https://spectrum.ieee.org/huge-ssds
At the SC24 supercomputing conference held in November in Atlanta, Hafþór Júlíus Björnsson (the actor who played The Mountain in Game of Thrones), deadlifted a custom barbell weighed down by 453 kilograms (1000 pounds) of solid state drives. The data stored in those drives totaled just over 280 petabytes.
"Without question, this is the most data lifted by a human in history," says Andy Higginbotham, senior director of business development at Phison Electronics. High-performance data platform VDURA orchestrated the record-breaking lift in collaboration with Phison.
Behind this publicity stunt is a real trend—to feed AI's insatiable data appetite, memory drives are getting larger, with no end in sight. Phison recently announced the largest SSD memory drive to date, storing 128 terabytes of data, and piled hundreds of them into Björnsson's barbell. Within a few weeks, Solidigm announced its own 123 Tb drive. Samsung and Western Digital also recently started carrying similar products.
The shift towards more AI workloads in data centers has led to very power-hungry chips, mostly GPUs. Since the overall power use in a data center is going up, people are looking for ways to use less power wherever possible. At the same time, large language models and other AI models require ever increasing amounts of memory.
"You can see where storage requirements are going," says Roger Corell, senior director of AI and leadership marketing at Solidigm. "You look at a large language model just a couple years ago, you had a half a petabyte per rack or lower. And now there's large language models that pair with between three and three and a half petabytes per rack. Storage efficiency to enable continued scaling of AI infrastructure is really, really important."
Crucially, this new crop of solid-state drives takes up the same area in a computing rackand power budget as their roughly 32 Tb and 64 Tb predecessors—although they are slightly taller—meaning they can be swapped into data centers for an easy win.
(Score: 4, Touché) by donkeyhotay on Saturday January 25, @12:59AM (5 children)
So much power and memory, and the result is generated imagery of humans with three arms and every other word is misspelled or written in some kind of non-existent alphabet. How can we trust anything AI produces?
(Score: 0) by Anonymous Coward on Saturday January 25, @03:06AM
Don't forget kiddie porn...
(Score: 0) by Anonymous Coward on Saturday January 25, @03:40AM
Well, it's getting nuclear power back online, so that's something. Right? I think?
(Score: 1) by Chromium_One on Saturday January 25, @07:26AM (1 child)
You're out of date. There are multiple "open" image generation models that can do halfway decent text now! Still can fall down on things like hands though.
But, those are tiny (not seen any over 30GB as of yet) in comparison to the more ambitions text generation and (attempts at) reasoning models. Consider Deepseek v3, which I believe is the current largest "open" model. Damn thing is 671 billion parameters and about 0.7 petabytes! By the time you add context memory, the hardware requirements are more than slightly obscene compared to what most orgs would need for anything else. And, of course, even the larger / "better" closed models can still fall down on lots of logic questions.
When you live in a sick society, everything you do is wrong.
(Score: 1) by Chromium_One on Saturday January 25, @07:34AM
(And I am having that moment. TB not PB!)
When you live in a sick society, everything you do is wrong.
(Score: 3, Funny) by FunkyLich on Monday January 27, @08:46AM
That's what you deserve for not being able to fit it within 640kB!
(Score: 2) by damnbunni on Saturday January 25, @01:47PM (2 children)
Are they calling SSDs "memory drives" now? I've not heard that.
I thought this was going to be about drives with more cache RAM or something.
(Score: 2, Interesting) by Chromium_One on Sunday January 26, @12:33AM
Seems every chunk of the tech industry that develops past a certain point wants to develop their own terms to cover things that are adequately covered already. (See Microsoft spending decades ignoring what everyone else was doing or saying to reinvent the wheel every other week, very often doing so in a far worse fashion).
When you live in a sick society, everything you do is wrong.
(Score: 2) by Freeman on Monday January 27, @05:41PM
Ugh, it's all "flash memory" and they're technically not wrong. The best type of not being wrong.
Joshua 1:9 "Be strong and of a good courage; be not afraid, neither be thou dismayed: for the Lord thy God is with thee"
(Score: 2) by VLM on Sunday January 26, @04:31PM (1 child)
I don't watch streaming slop, so I never got into game of thrones, but the unusual name was memorable and I've seen it before, turns out he's held the world record a hair over 1100 pounds. 1000 is about 10% less than his successful world record attempt so its "a lot" but not a new record. The world record goes up a few pounds every few years so he's probably not the record holder anymore.
Maybe if they set all the hard drive bits to "0" the drives would be heavier. Seems reasonable because punch cards are lighter the more "1" bits they have for obvious reasons.
I'm way too lazy on this slow Sunday to calculate it, but energy is mass and entropy is energy so changing the patterns stored on HDD should vary its weight. Obviously changing patterns on SSD would vary its weight more directly as the electrons stored (or not stored) on the capacitors have an obvious mass, although not much.
280 petabytes would be 2 or so exabits or 2e18 bits. Some handwaving that a flash cell at one point in history might have held 400 electrons according to google; seems plausible; 0.8 Zetta-electrons aka 0.8e21 electrons. Electron mass is arount 9e-28 grams so figure filling up that 1000 pounds of SSD would increase its weight by somewhat less than a millionth of a gram. Donno if thats impressive or not or if I made a "pre-tea drinking" math mistake this morning.
(Score: 1) by Chromium_One on Sunday January 26, @11:38PM
Vaguely recall an article some years ago about just those sorts of measurements being done on HDDs. Failing to remember if SSDs were done as well, and not tracking down the reference now. Yes, the difference was enough decimal places down that it was a *process* to reduce error bar on measurements enough.
When you live in a sick society, everything you do is wrong.