Silicon Motion's SM2508 Set to Launch in Q4, Edging Out Phison as Top SSD Controller

[TheLostSwede]

Silicon Motion's SM2508 was first revealed in August last year at the Flash Memory Summit 2023, but after that things went pretty quiet. However, the company was demoing the SM2508 up and running at Computex this past week and it's set to edge out Phison's E26 Max14um in the battle of fastest NVMe SSD controller. We're not talking about any massive gains here, but the reference drive from Silicon Motion was shown running CrystalDiskMark 8.0.4 at the show and if we do a rough comparison to a Phison E26 Max14um, the SM2508 beats Phison by about 800 MB/s in sequential read performance and 500 MB/s in sequential write performance.

This might not seem like a whole lot, but the SM2508 is built on TSMC's N6 node which results in a 3.5 Watt peak power consumption, or 7 Watts for the entire SSD at load. A typical Phison E26 based SSD draws in excess of 11 Watts of power at full load, which is a big difference in a mobile device. This should obviously also lead to lower thermals and we should finally see PCIe 5.0 drives that don't need massive heatsinks or active cooling. In fact, 7 Watts power draw is very similar to Phison's E18 PCIe 4.0 based SSDs. Silicon Motion is still working on fine tuning the firmware for the SM2508, so performance might yet improve to reach the promised 14 GB/s write performance. Currently the random performance is also looking a bit on the weak side compared to Phison. According to Tom's hardware, we should see the first drives with the Silicon Motion SM2508 appear in the market sometime in Q4 this year.



View at TechPowerUp Main Site | Source

 

[dgianstefani]

Finally, Gen 5 SSDs on a modern node.

The fact some gen 5 SSDs till now are literally double the price of a gen 4, yet controller is built on ~12 nm etc., is a bit absurd, when the main downside is heat.

Not that anything other than sequential is going up by anything significant with gen 4 and gen 5 NAND flash drives. Controllers are better at working around the downsides of the tech though with clever algorithms.

TBH I'm more impressed with modern DRAMless drives that perform well.

Q1T1 4K still at ~<100/300 I see.

 

[TumbleGeorge]

Finally, Gen 5 SSDs on a modern node.

The fact some gen 5 SSDs till now are literally double the price of a gen 4, yet controller is built on ~12 nm etc., is a bit absurd, when the main downside is heat.

Not that anything other than sequential is going up by anything significant with gen 4 and gen 5 NAND flash drives. Controllers are better at working around the downsides of the tech though with clever algorithms.

TBH I'm more impressed with modern DRAMless drives that perform well.

Q1T1 4K still at ~<100/300 I see.

I'm not convinced that the problem with the Phison controller generating so much heat is so much related to the lithographic node. Hmm. Although...it doesn't seem to me to be 12nm, for example in its latest variant cortex R5(R52+, 2022) if this one is used, it is 16nm(R5 original from 2011 use 28nm). Not sure for lithography node of RISK V chip in the Phison E26 controller. Probably bad design.

 

[AnarchoPrimitiv]

Just please make an all SLC (with something like Kioxia's XL-Flash) drive for consumers in the 250-500GB capacity so that I can use it as an OS drive....and for the skeptics out there, below is a comparison of a P5800X Optane drive versus an all SLC drive from Dapustor....you can see how the performance is very close to optane performance. The "D:" drive is the all SLC and the "O:" drive is Optane.

 

[TheLostSwede]

I'm not convinced that the problem with the Phison controller generating so much heat is so much related to the lithographic node. Hmm. Although...it doesn't seem to me to be 12nm, for example in its latest variant cortex R5(R52+, 2022) if this one is used, it is 16nm(R5 original from 2011 use 28nm). Not sure for lithography node of RISK V chip in the Phison E26 controller. Probably bad design.

The SM2508 uses four Arm Cortex-R8 cores. So in theory, the SM2508 should run hotter if we're basing temps on core type, which isn't how it works.
The node matters a lot when it comes to thermals, but yes, the actual design matters too, as we've seen least not from AMD and Nvidia.

 

[phints]

Only 81 MB/s in 4k rand? Yikes this will be poor performance for daily tasks. We already see >100 MB/s with Gen 4 drives like the SN850X and 990 Pro. Hard pass on this, will wait for eventual WD/Samsung solutions.

 

[TheDeeGee]

Cool, but pass on anything but the stock low profile heatsink.

 

[TheLostSwede]

Only 81 MB/s in 4k rand? Yikes this will be poor performance for daily tasks. We already see >100 MB/s with Gen 4 drives like the SN850X and 990 Pro. Hard pass on this, will wait for eventual WD/Samsung solutions.

This isn't the final firmware and as I wrote, they're still working on optimising it. It also depends to a degree on the NAND flash used as well and we have no idea what was used here.

 

[b1k3rdude]

Just please make an all SLC (with something like Kioxia's XL-Flash) drive for consumers in the 250-500GB capacity so that I can use it as an OS drive....and for the skeptics out there, below is a comparison of a P5800X Optane drive versus an all SLC drive from Dapustor....you can see how the performance is very close to optane performance. The "D:" drive is the all SLC and the "O:" drive is Optane.
View attachment 350556

Do you happen to have the undurance plot over time for the above 2 drives..? as in when do the sequencials fall off a cliff...

 

[MxPhenom 216]

The SM2508 uses four Arm Cortex-R8 cores. So in theory, the SM2508 should run hotter if we're basing temps on core type, which isn't how it works.
The node matters a lot when it comes to thermals, but yes, the actual design matters too, as we've seen least not from AMD and Nvidia.

Its pretty 50/50. Load up a smaller node with more gates and clock controlled elements like flops and itll be prone to heat issues just the same. Which is very easy to do. Partly the reason why the drive to smaller geos even happens in the first place. Can shove more gates into the same footprint.

 

[Wirko]

Surprisongly little news from Phison. The only thing I saw was a single announcement of a new Gen 5 SSD from MSI (?) using the E31T, but that one is 4-channel and DRAM-less.

 

[persondb]

Finally, Gen 5 SSDs on a modern node.

What node does this one uses? I couldn't find it anywhere.

I don't think it's that much of a node issue with a lot of those controllers actually but the fact that the first ones basically took PCIe 4 SSD controllers and mostly upgraded the PHY and then had to clock higher to support those rates instead of doing new architectures.

I am honestly more interested in more efficient SSD controllers and NAND over PCIe 5 speeds, as of right now, there is very little point for most people, including gamers. Games don't see any difference between a PCIe 4 and 5 and actually even the difference of a good PCIe 3 SSD to 4 there will be barely any difference. And that is also because there are barely any games that can take advantage of such speeds.

 

[RogueSix]

What node does this one uses? I couldn't find it anywhere.

You couldn't find it anywhere? Dude? I know it's a pretty wild idea but how about reading the news article you are commenting on?

 

[phubar]

Just please make an all SLC (with something like Kioxia's XL-Flash) drive for consumers in the 250-500GB capacity so that I can use it as an OS drive

pSLC that can be done with regular TLC flash is fine. You don't need proper SLC for what you're after (performance). You probably don't want to pay the price either. Virtually no one else does. Which is a big part of the reason why its mostly gone from the market outside of niche industrial grade stuff.

Its also worth pointing out too that for common desktop work loads its going to be random read/write Q1T1 4K performance that matters far Far FAAAAR and away more than sequential performance. The drive you're posting is around half as fast as Optane there.

Also Optane, 1st or 2nd gen, still will probably still win out big time in practical performance for desktop work loads since its latency is going to be a fraction of any client grade SSD. Yes even PCIe 5 SSD's.

Its a fundamental limitation of the consumer grade client SSD flash and not anything to do with the controller. SLC flash also won't help much here either. Optane was uniquely fast here vs any other flash on the market. Everyone else is focused on trying to get costs down and cranking storage capacity not low latency. L1Tech has a great vid on this if you want more info.

I'd recommend anyone get one of those Optane 905p's that you can get for $300-400 if you want a SSD that is incredibly reliable and damn fast for typical desktop work loads. They're still basically unbeatable right now. Biggest issue with them is getting a u.2 adapter of some sort.

If you want any faster you'd have to splurge for a P5800X.

 

[Wirko]

Just please make an all SLC (with something like Kioxia's XL-Flash) drive for consumers in the 250-500GB capacity so that I can use it as an OS drive....and for the skeptics out there, below is a comparison of a P5800X Optane drive versus an all SLC drive from Dapustor....you can see how the performance is very close to optane performance. The "D:" drive is the all SLC and the "O:" drive is Optane.
View attachment 350556

This is from the ServeTheHome's review of the Dapustor 800GB SSD. The result is hardly exciting; 50% advantage over a good TLC SSD in QD1 random reads, and AS SSD results are lower still? Is that all SLC has to offer? On the other hand, enterprise drives are not supposed to be optimised for QD1, and also the reviewer doesn't seem to care about that (he just notices good QD1 write performance).

Price unknown of course ... if anyone asks "how much?", the answer is "then go buy a Transcend". Transcend has a single current SLC model with a PCIe interface, the MTE560I, and I guess we won't se many reviews of it to learn about performance. Anyway, it's rated for 60,000 rewrite cycles and is about ten times as expensive per GB as a good TLC SSD.

pSLC that can be done with regular TLC flash is fine. You don't need proper SLC for what you're after (performance). You probably don't want to pay the price either. Virtually no one else does. Which is a big part of the reason why its mostly gone from the market outside of niche industrial grade stuff.

Do you know if anyone still manufactures proper SLC NAND chips anymore? Also, is there any fundamental difference between proper SLC and pseudo SLC? I suspect not, and TLC/QLC is just more modern and made to tighter tolerances than the old SLC, which makes it possible to resolve 8 or 16 voltage levels on reading.

Its also worth pointing out too that for common desktop work loads its going to be random read/write Q1T1 4K performance that matters far Far FAAAAR and away more than sequential performance.

True. At least that's what we (and me too) usually think. But it would be nice if someone could provide a more detailed analysis. Random reads in CDM, AS SSD etc. are fully random over a multi-gigabyte area, or over the entire disk. That's fine but these are worst-case results. Random reads in the OS, aplications and games are very probably less random or close to sequential (operations on many files in a single folder, where the files aren't totally fragmented).
Finally, if anyone is writing any kind of program that reads a lot of random data, it would be by far fa-a-a-a-ar the best if they took a trip to the hell and back and learn proper code design with queued access, not wait for technology to improve because it won't.
Random writes are always less problematic and faster than reads because they are cached.

The drive you're posting is around half as fast as Optane there.

Also Optane, 1st or 2nd gen, still will probably still win out big time in practical performance for desktop work loads since its latency is going to be a fraction of any client grade SSD. Yes even PCIe 5 SSD's.

Its a fundamental limitation of the consumer grade client SSD flash and not anything to do with the controller. SLC flash also won't help much here either.

Enterprise grade won't help much either, also because enterprise mostly isn't interested in QD1, at least not for server purposes (HPC and ML may be different). I count the outliers like the Samsung 983 ZET as exotic grade.

Optane was uniquely fast here vs any other flash on the market. Everyone else is focused on trying to get costs down and cranking storage capacity not low latency.

Rest assured that Intel and Micron were also focused on trying to get margins up and cranking capacity. Apparently they didn't find a practical way to build Optane in many layers (they made a two-layer model, not sure if it was a prototype or a production model). Plus those patent issues of course.

 

[TheLostSwede]

Do you know if anyone still manufactures proper SLC NAND chips anymore? Also, is there any fundamental difference between proper SLC and pseudo SLC? I suspect not, and TLC/QLC is just more modern and made to tighter tolerances than the old SLC, which makes it possible to resolve 8 or 16 voltage levels on reading.

Yes, SLC NAND is still being produced.

SLC NAND flash memory part catalog

www.micron.com

SLC Nand | Nand Flash | Samsung Semiconductor Global

Samsung SLC Nand, suited for high-level performance, manages industrial devices that call for data integrity and top-grade reliability.

semiconductor.samsung.com

The issue is that the largest NAND chips appear to be 16 GB and one 16 GB Micron NAND chip retails for €112.70 if you buy a roll of 1k.
That means a 500 GB SLC NAND based SSD would end up costing somewhere in the region of €4,000.

And there's at least one difference between proper SLC and pSLC and that's the size of the NAND chips. The other difference is the P/E cycles are about half for pSLC vs SLC.

Pseudo SLC (pSLC)

Discover Exascend's Pseudo Single-Level Cell (pSLC) technology, delivering enhanced durability, higher performance, and balanced cost-effectiveness for demanding applications.

exascend.com

4 reasons to use industrial 3D TLC NAND flash as pseudo-SLC

The ability to operate TLC in pseudo-SLC mode with one bit per cell continues to carve out a niche existence. However, for a vast number of applications the economic advantages of this configuration outweigh the disadvantages. We've compiled four arguments to consider.

www.swissbit.com

 

[Wirko]

The issue is that the largest NAND chips appear to be 16 GB and one 16 GB Micron NAND chip retails for €112.70 if you buy a roll of 1k.
That means a 500 GB SLC NAND based SSD would end up costing somewhere in the region of €4,000.

It would also have a maximum transfer rate of 1250 MB/s at that price, haha. I calculated that from Samsung's data (40 Mbit/s x 8 bit = 40 MB/s per chip), although all listed Samsung's SLC chips have reached EOL. The "high performance" touted by both manufacturers is only there when compared to NOR flash, not to anything modern.

This is 25nm planar NAND, it seems to be a close relative to the last planar NAND that Intel put in their SSDs a decade and half ago (but that one was MLC). It hasn't seen much development since. It's intended for microcontrollers, which likely don't even run an OS, and can't be bothered by file systems, let alone wear leveling. It's basically unusable in anything with PC architecture. Not only is the price per GB extremely high; the price per TBW is very high too, compared to today's TLC SSDs.

The article by Swissbit that you linked to nicely sums up these same things.