ssds Pure Storage Claims 300TB SSDs Could Arrive as Soon as 2026, but They Will Be Shockingly Expensive

[BiG StroOnZ]

Summary

Pure Storage, a supplier of all-flash storage devices, expects capacity of its proprietary direct flash module (DFM) solid-state drives to increase by six times in a few years, leading to 300TB capacities. The company expects advancements of 3D NAND areal density driven by increased number of layers and other factors to enable such a dramatic capacity increase. Pure CTO Alex McMullan showed a chart to Blocks & Files that reveals Pure’s plans and Pure Storage claims that it'll have these mammoth SSDs available by 2026. 

 

 

 

 

 

Quotes

Quote

Pure CTO Alex McMullan briefed Blocks & Files and presented a chart showing Pure’s DFM capacity expansion roadmap out to a 300TB module. He said: “The plan for us over the next couple of years is to take our hard drive competitive posture into a whole new space. Today we’re shipping 24 and 48TB drives. You can expect … a number of announcements from us at our Accelerate conference around larger and larger drive sizes with our stated ambition here to have 300TB drive capabilities, by or before 2026.”

 

This far outstrips other disk drive capacity roadmaps. For example, Toshiba sees its MAS-MAMR and HAMR technology taking it to 40TB in 2026. Seagate has said its HAMR tech should enable a 50TB HAMR drive in its fiscal 2025 and a 100TB drive “at least by 2030.” Pure Storage will have a 5 to 6x capacity per drive advantage by then if its engineers can deliver the DFM goods.

 

How will Pure get to a 300TB DFM? McMullan again: “All the chip fabs are shipping us somewhere between 112 and 160 layers. All the fab vendors have a plan and a path to get to 400-500 layers over the next five years. And the whole point of that will help us of course, on its own.”

 

Now a DFM is not your usual SSD; you won’t be able to take one of these and slot in your workstation or business PC. Instead, you have to buy an entire system, and they don’t come cheap. Pure Storage launched its FlashBlade//E on March 1 2023, and while it costs only $0.20 per GB, the minimum purchase order is a staggering 4PB (or $800,000), not your usual portable SSD, or NAS for that matter.

 

The newly-launched product uses 40 x 48TB DFMs per chassis (working in pairs to get to 4PB). 300TB DFMs would bring total capacity to 12PB, a near-sevenfold improvement that can be achieved by hoping that the likes of Samsung and Solidigm increase the number of NAND layers to around 500.

 

Other ways of increasing capacity would be by increasing the physical size of DFMs and moving to PLC (Penta-layer cell), the successor of mainstream QLC, a technology that is currently being tested and likely to be rolled out in 2023.

 

My thoughts

While this technology is used for enterprise solutions, consumers would eventually see its benefits at some point. As 3D NAND flash manufacturers will ultimately switch from the current 200-layer chips to the 400 or 500-layer chips, and this will drive storage density up. Presently, consumer SSDs offer up to 15TB solutions, while HDDs offer up to 22TB. Meaning, if these projections are correct, extra high capacities will at some point drive down prices of SSD per TB, while at the same time increasing the average capacity of an SSD. While an assumption, if SSD prices come down over the course of 5 years, this should also drive down HDD prices as well. Something to keep in mind, as Blocks & Files points out, is that 5 years brings us all the way to 2028, 2 years after McMullan’s purported 300TB DFMs by 2026 goal. Translation, 3D NAND layer increases will not get Pure Storage to 300TB by 2026 on their own; Pure Storage can't rely solely on increased number of active layers, but will need to improve upon other things too. Nonetheless, while Pure Storage's plans seem enthusiastic, it is still very exciting nonetheless; despite this being a primarily enterprise focused endeavor. I believe the benefits enterprise receives, eventually make their way to consumers at some point or another. 

 

Sources

https://www.tweaktown.com/news/90590/300tb-ssds-could-arrive-in-2026-but-with-big-catch/index.html

https://www.techpowerup.com/305422/300-tb-ssds-could-arrive-as-soon-as-2026-claims-pure-storage

https://www.techradar.com/news/300tb-ssd-are-coming-soon-but-they-will-be-shockingly-expensive

https://blocksandfiles.com/2023/03/01/300tb-flash-drives-coming-from-pure-storage/

https://www.tomshardware.com/news/pure-storage-300-tb-flash-drives-in-2026

[StDragon]

Hopefully Azure and AWS will be able to price their storage tiers accordingly. Meaning, performance and per GB should get cheaper.

[Levent]

Makes me wonder if controllers on these also has to get proportionally faster. On that note I thought DFMs didn’t have any controllers on board?

[BiG StroOnZ]

3 minutes ago, Levent said:

On that note I thought DFMs didn’t have any controllers on board?

 

From Tom's Hardware:

 

Quote

The modules use enterprise-grade SSD controllers albeit with fully custom firmware as well as 3D TLC or 3D QLC NAND memory devices though it is unclear whether the company uses any custom packaging to increase capacity of its flash chips. 

 

[TetraSky]

Eh, cool.

Too bad this won't really drive prices down, until all manufacturers decide to stop selling low capacity nand chips.

 

That "custom firmware" with full telemetry better just be for testing purpose and not in consumer products though.

[leadeater]

Quote

but They Will Be Shockingly Expensive

I dunno, that is probably the least shocking thing in this story lol

 

18 hours ago, TetraSky said:

That "custom firmware" with full telemetry better just be for testing purpose and not in consumer products though.

Pure Storage systems report back to a provided service which monitors performance, failures and other issues. This is common, many other vendors do the same. Customers of these systems want this, proactively picking up on configuration issues or identifying abnormal system behavior is something that is very much wanted.

[Distinctly Average]

8 minutes ago, leadeater said:

I dunno, that is probably the least shocking thing in this story lol

 

Pure Storage systems report back to a provided service which monitors performance, failures and other issues. This is common, many other vendors do the same. Customers of these systems what this, proactively picking up on configuration issues or identifying abnormal system behavior is something that is very much wanted.

Indeed, a bit like a more advanced SMART from spinning magnet days. All our drives have a similar technology, but it has to be enabled through the iRMC and an error reporting system that requires specific configuration as to where it will send these reports. I don't think consumers need worry.

[Doobeedoo]

Yeah we've seen these huge capacities in various formats, though on consumer side I wonder how it grows. Ignoring that SATA though, on M.2 stick form factor it's leas physical space. Now even today most don't need highest single capacity one let alone all slots. But yeah I expect for consumer needs form factor will suffice.

[Lurick]

Having worked with a few Pure Storage products, I'm excited to see what these bring and glad to see them pushing forward

[Salv8 (sam)]

oh the amount of 'homework' you could store...

the raw rips of your media...

the backups of your games...

i'm selling a kidney right now.

[williamcll]

Not sure how reliable QLC is though.

[TopWargamer]

The amount of Linux distros, git repos, and Wiki articles I could store on this makes me blush. May even run 2 in raid 0 if I'm feeling extra spicy.

[Kisai]

On 3/3/2023 at 2:13 PM, BiG StroOnZ said:

 

 

My thoughts

While this technology is used for enterprise solutions, consumers would eventually see its benefits at some point. As 3D NAND flash manufacturers will ultimately switch from the current 200-layer chips to the 400 or 500-layer chips, and this will drive storage density up. Presently, consumer SSDs offer up to 15TB solutions, while HDDs offer up to 22TB.

 

The only reason magnetic storage is even still around is because SSD's don't yet offer a $/GB price point that makes it competative. I bought a 16TB and a 18TB (or was it 20? "formatted capacity" always annoys me) WD Red Pro because they were essentially half price at the time. Would I love to have a SSD instead? Yes. Do I want to pay ... oh $800,000 for it? No.

 

https://www.techradar.com/news/300tb-ssd-are-coming-soon-but-they-will-be-shockingly-expensive

 

Quote

Instead, you have to buy an entire system, and they don’t come cheap. Pure Storage launched its FlashBlade//E on March 1 2023, and while it costs only $0.20 per GB, the minimum purchase order is a staggering 4PB (or $800,000), not your usual portable SSD, or NAS for that matter.

So, let's, for the sake of argument, reverse engineer that price into a smaller unit. Let's say I wanted a 60TB drive to replace all my existing mechanical drives. It would cost $12,000 at this price point. Meanwhile it cost me like $300 each 20TB hard drive (on sale). The price has to come down to around $0.025/GB before it hits a point where mechanical drives are no longer viable at all. The list price of 20TB Mechanical drives is $500. So we're still looking at the price needing to drop by 90% before it replaces mechanical drives.

 

That said, not every piece of thing ever needed to be stored on SSD's, but I can't imagine mechanical drives still being used beyond 2030, they will just be too expensive to make to be performance/capacity-competitive with a SSD.

 

[Sauron]

On 3/3/2023 at 11:37 PM, Levent said:

Makes me wonder if controllers on these also has to get proportionally faster.

Well at some point you'll run into the same problem huge hard disks have; reading the entire drive for cloning or recovery is going to take hours. Controllers aside this would become a problem for bus interfaces where you'll have to decide how much bandwidth you're even willing to dedicate to your storage. Even at 60GB/s (about the bandwidth of an entire pcie 5 16x slot) it would take you over an hour to read through a 300TB drive.

[leadeater]

1 hour ago, Sauron said:

Even at 60GB/s (about the bandwidth of an entire pcie 5 16x slot)

Sustained 60GB/s, you say that so casually lol. Hell I wish that were practical, sadly not.

[Kisai]

4 hours ago, Sauron said:

Well at some point you'll run into the same problem huge hard disks have; reading the entire drive for cloning or recovery is going to take hours. Controllers aside this would become a problem for bus interfaces where you'll have to decide how much bandwidth you're even willing to dedicate to your storage. Even at 60GB/s (about the bandwidth of an entire pcie 5 16x slot) it would take you over an hour to read through a 300TB drive.

Consider the other half of the problem. What if one of the SSD chips explodes/fails/etc, and takes the entire drive with it. 

 

Like at a certain point, you have to consider the possibility that the entire drive will just eat it, which is why it's better to have like 4 physical drives to spread your data/programs/bulk storage around on, rather than put everything on one large drive. Like if your OS drive is 1TB PCIe3, and your game/application drive is 4TB PCIe5, and everything you're not actively using daily, but could re-download if the drive dies is 20TB, can be a mechanical drive. Which is how I have stuff setup on my PC, but the various things I do (like capture/edit video) would benefit from large SSD's, but not generally be full-read through in work processes.

 

But oh do I hate the idea of having to backup what I have. Buying additional drives to cover it is a as much as another computer, and paying for cloud storage is impractical since it would take ... oh a year and a half to sync.

[StDragon]

6 hours ago, Sauron said:

Well at some point you'll run into the same problem huge hard disks have; reading the entire drive for cloning or recovery is going to take hours. Controllers aside this would become a problem for bus interfaces where you'll have to decide how much bandwidth you're even willing to dedicate to your storage. Even at 60GB/s (about the bandwidth of an entire pcie 5 16x slot) it would take you over an hour to read through a 300TB drive.

 

1 hour ago, Kisai said:

Consider the other half of the problem. What if one of the SSD chips explodes/fails/etc, and takes the entire drive with it. 

 

Like at a certain point, you have to consider the possibility that the entire drive will just eat it, which is why it's better to have like 4 physical drives to spread your data/programs/bulk storage around on, rather than put everything on one large drive.

 

Pure basically uses RAID6 for fault-tolerance. They call it "RAID-HA", which is more specific to flash storage, but either way it's dual-parity.

 

[leadeater]

5 minutes ago, StDragon said:

Pure basically uses RAID6 for fault-tolerance. They call it "RAID-HA", which is more specific to flash storage, but either way it's dual-parity.

That's still across devices though, if for a consumer SSDs are really large and you buy one SSD then a chip fault is the end of that device. I'm not worried much about that myself, more the controller but yea... single storage device has never been the most resilient of setups.

[StDragon]

23 minutes ago, leadeater said:

That's still across devices though, if for a consumer SSDs are really large and you buy one SSD then a chip fault is the end of that device. I'm not worried much about that myself, more the controller but yea... single storage device has never been the most resilient of setups.

Franky, this isn't a new problem. If you need HA, you purchase two of the same drive and put them into RAID1 for fault-tolerance. But as everyone should know by now, RAID isn't a backup. That's why I've always recommended 2x drive at whatever capacity you need. So in a theoretical scenario involving a single 300TB drive, you purchase another 300TB and target that for backups.

If you can afford 3x 300TB drives, but only need 300TB worth of storage, then you put two of them in RAID1, and make the remaining a repository for backups.

But above all else people, never dedicate your secondary drive as a mirror without first having backups in place; because a delete from one drive will delete from the other in RAID1.

[leadeater]

1 hour ago, StDragon said:

But above all else people, never dedicate your secondary drive as a mirror without first having backups in place; because a delete from one drive will delete from the other in RAID1.

Also a mirror has the same write workload where a backup doesn't. Mirror gets all those transient writes etc while a daily backup etc just gets what's changed between each backup

[Sauron]

10 hours ago, leadeater said:

Sustained 60GB/s, you say that so casually lol. Hell I wish that were practical, sadly not.

well, one can dream