Solidigm Introduces the D5-P5430 QLC Data Center SSD

[TheLostSwede]

Solidigm, a leading global provider of innovative NAND flash memory solutions, is expanding its D5 Product Series with the Solidigm D5-P5430, a new QLC solid-state storage drive (SSD) optimized for mainstream and read-intensive workloads. With most of today's enterprise applications read-dominant, the D5-P5430—a 4th gen PCIe QLC SSD—offers substantial storage density and Total Cost of Ownership (TCO) opportunities while delivering read performance that is equivalent to the most widely-adopted TLC SSDs.

The D5-P5430 is optimized for mainstream workloads (e.g., email/unified communications, decision support systems, object storage, and virtual desktop infrastructure) and read-intensive workloads (e.g., content delivery networks, data lakes/pipelines, video-on-demand). These workloads are typically 80% reads or higher and need to move massive amounts of data at high throughput.




As a drop-in replacement for TLC NAND-based PCIe-based SSDs, the D5-P5430 can reduce TCO by up to 27% for a typical object storage solution with a 1.5X increase in storage density and 18% lower energy cost. In addition, Solidigm's newest drive can deliver up to 14% higher lifetime writes vs. leading TLC SSDs.

Important data center issues like power efficiency, edge computing, and sustainability of infrastructure, all benefit from the density of the D5-P5430, which can reduce power and cooling costs and lower the number of drives needed by up to 2x when compared to alternative solutions.

"Data centers need to store and analyze massive amounts of data with cost-effective and sustainable solutions," said Greg Matson, VP, Strategic Planning and Marketing, Solidigm. "Solidigm's D5-P5430 drives are ideal for this purpose, delivering high density, reduced TCO, and 'just right' performance for mainstream and read-intensive workloads."

The D5-P5430 supports a wide range of 1U and 2U server and storage configurations with a broad range of capacities that support the following legacy and modern EDSFF form factors:

• U.2 15 mm 3.84 TB - 30.72 TB
• E1.S 9.5 mm 3.84 TB - 15.36 TB
• E3.S 7.5 mm 3.84 TB - 30.72 TB

"Supermicro works with leading storage suppliers like Solidigm to bring the latest technology to our global customers that expect high-performance and secure storage technology," said Vik Malyala, Managing Director and President, EMEA; SVP, WW FAE, Supermicro. "Solidigm's new QLC-based E3.S SSD, the D5-P5430, gives customers a compelling combination of high density and efficiency for the data center infrastructure."

View at TechPowerUp Main Site | Source

 

[chrcoluk]

All my work in enterprise environments, they are so write intensive which is why enterprise storage tends to be focused on sustaining high amounts of writes.

You can cache reads to oblivion, but writes have to make it to the storage.

What an odd product. If they can make a 30TB QLC drive, they can also make a 30TB TLC drive, QLC only adds 33% density not 100%.

 

[HairyLobsters]

Never seen an E1.s or E3.s drive before. Seems like a better form factor that M.2.

 

[TheinsanegamerN]

QLC is disgusting and is suited for nothing except low end consumer media PCs. Anything else TLC or bust.

 

[Wirko]

All my work in enterprise environments, they are so write intensive which is why enterprise storage tends to be focused on sustaining high amounts of writes.

You can cache reads to oblivion, but writes have to make it to the storage.

What an odd product. If they can make a 30TB QLC drive, they can also make a 30TB TLC drive, QLC only adds 33% density not 100%.

Obviously not everyone needs the highest possible density, or else why would Solidigm, and Samsung, and Micron, and others bother making and selling 4TB and 8TB models for the enterprise? And even smaller ones too?

 

[Maxx]

QLC is disgusting and is suited for nothing except low end consumer media PCs. Anything else TLC or bust.

I disagree, but aside from that, this looks suspiciously like Intel's 192L PLC, so you could be even more disgusted. They detailed this at ISSCC earlier this year. These were 1.67Tb dies but reading through the presentation it looked like it was based on QLC (which would be 1.33Tb dies). I didn't see anyone open the drive but this gives an idea of the flash's characteristics (e.g., can estimate performance). Hynix has their own 176L QLC design from ISSCC 2022, although that is CTF. FG is more likely given the endurance here.

What an odd product. If they can make a 30TB QLC drive, they can also make a 30TB TLC drive, QLC only adds 33% density not 100%.

Check my reply right above this one, but I have more to add to your point. Intel has done this in the past, using QLC in TLC (pTLC) mode for enterprise. This appears to be PLC in QLC mode, but I wouldn't call it pQLC per se, it operates a bit differently because of the latch layout. In any case, the purpose would be to use floating gate for endurance, and actually the density is quite high on their PLC even in TLC mode. In fact in their 2023 presentation they state "the chip can operate in QLC ... delivering 24% higher bit density" when compared to the best-reported QLC densities. (176L Micron/Hynix QLC is at 14.7-14.8Gb/mm^2, this would be up to 18.6)

It can actually also operate in TLC but there is some special sauce here in how they designed this flash for which it makes sense in this application. (see presentation linked above)

 

[Scrizz]

All my work in enterprise environments, they are so write intensive which is why enterprise storage tends to be focused on sustaining high amounts of writes.

You can cache reads to oblivion, but writes have to make it to the storage.

What an odd product. If they can make a 30TB QLC drive, they can also make a 30TB TLC drive, QLC only adds 33% density not 100%.

They clearly mentioned CDNs etc. Those would be read-intensive workloads. I know I've seen mentions of Solidigm releasing 61TB SSDs. https://www.tomshardware.com/news/solidigm_61tb_ssd

 

[chrcoluk]

Yeah for node caches I accept the point, the toms hardware link suggests they made some kind of gains on QLC somewhere, although I do suspect toms hardware may be being fed pSLC numbers especially on the quoted sequential numbers. So these are likely "up to" dependent on if data is read/written from pSLC?

Kind of scary how obsessed the manufacturers have become with increasing the density, its like a cult behaviour almost. Kind of puts pressure on to buy SSDs now as whats on the market today will probably be gone within 1-2 years.

 

[Maxx]

Yeah for node caches I accept the point, the toms hardware link suggests they made some kind of gains on QLC somewhere, although I do suspect toms hardware may be being fed pSLC numbers especially on the quoted sequential numbers. So these are likely "up to" dependent on if data is read/written from pSLC?

Kind of scary how obsessed the manufacturers have become with increasing the density, its like a cult behaviour almost. Kind of puts pressure on to buy SSDs now as whats on the market today will probably be gone within 1-2 years.

I'd have to check out TH's source but generally you're not using SLC/pSLC for DC drives. Now, this flash (as listed above in the presentation I linked) was designed with special pSLC to shore up the endurance to 1K PEC in 5-bit/PLC mode so I'm not sure what impact that has here as they use the SLC as a sort of slower latch when doing program suspend but this is not the same as pSLC on a consumer device. Considering this flash was designed for PLC (but made to operate as TLC or QLC, presumably as we saw the 144L QLC in TLC for previous Intel/Solidigm DC drives) it probably would blow the doors off of something like last-gen BiCS QLC (which was garbage in performance).

 

[chrcoluk]

I'd have to check out TH's source but generally you're not using SLC/pSLC for DC drives. Now, this flash (as listed above in the presentation I linked) was designed with special pSLC to shore up the endurance to 1K PEC in 5-bit/PLC mode so I'm not sure what impact that has here as they use the SLC as a sort of slower latch when doing program suspend but this is not the same as pSLC on a consumer device. Considering this flash was designed for PLC (but made to operate as TLC or QLC, presumably as we saw the 144L QLC in TLC for previous Intel/Solidigm DC drives) it probably would blow the doors off of something like last-gen BiCS QLC (which was garbage in performance).

Yep I am not sure, my DC drive for instance doesnt use pSLC. Just the numbers dont seem realistic hence the suggestion, sadly we may never know as these are unlikely to get reviewed.

Of course if we see future consumer QLC drives with these types of numbers, then we may get the answer then when those get reviewed, interesting the things learnt here about the PLC running in QLC mode.

 

[Maxx]

Yep I am not sure, my DC drive for instance doesnt use pSLC. Just the numbers dont seem realistic hence the suggestion, sadly we may never know as these are unlikely to get reviewed.

Of course if we see future consumer QLC drives with these types of numbers, then we may get the answer then when those get reviewed, interesting the things learnt here about the PLC running in QLC mode.

I've pegged the performance increase at around 11% over the previous gen for sequential per die, if that's what you mean. IMFT -> Intel QLC went from 30MB/s to 40MB/s per die and this seems higher. I thought you had been referring to the QLC comparison numbers (which compares inferior QLC). If you mean sequential, that's straightforward since it'll be in QLC mode. These drives have 16-channel controllers and can interleave up to 6 dies a channel at least. Napkin math would put the new dies at 44-45MB/s. This is certainly not TLC levels of performance but is pretty good giving they were going for density primarily with the flexibility of TLC & PLC modes (TLC density wouldn't be quite as good as YMTC's 232L WoW TLC or Micron's 232L TLC, but way ahead of any 176L).

 

[chrcoluk]

I mean this.