AFAIK this has already been a problem, you can find Samsung M.2 SSDs for cheaper than Samsung SATA SSDs at the same capacity, because their cloud customers have all flown past classic SATA/SAS for NVME U.2 and U.3, which is much more similar to M.2 due to NVME.
I was planning on adding a big SSD array to my server which has a bunch of external 2.5 SAS slots, but it ended up being cheaper and faster to buy a 4 slot M.2 PCIe card and buy 4 M.2 drives instead.
Putting it on a x16 PCIe slot gives me 4 lanes per drive with bifurication, which gets me the advertised maximum possible speed on PCIe 4.
Whether or not the RAM surge will affect chip production capacity is the real issue. It seems all 3 OEMs could effectively reduce capacity for all other components after slugging billions of dollars into HBM RAM. It wouldn’t just be SSDs, anything that relies on the same supply chain could be heavily affected.
Exactly this. Micron ended their consumer RAM. Sansung here is just stopping producing something that is arguably outdated, and has a perfectly fine, already more available, most often cheaper or equivalent modern replacement.
Aside: WTF are they using SSDs for?
LLM inference in the cloud is basically only done in VRAM. Rarely stale K/V cache is cached in RAM, but new attention architectures should minimize that. Large scale training, contrary to popular belief, is a pretty rare event most data centers and businesses are incapable of.
…So what do they do with so much flash storage!? Is it literally just FOMO server buying?
Storage. There aren’t enough hard drives, so datacentres are also buying up SSDs, since it’s needed to store training data.
since it’s needed to store training data.
Again, I don’t buy this. The training data isn’t actually that big, nor is training done on such a huge scale so frequently.
As we approach the theoretical error rate limit for LLMs, as proven in the 2020 research paper by OpenAI and corrected by the 2022 paper by Deepmind, the required training and power costs rise to infinity.
In addition to that, the companies might have many different nearly identical datasets to try to achieve different outcomes.
Things like books and wikipedia pages aren’t that bad, wikipedia itself compressed is only 25GB, maybe a few hundred petabytes could store most of these items, but images and videos are also valid training data and that’s much larger, and then there is readable code. On top of that, all user inputs have to be stored to reference them again later if the chatbot offers that service.
The lines used to produce vram also do ssd nand flash, so they make less ssds to make more vram
Glad that I recently bought a bunch of storage so that I’ll be covered for a good amount of time.
Bought a new pc in October and I’m soooo glad I did
Who needs computers anyways?
Just go outside and play with real people
Crap, I really wanted to buy a new external HD for my home server setup sometime soon.
HDDs should be fine. No?
Until next month when they cancel those too
I suppose so, it just takes forever to transfer files.
Syrup of Squill.
glad i recently bought an SSD. it was expensive, and it will be more expensive in the future, damn :(
I wonder what changed, prices were being driven down on SSDs for a while there
Put a 1tb 850 Evo in our PS4 years ago for a pretty reasonable price. Kind of expected prices to continue to fall back then
AI happened. It requires an immense amount of RAM and storage for data centers, and manufacturers don’t have enough production capacity to keep up.
Prices were going down because the consumer market didn’t have as much of a demand for both after covid.
Well. I just FOMO ordered a SATA SSD. Thanks, OP.
Mostly because I got 2x64Gb sticks 2 months before they shot up to nearly 4X the price! And it doesn’t feel like this bump is going away soon.
I got an old Nitro 5 with a rickity old 500gig hard drive. Will a Crucial BX500 1TB 3D NAND SATA 2.5-Inch Internal SSD be a good Christmas present for it?
Probably should get something while prices are somewhat more reasonable.
I remember when 8TB SATA SSD was $350
pepperidge farm remembers
sure grandma, lets get you to bed
This bubble is going to become the entire market, isn’t it. Until it becomes too big to fail because 80% of the workforce is tied up in it. Then it is allowed to pop, costing the western world everything, all going into the pockets of the super rich, and we get to start over.
That’s the entire point. It’s a scam.
Then it is allowed to pop, costing the western world everything, all going into the pockets of the super rich, and we get to start over.
After the bailouts at the expense of the poor, of course.
it becomes too big to fail because 80% of the workforce is tied up in it
In 2008, banking sector and auto industry needed bailouts for the investor/financial class. Certainly, there was no need to layoff core banking employees, if government support was the last resort to keep the doors open AND gain controlling stake over future banking profitablity in a hopefully sustainable (low risk in addition to low climate/global destruction) fashion. The auto bailout did have harsher terms than the banking bailout, and recessions definitely harm the sector, but the bailouts were definitely focused on the executives/shareholders who have access to political friendships that result in gifts instead of truly needed lifelines, or wider redistribution of benefits from sustainable business.
The point, is that workforce is a “talking point” with no actual relevance in bailouts/too big to fail. That entire stock market wealth is concentrated in the sector, and that we have to all give them the rest of our money (and militarist backed surveillance freedom) or “China will win” at the only sector we pretend to have a competitive chance in, is why our establishment needs another “too big to fail moment”. We’ve started QE ahead of the crash this time.
Work force is relatively small in AI sector. Big construction, but relatively low operations employment. It displaces other hiring too.
the shoe event horizon.
AI has taken more things since it’s big push to be adopted in the public sector.
Clean Air
Water
Fair electricity bills
Ram
GPUs
SSDs
Jobs
Other people’s art and writing.
There are no benefit to this stuff. It is just grifting.
Also free and fair elections. Fidesz published a clearly AI-generated document claiming it was a leak from current oppposition party Tisza, as a real program.
true
Yet another chapter in the fucking AI craze started up by them fucking techbros.
Also, someone forgot that in some places in the world, people have to use older PCs with SATA drives. That, until their discontinuation announcements, Crucial and Samsung SATA drives were several tiers better than, say, those cheapo Ramsta drives.
Discontinuing outdated tech has nothing to do with AI. SATA SSDs need to be retired. NVME is superior and widely available.
Especially since you can get M.2 to SATA adapters, so people stuck with SATA only motherboards can still upgrade their storage.
Literally the same deal when companies stopped making IDE drives, people just used SATA to IDE adapters instead.
Do you know of any m.2 to SATA adapters that support NVMe? Or are these only for Sata M.2s?
Man, it sure would be helpful for my argument if I could.
I went back and checked the ones I was looking at, very helpful fine print stating “not for NVEM ssds”, so they all only work with mSATA M.2 SSDs, hell of a let down.
Damn… I was hoping you were aware of things that I had missed…
Anyway, if you run out of M.2 NVMe slots on your motherboard there are still options, since NVMe is just PCIe:
https://www.amazon.com/Adapter-Converter-Reader-Expansion-Internal/dp/B0BK2R7T57
Despite how similar the interface is the protocol is completely different. NVMe is basically just PCIe, so adaptating it so that it runs “under” SATA would be difficult if not almost impossible. And most definitely not worth the extra price.
Well no you can translate, but it just seems that nobody has actually made a product to do so.
e.g. those M.2 SSD to USB adapters, those aren’t speaking NVMe to the host device. They either talk the traditional USB “bulk transfer” protocol, or potentially SCSI, translating that to NVMe for the SSD itself.
Can we just burst this damn AI bubble already?
When I built a PC a couple of years ago when I really didn’t need one, then over specced it just because. I’m very happy right now as the prices are insane, feel like I could sell the PC for more than it cost me which mental.
