245TB Micron 6600 ION Data Center SSD Now Shipping(investors.micron.com) |
245TB Micron 6600 ION Data Center SSD Now Shipping(investors.micron.com) |
EDIT- The same 2TB ssd is now $329.99 at bestbuy.
That's a bit off the mark. Processors, RAM, and flash memory each require their own specialized fabs. TSMC makes processors but not DRAM or NAND flash. Kioxia makes flash only. Samsung has all three types of fabs. Micron does DRAM and NAND, in different fabs.
The increase in SSD prices is not because there are many components competing for the same constrained resources, it's that they are complementary goods being subject to the same dynamics in parallel because the servers that are causing this demand spike need all of those components. Where we do see competition for the same fab capacity is in the mix of DRAM types, where GPUs want HBM, server CPUs want DDR5, laptops and phones (and the occasional server CPU) want LPDDR. And competition among different types of processors for TSMC's fab capacity.
The pandemic caused once such boom-bust that resulted in a rather large downturn in demand in 2022-2023 referred to as the pandemic hangover. During that time demand dropped following overspend during the pandemic and members of the cartel drastically cut production at times to keep prices above cost. Even after the demand recovery began in 2023, the cartel members were slow to increase production and made little to no investment in production capacity in 2024-2025. Creating a shortage.
The AI hype cycle has exacerbated the shortage by creating speculative purchases and then panic buying. Remember the shoe company that pivoted to AI?
So Cartel market manipulation is partially to blame for the over 100% increase in prices and the shortages.
The reality is all the things we learned from the Great Depression have been forgotten, and half the voting public seems to champion monopolies while simultaneously acting dumbfounded that the end result will always be an increase in costs.
NIMBYs and degrowth activists own this particular "market failure". The companies have been trying to invest in capacity ahead of demand but have been prevented from doing so.
Last half year ate up three to four years of earlier price regression, that's about it.
As long as this plateaus here, as prices did for last 4 months, that's just the new equilibrium where it has the chance to get better again, would no be all doom and gloom about personal computing yet.
The doomerism surrounding this topic is wild.
Gas prices have also gone up recently, but I don't see the same claims about how it's the end of personal vehicles, that the prices are never coming down, or that we're all becoming renters instead of buyers.
Saying that there isn't a force countering this is peak doomerism. New factories take a long time to build. These companies didn't have spare production lines waiting for demand to go up.
It's intuitively obvious to a lot of people that the era of personal, wholly owned transportation is waning. A lot of people seem to miss the second clause of that old "you'll own nothing" phase, the part where most people are happy about it!
When vehicles drive themselves, and there's a large enough pool that one can show up pretty reliably within a few minutes of your needing one, how many people are going to choose to own when renting is cheaper and easier?
I remember selling some ancient GPUs that I was going to throw away during one of the previous booms.
I also thought about upgrading my gen4 SSD at the same time. It was still meeting all my needs, so I thought if I waited, I could get a bigger, faster one for cheap later. Mistake!
Things will normalize, semis are a boom and bust industry, and it takes a massive amount of capital investment to increase supply. Because of that boom and bust, the producers are wary of growing their supply to keep up in lockstep with demand, because they might end up with a glut that they have a hard time selling, and a large pile of debt from building out that glut. Semiconductors at the leading edge take very large, very expensive, very rube goldbergian machines to make them, and a very skilled workforce to make those work.
If this stuff stays this expensive/profitable, industry will grow capacity to boost earnings until they overshoot, there's not some grand conspiracy to make you rent everything. We're just in a massive dislocation right now because we've discovered this awesome new tool that's boosting demand by a ridiculous degree. Last year we were at ~$300B of capital investment in DCs in the US, this year it's looking like ~$750B.
Production cartels exist to prevent individual greed from "ruining" it for the rest of the group. OPEC does it in the overtly, memory manufactures do it covertly, and have been convicted of conspiring in multiple jurisdictions, on multiple occasions.
Basically I like to use them as a cheap “better” thumb drive, putting ventoy on them with no concern for space, along with a fat32 partition for bios updates.
These along with a nanoKVM box my siblings can attach to the computers I built for them makes remote troubleshooting relatively painless when rustdesk isn’t sufficient.
Same a consumers (car drivers) subsidizing businesses (trucks). Road damage is almost entirely caused by trucks, 4th power of weight, but the cost is borne by us.
This is the first time I've had a computer have a complete failure and no way to fix it as I look around at the used market and the latest generation market. I'm kind of just disillusioned.
what happens is that the industry goes through certain discrete technical upgrades. for instance, EUV in fabs, or GMR disk heads. none of these are really planned, none of them are exponential. and they usually interact with other phenomena (such as Dennard scaling).
in a sense, the phenomenon is more like "expectations are exponential, and this motivates manufacturers to schedule updates".
hard disks are still improving, arguably similar to how they have in the past, but there are limits to demand. the consumer market has mostly dropped out, for instance due to flash.
even in flash, there is no exponential scaling in devices. people got excited in the initial startup, when for instance, mature TLC is so much better than early SLC. but all that's over: it's both mature and we'll probably never see PLC. even QLC is interesting in that it illustrates that most of our storage is very cold.
ebbs and flows
Funny thing is that one of the best you can get is the Crucial (Micron) 8TB one but even that one gets more expensive. I have the feeling it will be gone completely soon.
The interface looks equiv to 4x PCIe 5.0.
> Sequential read (MB/s): 13,700
> Sequential write (MB/s): 2,700
The same EBS GP3 used to be specified with 16K max IOPS at 16 KiB random transfers until pretty recently.
Also: price is expected to be $80k. I suppose density is the selling point here, not speed.
Rather silly of them to hide investor relations material behind an anonymity-hostile CDN.
PDF for those who want it. https://web.archive.org/web/20260506084407if_/https://invest...
You don't have permission to access
"http://investors.micron.com/news-releases/news-release-detai..." on this server.
High security on this press release.
I haven't bought a hard drive or an SSD in at least a decade (I get stuff for free, basically) but…that seems a bit high, right?
Seems like well-rated consumer-level SSDs cost around $250 for 1TB right now.
What accounts for the premium price/TB of these extremely high capacity enterprise-targeted drives?
Dell is getting first dibs.
How big of a deal is this part in relation to the initial upfront costs? I’m not privy to the cost of power for SSD
Just imagine something like 2S 9565 in at least 2U machines: with 10 server x 2U x 2 CPU you would have 8kW in the processors alone and you didn't even fill half of the 42U rack.
https://www.amd.com/en/products/processors/server/epyc/9005-...
Getting rid of 30 watts of heat is trivial compared to say, 300 (I don't quite know how to read that ratio since a 2.5kW SSD seems a little high to me).
Fewer servers means fewer cpus, less RAM, fewer fans, and maybe fewer switches.
https://www.storagereview.com/news/245tb-kioxia-lc9-ssd-sets...
Furthermore, 15-60x density improvement reduces server and infrastructure costs because it requires vastly less of everything per EiB.
Cooling and power are the limiting factors of density.
So it's not exactly about cost savings, but having the option to do more, faster.
Also, you could also get much higher bandwidth density out of this vs HDD, and this is great for AI training
The datasheet shows 3GB/s sequential write, which for 245.76TB means writing the whole drive takes around 22h45m. Odd that the endurance is specified as "1.0 SDWPD", which is almost meaningless since the drive takes roughly that long to write at full speed.
At scale, 1.9 times more energy is required for an HDD deployment
...but those HDDs are going to hold data for far more than twice as long. It's especially infuriating to see such secrecy and vagueness around the real endurance/retention characteristics for SSDs as expensive as these.
On the other hand, 60TB of SLC for the same price would probably be a great deal.
Those drives aren't going to be used for cold storage, and it is basically a guarantee that there will be checksums and some form of redundancy. Who cares whether the data is retained for 10 or for 15 years after writing when you can do a low-priority background scrub of the entire drive once a month, and when there are already mechanisms in place to account for full-drive failure?
Very cool bit of tech.
* October 2020: around 200 EUR
* May 2026: around 230 EUR
But if you're buying a $500k GPU server putting 100TB of nvme in there for $50-100k is justifiable.
Nowadays it feels like that this time and price region is like decades away in the future. I was hoping I can store more data in future on modern tech like SSDs and not less.
The old SATA SSD form factor is dead and wont come back.
OWC ThunderBlades exist, but 32TB will set you back 9 grand.
You should be able to assemble something with USB-C for under 5k. That's not a mass consumer market thing, but perfectly doable, if your use case warrants it. We are stuck with 2021 pricing, but now with options of 8TB per NVMe drive at way higher speed.
36TB+ HHD external WD drive combos were always around EUR 1000 over last 5 years. With a short low end around EUR 600 in 2023
https://www.owc.com/solutions/thunderblade?sku=OWCTB3TBL8X32
That's more expensive than buying a single large capacity drive. It's also a terrible idea. I would never trust a low cost chinese controller with terabytes of my data.
>The old SATA SSD form factor is dead and wont come back.
That's true and very unfortunate.
there is no sign of any impending breakthroughs that would change flash economics much.
slc-mlc-tlc-qlc was very nice but plc will not happen. layer-based flash was also nice but it is ultimately linear (more layers, more cost, lower yield). dimensional shrinks are already stalled because of a tragic electron shortage (per cell).
I guess there's no harm in pining for some other NVRAM technology (spins, etc).
Despite the recent AI bubble you can still buy HDDs in the tens of TBs for a few hundred EUR/USD and you still don't see them in every computer. How high could the 30TB SSD demand be to justify the kind of volumes that drive price down?
In the DC it's the opposite, large and efficient drives are a must to save support all those fancy workloads while driving down space, power, cooling needs.
With regards to the new Micron SSD - I wonder how they keep it cool? I don't see coolant ports on it so they must strap a heatsink on.
The average user consumes that much quite regularly. They've been taught to stream it off of someone else's computer, mostly so that the next time they stream it they can be compelled to pay for it again. It's fun going back to dumb terminals.
This is all fine and even desirable for a consumer who will only be writing for at most a minute or two, but with a 245TB server drive you need to assume the performance will be needed constantly. They target sustainable and predictable performance.
What should worry way more is DWPD which is abysmal... on the first glance. But if you punch it in the calc it still would take ages to wear it out.
SSD #1 SSD #2 SSD #3
Capacity (GB) 245000 245000 245000
Warranty (yr.) 3 3 3
DWPD 0.3 1 0.075
TBW (TB) 80482 268275 20121
TBW (PB) 80.483 268.275 20.121
PBW 80.483 268.275 20.121
GB/day 73500 245000 18375
Time period Average host-side write data rate (MB/s) needed for reaching DWPD value within specified time period
8 hr. 2552.08 8506.94 638.02
12 hr. 1701.39 5671.30 425.35
24 hr. 850.69 2835.65 212.67
now, nobody cares. I have over 500 NVMe drives in our deployment and the drive deaths are not due to wear.
"U.2" does not change anything in the mechanical characteristics of a 2.5" drive, it just replaces the SATA or SAS electrical interface with a NVMe electrical interface.
You can mount a U.2 drive in any location intended for 2.5" drives, as long as its height can fit there.
However, 2.5" drives come in various heights. Many laptops and mini-PCs that accept 2.5" drives accept only some of the smaller heights and they do not accept the greater heights, like 15 mm, which are typical for enterprise SSDs and HDDs, regardless whether they have a NVMe, i.e. U.2, or a SAS interface or a SATA interface.
This new high-capacity U.2 SSD has the standard 15 mm height of the 2.5" form factor.
1. For comparison, an HDD usually comes in around ~10 watts
Now, I would LOVE to see this much SLC flash on a direct to bus attachment setting.
Note that the 245TB is an E3L, the half size version of it come in smaller size.
https://americas.kioxia.com/en-ca/business/ssd/solution/edsf...
https://www.exxactcorp.com/blog/storage/edsff-e1s-e1l-e3s-e3...
https://www.simms.co.uk/tech-talk/e1s-e1l-the-new-server-for...
No need to worry about cooling when each layer in the sandwich is only a fraction of a micrometer thick!
P.S. I had to shuck 20TB WD drives that cost 350EUR on sale (now at 400EUR). 26TB drives are now ~700EUR. These external drives were the cheap option. Standalone drives usually cost more.
Now that's a terrible idea. You shouldn't trust any single device with terabytes of your data, regardless of whether it is low-cost or Chinese.
surely you don't actually think that's realistic pricing?
It is very real, that is the price they quote. You can buy it through Dell at that price.
3.84TB SSD SAS ISE, Read Intensive, up to 24Gbps 512e 2.5in with 3.5in HYB CARR, AG Drive
Dell Price $8,825.13 /ea.
3.84TB SSD SATA Read Intensive 6Gbps 512e 2.5in Hot-plug AG Drive,3.5in HYB CARR, 1 DWPD
Dell Price $7,893.91 /ea.
3.2TB Enterprise NVMe Mixed Use AG Drive U.2 with carrier
Dell Price $6,596.39 /ea.
The counter-example I hold up is Harley-Davidson. They could hardly pay a bill in the late '70 through the '80s. They'd even get free samples from suppliers and put them into production. They survived with very strong brand loyalty from customers and after-market companies.
Samsung sells directly to customers of all sizes, so they have a very good idea as to where products end up. Cutting the legs off of innumerable small businesses is despicable behavior.
I don't care if Samsung is literally starving to death, where I was previously at least a marginally loyal customer not even looking at other brands because prices were fine.
The Market™ is perhaps going after higher margins. If you can charge $200 instead of $100 for a widget, why wouldn't you?
Ideally at some point someone will see the margins and be motivated to go after some of it and offer things for $190, but the ROI on upfront fab costs given market risks may not be high enough for the business uncertainly that needs to be taken (boom-bust cycles).
No participant in The Market is obligated to "cater" to demand if they do not think the juice is worth the squeeze.
Spare capacity, mostly. That’s why they have higher endurance. If you want to double the endurance of a given drive, tell the controller to allocate twice as many spare blocks and report less capacity than you would otherwise.
In this case, you are also paying a premium for the PCIe attachment instead of SAS, and a lot for price elasticity. You see, with drives like these you slash space and energy consumption in relation to HDDs by a large number, and that allows you to pay a premium for the device, because, at the end of its lifetime, it’ll have more than covered the cost difference in saved space and energy.
The word "enterprise".
They’re currently selling for $942.72 on Amazon.
The price graphs of these are wild, wow. https://camelcamelcamel.com/product/B0CHGT1KFJ
Looks like prices for NAND spiked at the end of last year, a couple of months behind DRAM.
Happy not to be building or buying any new computers these days.
The extremely high capacity and the enterprise targeting.
https://www.opencompute.org/documents/datacenter-nvme-ssd-sp...
Why would you want to store every movie or series you watch? 30TB of data is something like 1 year of uninterrupted streaming at average Netflix 4K bitrates. Even more at HD bit rates. How many people would ever store years worth of movies on SSDs no less? Enough for it to drive huge sales in the market?
It isn't. The number of books I've read again, that I've wanted to have my children read. Music. Shows. The old VHS tapes were only disposable because the format was so deteriorating. It was inconvenient because of the physical size.
But, if people realize that media could be non-disposable, how could the recording industry get rich selling you the same albums you'd already bought a half dozen times before? I've been storing everything as FLAC... it'll be good for centuries. Books already have a "good for centuries" file format. I suspect very strongly that 4K is the ultimate res for all media made up until today (no way to do 8k remasters). It doesn't need to be disposable, it just is because there are people who want to get rich off of you renting your own culture back from them.
>How many people would ever store years worth of movies on SSDs no less?
Send me a boxfull of these 245TB SSDs, I'll show you. My film count surpassed Netflix's about 12 years ago, and I haven't slowed down once.
This is a big bet. Look at what happened in 2001 with the dot-com boom. We're still trading on their dark fiber over-build today. Meanwhile any overcapacity built in fabs will quickly be made obsolete by newer and better fab technology (or at least, that's been the pattern for the past 30 years).
I think you're missing the fact that building new supply takes time and sustained commitment, and there's simply nobody in a good position to make that commitment without losing big if your thesis turns out to be wrong.
If the demand for new AI builds is eventually satisfied, or worse, craters overnight, then who will be left holding the bag? It sure won't be Google or Apple, or even NVIDIA - it will be TSMC and Samsung.
But as you mention, this is all temporary. Either you're right, and demand will remain sustained long enough for some of the providers to decide to take that risk, or demand will crater and prices will fall.
The fact that the S&P 500 is near record highs at the same time as consumer confidence is at a 70 year low is not encouraging for continued all steam ahead in my mind... but then it's easy to predict a general future recession, and much harder to predict it to the day.
I doubt consumers will regret these compute purchases in 3 years, my 3 year old GPU is still holding value, actually increased in value, and I use it more now than ever.
I did predict or expect that if oil went to $150 we'd be in recession territory, currently hovering below that level and folks are feeling the squeeze and aren't happy. Things could get worse or better, tough to tell, but I think silicon demand is more or less secular and will do relatively well in a variety of macro conditions.
Bad metaphor. Not being forward enough is a common mistake. When learning, you may want to make yourself _feel_ like you're too far forward. We can commonly think we're too far forward when actually overall we're still too far back. Teaching it is very hard.
Although it is dependent on your style of skis (e.g. carving) and style of skiing (powder, racing, cruising).
For me, it is very rare that I go too far forward (I began with a leant back style and I haven't rectified that after many years of skiing). I try to prioritise fun over ability.
The fact that we _haven't_ seen such deals be made, and that ~50% of new datacenter builds have been quietly cancelled[1], suggests to me that we're dealing with paper demand more than true sustained demand.
1: https://www.tomshardware.com/tech-industry/artificial-intell...
The major companies have been trying to build >$100B of new production capacity in the US for years now. All of these manufacturing facilities have been significantly delayed by NIMBYs using the same "environmental and community concerns" advocacy slop that hinders almost all productive industry in the US.
Blaming the suppliers is a lazy take. They aren't responsible for degrowth activists being able to dictate what we are allowed to build.
It doesn't change your point that these lines are different and the immediate price spike is not about them competing for capacity in the same facilities, but the fact that manufacturers are committing to large enough future purchases to drive new fab construction means that future pricing outlook (which does impact the current prices to extent) does involve some amount of competition between different types of semiconductor products.
> I have over 500 NVMe drives in our deployment and the drive deaths are not due to wear.
Like, there's so many things that could benefit from a cheap processor involved in their operation, the growth seems effectively unlimited.
By the way, interesting podcast on that topic: https://www.youtube.com/watch?v=mDG_Hx3BSUE
When supply is constrained, widgets to get sold to whoever is willing to pay the most. It’s capitalism, baby. Arguably this is a good thing, because the limited resources are allocated to the people who can create the most value from them. And high prices provide an economic incentive to bring more supply online.
Nvidia is making big profits right now because they don’t have a lot of competition for their fancy AI chips. The prices will come down when they have more competition.
I think a lot of people will accept industry when industry tries to be a good steward of the community's environment.
As with the sudden inexplicable "concern" about data center construction, I don't think this activism is organic.
I don't think it is sudden, there was much of the same concern about Amazon warehouses and anything else that cuts down trees, disrupting wildlife, and human neighbors. The reason it is more of a concern with modern datacenters is because they don't come with a significant amount of new jobs for the communities they are affecting. People will put up with a lot if means their unemployed loved ones can get on their feet.
With AI being a looming threat to everyone's livelihood it is compounding the reaction, not only do these datacenters not provide new jobs, they are actively contributing to people losing their current jobs.
If these new datacenters were going into empty buildings in cities that already had adequate infrastructure, then no one would care. They would still be against ai in general, but the datacenters wouldn't be a problem.
edit:
They could probably get a ton of support if they went into one of the many cities with crumbling infrastructure, took over abandoned buildings, and paid to fix up the roads and stuff. Like if they moved into an abandoned factory in flint, and fixed the water supply they could be heros.
And some do. Most don’t, for the same reasons self driving doesn’t matter.
Or it could be that I mostly talk to people in the real world, and less so follow the echo chambers online that think "you'll own nothing" is a foregone conclusion about the future and fit their worldview to match.
A single speck of dust could throw off that measurement (~ 1.6 x 10^-7 grams)
I just want....I just want hard drive prices to come back down. *sniffle*