD, and even spinning rust HDDs.

      And yes, Apple is dropping the lower-storage M4 Mini model so you have to buy the next size up, an extra $200 for $50 of storage that used to be an extra $20, instead of changing their prices.

      They're also running out of stock on those, and rumors are they might release the M5 mini next month if they're ready.

      They ARE also dropping the 128GB RAM model of the Mac Studio, their high-end desktop, which now maxes out at 96GB RAM.

  2. Anonymous Coward
    Anonymous Coward

    both RAM

    in the 80s we just used the term SRAM or DRAM, both RAM. NAND is just a form of SRAM. like NOR and others. both prices have gone mad. even a 2nd hand 128GB NGFF PCI or SATA is now 5x the price of 6 months ago. those got nothing to do with AI other than the supply crunch. sooner or later we will see DIY DDR and SSD stuff becoming a thing. I started building my own SSD drives, for fun. AI is very useful but the way it divides society right now isnt right.

    1. williamyf Silver badge

      Re: both RAM

      Flash (NOR Flash in particular) is more akin to EEPROM than to RAM.

    2. williamyf Silver badge

      Re: both RAM

      When I answered Anon Above, I was out and About, so let me Elaborate:

      1.) In the '80s SRAM and DRAM would lose all content if power went out, meanwhile ROM, PROM, EPROM and EEPROM would keep their data for years on end without power, just like NAND/NOR Flash would do today.

      1.1) ¿Do you know what else did not lose the data when the power went out? HDDs

      2.) In the 80s you could modify SRAM and DRAM in 8, 16 or 32 Bits without erasing (modify in place). Nowadays, the size is 64 Bits, meanwhile EPROM and EEPROM, you needed to erase first and modify latter, and the Erase window was the whole chip's contents (and in the case of EPROMS, there was a literal quartz window involved). Meanwhile, modern QLC NAND Flash has Erase Blocks of 48~96MB and NOR Flash has Erasure Blocks of 64KB~256KB

      3.) in the '80s you could read SRAM and DRAM in 8, 16 or 32 Bits, Nowadays , the size is 64 Bits. Meanwhile NAND Flash has to be read ~16KB at a time, but NOR Flash can be read 8~64Bits at a time. You know what else could be read 8~64 bits at a time, ROMs, PROMs, EPROMs and EEPROMS.

      4.) ¿You know what technology in the '80s had to be read and written in large chunks (Blocks), HDDs, that's why they were called "Block Devices".

      So, as you see, DRAM and SRAM are more similar to current DDR5 (or GDDR, or HBM) than they are to Flash. And Flash is more similar to ROM, PROM, EPROM and EEPROM (and to HDDs) than to DRAM or SRAM (or the modern equivalents listed).

      So, in a word no. If NAND (or NOR) Flash gets more expensive, is not the fault of the RAM-Pocalypse. If HDDs get more expensive, is not the fault of the RAM-pocalypse either. If processors get more expensive, is not the Fault of the RAM-pocalypse, ditto for BMCs, Power Delivery circuits, and other stuff needed for Large servers.

      Is the AI boom that is making RAM more Expensive, and that RAM-pocalypse is driving higher the prices of everything that uses RAM, but NOR and NAND Flash barely use any RAM, so their price increase is not due to the RAM-pocalypse, is due to the AI boom. So, the people criticizing the author are correct...

      As per your idea of DIY RAM and SSDs, in the current climate, the cost of the actuall RAM and Flash chips is the bulk of the cost increase, the PCB and ancillary components for the (SO)DIMMs or M.2 PCBs and controllers are peanuts in comparison, so ¿Where do you plan to get cheap naked DDR or NAND Flash chips for your DIY stuff? Desoldering them from discarded (SO)DIMMs and M.2 Drives and Resoldering them DIY Style in a more dense configuration is the only way forward for that "brilliant" idea... Good luck with that.

      (and yes, I am aware if the russians who designed their own DIMM PCBs for DIY RAM Sticks, maybe ask them where to get inexpensive chips)

  3. sebacoustic

    raspi

    News: some Raspberry Pi now more expensive than a Mac mini

    Apple: wait, what? We'll see about that!

    1. Anonymous Coward
      Anonymous Coward

      Re: raspi

      Mac Neo …. Inbound.

  4. Rich 2 Silver badge

    AI

    Are people really buying these things primarily for “AI” (as the article suggests) or are they buying them to do …you know …normal useful stuff?

    1. wolfetone Silver badge

      Re: AI

      I was considering buying a Mac Mini to replace the Intel i5 one I've had since 2013. I use mine now more for Garageband and scanning negatives, so thought this Mini would be a reasonable upgrade.

      When it was cheap. Now? No.

      1. Steve Davies 3 Silver badge

        Re: When it was cheap. Now? No.

        There are plenty of slightly used ones with slightly older M series CPU's available. You don't have to buy new kit all the time. I recently purchased a used 14in MBP with an M5 cpu for £500 less that retail.

        1. wolfetone Silver badge

          Re: When it was cheap. Now? No.

          I know, and to be honest post-COVID I got burned with a few car purchases which have tainted my mind in regards to second hand stuff, especially when it's north of £100 I'm spending on it.

          But I think sooner or later I'll pull the trigger on an M3 Mini, given that they run LLMs fairly well locally for less than the cost of an nVidia graphics card.

          1. Lazlo Woodbine Silver badge

            Re: When it was cheap. Now? No.

            I don't think Apple ever made a Mac Mini with the M3 chips.

            I've got a base M1 model and it's getting a little bogged down with some of my tougher workflows, so I'm thinking of buying an M4, the £200 price jump for the new base model albeit with double the SSD, is a little off putting, but I'll have a redundancy cheque soon, so I may just take the leap...

    2. doublelayer Silver badge

      Re: AI

      Probably those buying them for LLM usage are buying the higher-priced versions with even more RAM. Models with 16 GB can run some small models quite quickly, but for people who want to run bigger ones or more tasks through them, more RAM speeds it up a lot. Then again, some OpenClaw fans were running that on a Raspberry Pi, so for anyone in that camp, the base Mini is sufficient for those purposes.

      1. MachDiamond Silver badge

        Re: AI

        "Models with 16 GB can run some small models quite quickly, but for people who want to run bigger ones or more tasks through them, more RAM speeds it up a lot. "

        Apple is very proud of their RAM and charge better than top dollar for it. Now that it's soldered on, there's no buying the base model and sourcing RAM elsewhere to save a few thousand.

      2. williamyf Silver badge

        Re: AI

        Actually, while OpenClaw runs the agent locally, the "BackEnd" LLM can run locally or externally (in a server you own or in the Datcenter of Anthropic or OpenAI).

        People wanting to run OpenClaw Completely localy will need a powerfull machine (say, Mac Studio(s), AMD Strix Halo(s), or nVIDIA DG10(s)). But people running the model using an LLM on someone else's cloud can make do with a Raspberry Pi or a Mac mini for locally running the Agent.

        This is what OpenClaw themselves have to say about running the models locally:

        «Local is doable, but OpenClaw expects large context + strong defenses against prompt injection. Small cards truncate context and leak safety. Aim high: ≥2 maxed-out Mac Studios or equivalent GPU rig.»

        https://docs.openclaw.ai/gateway/local-models

        For Reference, a fully decked Mac Studio has 256GB of RAM and can be clustered 4x, AMD's Strix Halo has 128GB of RAM and can be clustered 4x, nVIDIA's DG10 has 128GB of RAM and can be clustered 2x (but has the most raw compute power for AI).

        Oh, and by the way, Apple and AMD can instantiate 2 different models (or 2 instances of the same model) without fighting for compute, one in the GPU and the other in the NPU, meanwhile, the DG10 can only instantitate one model (it lacks an NPU). If you are testing various local LLMs in order to choose "the one model" to run with your OpenClaw, this can significantly accelerate your testing.

    3. Marty McFly Silver badge
      Gimp

      Re: AI

      >…normal useful stuff?

      I have a bunch of Mac Mini's, most are running TV displays. Sure, they cost more than a Roku or Apple TV or the Smart TV with their built-in spyware. But I can apply my own privacy and advertising controls to what I watch. That value is worth it to me. And I submit that most people don't know they are selling their privacy so cheaply. If you want an eye opener, monitor the traffic from your "Smart" TV and ask yourself why it is so talkative to unknown outside servers.

      Oldest is a 2011 i7 former Mac OS Server. It gets the job done, but will probably be next to be replaced when the M5 Mini debuts. And to be completely transparent, if possible I will to purchase a Made in USA Mac Mini too.

      Yeah, okay, fanboi icon. The thing is I did not intentionally start down this path. It just kind of happened with the 2011 Mini and slowly grew from there. Mac's are now my default go-to, Linux second on my network, and Windows in last place.

      1. MachDiamond Silver badge

        Re: AI

        "Oldest is a 2011 i7 former Mac OS Server."

        I've got a 2008 Mini that runs my accounting software. It's bolted under a desk and runs headless. If I spot another one for $20-$40, I'll get it as a spare. My Mac Pro 4,1 is my media server that I think runs High Sierra so I have a flavor of iTunes that mates with my iPods the best. It's also a headless machine that I remote into with my iMac ($25 + $75 to fetch back). A lot of older gear can be perfectly fine for many tasks. Once I get a CNC router built and going, it will be run with an old PC (Win7 likely). It won't need anything more powerful than that. Being a win machine, it will not be on the network. I have plans to stick an iMac in the garage so I can look things up and can use that to store router files I create in the office to use SneakerNet as a way to get files into the CNC. I suppose I could run another network that doesn't connect to the internet.

  5. NerryTutkins

    As a developer, I'm focused on Android but looking at iOS if I need to expand as I have a codebase on Maui that I can work for iOS too. The downside is needing a Mac to build the package and an iphone to test on.

    The cheapest mac option is now the Neo which I looked at in a local store. About 600 EUR which seems quite reasonable. The build quality looks ok, and the specs are not terrible.

    The costly part is the phone. Even several-year-old second hand iphones here cost more than that.

    The iphone costs are just insane. There really isn't any model in their range where the price is even remotely reasonable in comparison to even good Android brands. Samsung has offerings from top to bottom, for example.

    1. Excused Boots Silver badge

      "The iphone costs are just insane. There really isn't any model in their range where the price is even remotely reasonable in comparison to even good Android brands. Samsung has offerings from top to bottom, for example.”

      It depends on what you mean by ‘reasonable’. They cost what they do because people want to buy them and are prepared to pay that price.Therefore those prices must be ‘reasonable!’

      Anyhow if you want to develop for iOS that that might be just the ‘cost of doing business’, pay it or not. But although I believe you absolutely have to have a Mac for building , there are iOS emulators available to test it on. Yes, probably not quite as reliable as actually running your app on a physical device - but might suffice.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon