Monday, July 24th 2023

Solidigm Introduces D5-P5336 - the World's Highest Capacity PCIe SSD

Solidigm over the weekend announced the Solidigm D5-P5336 line of enterprise SSDs. The company earns bragging rights for having the highest capacity among production SSDs, with these coming in capacities of up to 61.44 TB—an incredible amount of storage that HDDs have yet to catch up to. These drives are targeted at the data-center, where they not just replace HDDs due to their sheer density, but also provide the advantage of responsive flash storage. The drives aren't gunning to top SSD performance charts, the design goal is simply capacity and reliability. The D5-P5336 series feature enterprise-grade QLC NAND flash memory that's been extensively tested for reliability and write endurance.

The Solidigm D5-P5336 series comes in three data-center relevant form-factors—15 mm-thick U.2, 7.5 mm-thick E3.S, and the E1.L (ruler) form-factor. Capacities start at 7.68 TB for the U.2 and E3.S models, and 15.36 TB for the E1.L. The maximum capacity on offer is 61.44 TB for the U.2 and E1.L form-factors, and 30.72 TB for the E3.S. Among the capacity variants are 15.36 TB, 30.72 TB, and 61.44 TB. The drives use a proprietary controller that takes advantage of the PCI-Express 4.0 x4 host interface, and NVMe 1.4c protocol. The star attraction, however, is the 192-layer 3D QLC NAND flash memory made in-house by Solidigm.
All variants offer sequential read speeds of up to 7 GB/s. Since these are high-density QLC drives, the sequential write speeds are low, but rated at up to 3.3 GB/s. Solidigm claims this is within 20% of a theoretical TLC-based drive at comparable densities. 4K random access speeds at QD-256 is rated at 1.005 million IOPS, and 16K writes at QD-256 at 43,000 IOPS. The company didn't put out variant-specific endurance numbers, yet; but the 61.44 TB variant can take 0.58 DWPD, while the base 7.68 TB variant 0.42 DWPDs. The 61.44 TB variant should typically offer 213,000 TBW, and the 30.72 TB variant 105,000 TBW.

Solidigm Press Release

Solidigm is proud to announce another industry-first quad-level cell (QLC) solid-state storage drive (SSD) for the data center—the Solidigm D5-P5336. Offered in capacities from 7.68 TB to 61.44 TB, the Solidigm D5-P5336 makes it possible to store up to 6X more data in the same space compared to an all hard disk drive (HDD) array.

"Modern workloads like AI and capabilities like 5G are rapidly reshaping the storage landscape," said Greg Matson, VP of Strategic Planning and Marketing at Solidigm. "Businesses need storage in more places that is inexpensive, able to store massive data sets efficiently and access the data at speed. The D5-P5336 delivers on all three -- value, density and performance. With QLC, the economics are compelling -- imagine storing 6X more data than HDDs and 2X more data than TLC SSDs, all in the same space at TLC speed."

Built to handle massive amounts of data from the core to the edge, the read performance of the Solidigm D5-P5336 exceeds some of the latest cost-optimized triple-level cell (TLC) SSDs on the market today. Artificial intelligence (AI), machine learning (ML), content delivery networks, scale-out network attached storage (NAS), and object storage are all read-intensive workloads that continue to drive enormous volumes of data, regardless of where they reside.

Massive scalability and improved TCO across a range of configurations
With its high capacity enabling smaller storage footprints, Solidigm's D5-P5336 enables a lower total cost of ownership and more sustainable infrastructure than all-TLC arrays, SAS HDD arrays, or hybrid arrays. Consider the following, based on a 100PB object storage solution deployment:

"For years there has been debate about endurance and reliability of SSDs, QLC in particular, but Solidigm might have ended that debate with the D5-P5336," noted Avery Pham, VP Operations, VAST Data. "Any number of applications will see notable benefits from these drives from AI and machine learning to object storage."

"Today, it is clear that the primary constraint for edge workloads is the limitation of bandwidth rather than latency," said Doug Emby, VP of Sales and Business Development at Cheetah RAID Storage. "Solidigm's D5-P5336 QLC SSDs offer an impressive combination of capacity, performance, and reliability as a solution to overcome this challenge. The seamless integration of these Solidigm QLC SSDs with Cheetah's high-performance servers makes them highly suitable for the efficient deployment of edge solutions."

The D5-P5336 is shipping now in E1.L form factor with up to 30.72 TB, with subsequent availability extending to 61.44 TB in both U.2 and E1.L later this year. In the first half of 2024, Solidigm will be shipping E3.S form factor with up to 30.72 TB.

To learn more about the D5-P5336 visit this page. Additionally, the D5-P5336 will be on display at the Solidigm booth (#107) at Flash Memory Summit in Santa Clara, Calif. August 8-10.

Solidigm Slide Deck

Sources: Solidigm News, D5-P5336 Product Page
Add your own comment

20 Comments on Solidigm Introduces D5-P5336 - the World's Highest Capacity PCIe SSD

#1
TumbleGeorge
I hope Solidigm with this product will "drag a leg" and cause a stir in capacities and SSDs for ordinary users. The time for 16TB m.2 or other format in terms of low production costs due to cheap nand chips has never been better.
Posted on Reply
#2
yannus1
They definitely need to make better charts. They are confusing and even give a bad impression of the product.
Posted on Reply
#3
Denver
I think I'm going to need one of these for the next "wonderful" ports. :)
Posted on Reply
#4
Chrispy_
One of the charts says their new QLC has 3000 p/e cycles, better than man TLC drives:



Other charts show endurance figures of 790 p/e cycles for the 7.68TB model, up to 1090 p/e cycles for the 61.44 TB model:



So no, they haven't magically solved the QLC endurance issue, it's still around a third of TLC's endurance. Is half a drive write per day a lot of writing on a 61.44TB drive? Absolutely yes, but it's still QLC, and still a long way from ideal when these things are likely to cost around $4000 each.

I can see their use case for content hosting where the data is almost static yet hundreds of thousands of users could be pulling data off it simultaneously - but AFAIK most of the large content hosts go down the route of slow, cheap, spinning rust with enormous amounts of redundancy. Perhaps that thinking is being overwhelmed by increasing datacenter energy and size constraints. I don't know for sure, I'm not a Microsoft/Amazon/Google/Facebook employee...
Posted on Reply
#5
JAB Creations
What is the form factor? Wait a minute, I've got a meme picture here somewhere...
Posted on Reply
#6
Wirko
Chrispy_One of the charts says their new QLC has 3000 p/e cycles, better than man TLC drives:

Other charts show endurance figures of 790 p/e cycles for the 7.68TB model, up to 1090 p/e cycles for the 61.44 TB model:
Both can be true. P/E cycles is a term that's used to describe NAND chips; TBW and DWPD are associated with SSDs and are calculated with some assumptions regarding write amplification.

But the charts dont say as much, they are misleading, that's for sure.
Chrispy_Perhaps that thinking is being overwhelmed by increasing datacenter energy and size constraints.
This, and IOPS too. A single SSD has the IOPS of a large van filled with hard disks, and when serving thousands of users, this is crucial.
Posted on Reply
#7
Prima.Vera
We need more and cheaper 8TB SSDs... For some reason we are stuck with 4TB SSD for so long and same price almost....
Posted on Reply
#8
Chrispy_
Prima.VeraWe need more and cheaper 8TB SSDs... For some reason we are stuck with 4TB SSD for so long and same price almost....
I could be wrong on this but I believe most of the players in the industry are currently capped at 512GiB NAND dies, and since the affordable mainstream controllers are 4-channel you have a 4x512GiB sweet spot of 2TB. 4TB requires stacking/interleaving NAND dies which results in higher costs, and more expensive 8-channel controllers like the Phison E18 can then double that as well, just about squeezing everything onto a double-sided 2280 M.2 form factor with no room to spare, but it's an engineering marvel using the most expensive parts available, and as many of them as physically possible. Nothing about that will ever be cheap, hence the 4TB reasonably-priced limit for 2280 SSDs.

Once you go beyond 8TB drives at the moment, I'm not entirely sure what's happening under the hood. Intel/Solidigm/WD drive reviews I've read test performance but not the internal layout. I don't know if the 30.7TB/61.4TB U2 form factor drives are internal RAID/JBOD arrays behind a RAID controller, or whether they use dedicated SSD controllers that have 16 or 32 channels.

So, until we get higher-density NAND dies, we're stuck at 16 multiples (8-channel, double-stacked) of the current 512GiB dies for 2280, consumer-friendly 2280 M.2 SSDs.
Posted on Reply
#9
Wirko
Chrispy_I could be wrong on this but I believe most of the players in the industry are currently capped at 512GiB NAND dies, and since the affordable mainstream controllers are 4-channel you have a 4x512GiB sweet spot of 2TB. 4TB requires stacking/interleaving NAND dies which results in higher costs, and more expensive 8-channel controllers like the Phison E18 can then double that as well, just about squeezing everything onto a double-sided 2280 M.2 form factor with no room to spare, but it's an engineering marvel using the most expensive parts available, and as many of them as physically possible. Nothing about that will ever be cheap, hence the 4TB reasonably-priced limit for 2280 SSDs.

Once you go beyond 8TB drives at the moment, I'm not entirely sure what's happening under the hood. Intel/Solidigm/WD drive reviews I've read test performance but not the internal layout. I don't know if the 30.7TB/61.4TB U2 form factor drives are internal RAID/JBOD arrays behind a RAID controller, or whether they use dedicated SSD controllers that have 16 or 32 channels.

So, until we get higher-density NAND dies, we're stuck at 16 multiples (8-channel, double-stacked) of the current 512GiB dies for 2280, consumer-friendly 2280 M.2 SSDs.
NAND dies are 512 gigabits, not gigabytes! Stacks of 8 are a common thing. 16 is apparently when things become technologically challenging for some reason, which means expensive.
You can often find the number of CE (chip enable) signals in controller data sheets. The E18 has 32, which means that 32 chips (dies) can be connected to it. But maybe some kind of multiplexing exists in addition to CEs because 32 x 512 Gbit isn't enough for a 4 GB drive.
Posted on Reply
#10
Chrispy_
WirkoNAND dies are 512 gigabits, not gigabytes! Stacks of 8 are a common thing. 16 is apparently when things become technologically challenging for some reason, which means expensive.
You can often find the number of CE (chip enable) signals in controller data sheets. The E18 has 32, which means that 32 chips (dies) can be connected to it. But maybe some kind of multiplexing exists in addition to CEs because 32 x 512 Gbit isn't enough for a 4 GB drive.
Ah okay.
Same net result by the sound of it though?
If it's bits not bytes, but they stack up to 8, then there's still this cost/complexity threshold at 512GiB per package.
Posted on Reply
#11
chrcoluk
Chrispy_One of the charts says their new QLC has 3000 p/e cycles, better than man TLC drives:



Other charts show endurance figures of 790 p/e cycles for the 7.68TB model, up to 1090 p/e cycles for the 61.44 TB model:



So no, they haven't magically solved the QLC endurance issue, it's still around a third of TLC's endurance. Is half a drive write per day a lot of writing on a 61.44TB drive? Absolutely yes, but it's still QLC, and still a long way from ideal when these things are likely to cost around $4000 each.

I can see their use case for content hosting where the data is almost static yet hundreds of thousands of users could be pulling data off it simultaneously - but AFAIK most of the large content hosts go down the route of slow, cheap, spinning rust with enormous amounts of redundancy. Perhaps that thinking is being overwhelmed by increasing datacenter energy and size constraints. I don't know for sure, I'm not a Microsoft/Amazon/Google/Facebook employee...
3D TLC endurance I think has also expanded past 3000 P/E, the MX500 chart I posted some time ago which shows the capability of Micron 176l 3D TLC showed it hitting if I remember right 5000 P/E and the chart was given credibility by a respected member of this community. So bear in mind whilst they make progress on QLC, TLC isnt standing still. But yeah QLC is getting better than before.



Thread

www.techpowerup.com/forums/threads/rumour-that-mx500-now-has-qlc-variants.302350/
Posted on Reply
#12
Chrispy_
I guess half the reason that some TLC drives were listed as lower endurance is nothing to do with the actual TLC endurance, but because the manufacturers of those drives were baking shitty QLC endurance into the spec to allow them to bait-and-switch to QLC further down the line.

Just about every SSD manufacturer has been guilty of that so far, and it's not that the TLC has low endurance, it's that the model "currently using TLC" still needs to meet its rated and sold specification when they later downgrade the flash so that they don't get hit with class-actions.
Posted on Reply
#13
chrcoluk
Chrispy_I guess half the reason that some TLC drives were listed as lower endurance is nothing to do with the actual TLC endurance, but because the manufacturers of those drives were baking shitty QLC endurance into the spec to allow them to bait-and-switch to QLC further down the line.

Just about every SSD manufacturer has been guilty of that so far, and it's not that the TLC has low endurance, it's that the model "currently using TLC" still needs to meet its rated and sold specification when they later downgrade the flash so that they don't get hit with class-actions.
For sure, this is why I try to buy SSD's from brands who control their entire supply line directly, then I think bait and switch is less likely. So would be Samsung, WD, not sure if is any others?
Posted on Reply
#14
Wirko
chrcolukFor sure, this is why I try to buy SSD's from brands who control their entire supply line directly, then I think bait and switch is less likely. So would be Samsung, WD, not sure if is any others?
Also Micron/Crucial and SK Hynix/Solidigm. Micron also uses their own controllers, possibly for all their SSDs (sources: Tom's, Blocks and Files). SK Hynix makes controllers too (one is found in the Solidigm P44 Pro).

But if endurance is specified, and sufficiently high, I would be confident I'm not going to get a QLC-filled drive. As an example, the Crucial P5 Plus (1 TB) has an endurance of 600 TB, and the Kingston KC3000 and Fury Renegade have more. Consumer-level QLC isn't there yet.
Posted on Reply
#15
chrcoluk
Didnt know Micron own Crucial, problem there then is MX500 was bait and switched.

So have no high endurance rated drives been switched then?
Posted on Reply
#16
Wirko
chrcolukDidnt know Micron own Crucial, problem there then is MX500 was bait and switched.

So have no high endurance rated drives been switched then?
I remember a thread ... Ah, it was your thread! But the conclusion was that all QLC samples came from Aliexpress, right? Counterfeit drives?

Drives such as the Kingston NV2 have pretty low endurance, it indeed has TLC and QLC variants in the TPU database. I don't know about other confirmed examples that magically became QLC during their lifetime. The SX8200 Pro, the grandfather of bait-and-switch, had countless variants but all were TLC.
Posted on Reply
#17
chrcoluk
Just had another look, yeah I see, doesnt seem 100% conclusive, but a pattern emerged.

Ideally someone can take a punt and buy a 4tb from amazon or something. :)
Posted on Reply
#18
Scrizz
so Intel and Micron NAND would be similar since they developed it together.
www.networkworld.com/article/3247746/impact-of-intel-and-micron-ending-their-nand-partnership-is-negligible.html

and Solidigm is Intel NAND Division sold to SK Hynix
www.intel.com/content/www/us/en/newsroom/news/intel-sells-ssd-business-dalian-facility-sk-hynix.html#gs.31ozln
www.anandtech.com/show/17134/intel-sells-ssd-business-to-sk-hynix-as-new-subsidiary-solidigm

Let me see if I can pick up some used Enterprise drives on eBay.
Posted on Reply
#19
Wirko
ScrizzLet me see if I can pick up some used Enterprise drives on eBay.
Mr. Chrcoluk has shared some experience with those, and it's not brilliant through and through (but if you're buying by the pound, you sure get many lbs per £).
Posted on Reply
#20
Scrizz
WirkoMr. Chrcoluk has shared some experience with those, and it's not brilliant through and through (but if you're buying by the pound, you sure get many lbs per £).
I only buy Intel Enterprise SSDs ;)
Posted on Reply
Add your own comment
Apr 29th, 2024 11:46 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts