Silicon Motion's PCIe 5.0 SSD controller is finally coming in Q4 — low-power SSDs will leverage SM2508
Silicon Motion's SM2508-based SSD can achieve a 14.9 GB/s speed at around 7W.
Silicon Motion re-emphasized its plans to release the SM2508 platform for PCIe Gen5 client SSDs by the end of the year and revealed some additional details about the controller and its expected performance.
The Silicon Motion SM2508 SSD controller is based on four Arm Cortex-R8 cores, which makes it one of the most capable controllers on the market now — even when compared to some of the enterprise-grade platforms. The chip boasts eight NAND channels with 32 CE targets, each supporting interface speeds of up to 3600 MT/s, and can achieve sequential read and write speeds of up to 14.5 GB/s and 14 GB/s, along with random read and write speeds of up to 2.5 million IOPS.
At least on paper, the SM2508 offers considerably higher random performance than Phison's PS5026-E26.
Another key aspect of the SM2508 is that it is made on TSMC's N6 process technology, which allowed the company to reduce its peak power consumption to around 3.5W. According to the company, drives based on this controller will consume around 7W of power under loads — lower than that of existing competitors.
At present, SMI and its partners are polishing off the firmware of the SM2508 SSD controller and are qualifying it with various types of memory, including 3D TLC NAND for high-performance SSDs. The drive can support 3D QLC NAND as well, though it remains to be seen when SSD makers will adopt this type of memory for high-performance PCIe Gen5 drives.
Speaking of memory, the fastest 3D NAND flash memory devices on the market today (at least in more or less significant quantities) have a 2400 MT/s interface. Using such memory is essential to fully saturate a PCIe 5.0 x4 interface and achieve sequential read/write performance between 13 and 14 GB/s. The support for a 3600 MT/s ONFI/Toggle DDR interface will allow for the building of higher-end SSDs without using numerous memory devices, which is critical as next-generation 3D TLC devices have capacities that start at 1TB and get even bigger.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.
- Paul AlcornManaging Editor: News and Emerging Tech
-
bit_user That's progress. So, what happens to power consumption when you run it at PCIe 4.0 speed?Reply
I guess we should also like to know about 5.0 x2 - is there any power savings from using fewer lanes? -
DougMcC
Almost certainly you'd save power either way. But who is going to build x2? x4 is completely entrenched. x8 is far more likely than x2, since the bragging rights from the faster speed might actually sell units.bit_user said:That's progress. So, what happens to power consumption when you run it at PCIe 4.0 speed?
I guess we should also like to know about 5.0 x2 - is there any power savings from using fewer lanes? -
bit_user
I have a mini PC with a PCIe 3.0 x2 M.2 slot. I've even seen a couple boards that have just x1!DougMcC said:But who is going to build x2? x4 is completely entrenched.
Uh, well that'd have to be in an AIC form factor, because neither M.2 nor U.2 support more than x4.DougMcC said:x8 is far more likely than x2, since the bragging rights from the faster speed might actually sell units.
I think you'll also find a lot of desktop users are reluctant to run an x8 AIC, because that means cutting down their GPU to run at just x8 lanes. For some of the fatter GPUs, the slot might even be blocked! That means it would be just a niche product for servers and workstation users, and thus it's not a good way to "sell units". -
Makaveli The sequential speeds are great but there has been no movement in RND4k Q1T1 speeds in like a decade now. With Intel optane being the lone exception.Reply -
Notton It's always nice when RND4k is faster, but games already load insanely fast on a WD SN850X.Reply
I don't think I would feel 0.05s faster loading times.
and if I was running a program that required fast RND4K, I would get the appropriate server grade SSD. -
dimar SN850X is 6.8W max, 990 Pro 6W. So 7W for such speed is not bad I guess.Reply
I'd like to know the TBW info as well.
I'd like to have some kind of battery saving mode on SSD where the power goes to 0.5W for example when needed. -
bit_user
Weren't there some drives that used SLC to achieve nearly Optane-level performance?Makaveli said:The sequential speeds are great but there has been no movement in RND4k Q1T1 speeds in like a decade now. With Intel optane being the lone exception.
Unless you're running something like a database server, you probably don't need super-low latency as much as you think you do. And if you are, then what you probably care about even more are tail latencies. -
bit_user
A lot of drives will go even lower, if you have ASPM enabled. For something like a laptop, idling at even 0.5 W is actually rather a lot.dimar said:I'd like to have some kind of battery saving mode on SSD where the power goes to 0.5W for example when needed.
This might be informative, but beware that it's provided by a SSD manufacturer:
https://sabrent.com/blogs/storage/ssd-power-consumption
The reason desktop users might not want to enable ASPM is that it measurably hurts responsiveness. It takes time for the drive to climb back up to higher power states. -
Diogene7 Makaveli said:The sequential speeds are great but there has been no movement in RND4k Q1T1 speeds in like a decade now. With Intel optane being the lone exception.
I agree as well that it is a shame since Intel Optane is gone that there isn’t any lower latency persistent memory on the market :(…
I hope that CXL will help enable the introduction of new persistent memory with low latency that will likely be very expensive at first (ex: Nantero carbon nanotubes NRAM, Weebit Nano ReRAM, Avalanche technology MRAM, Antaios SOT-MRAM,…)
I also wish that the work done by the European research center IMEC on SOT-MRAM would help create DRAM-like Non-Volatile-Memory (NVM) : that would open so many new opportunities.
https://www.imec-int.com/en/press/imecs-extremely-scaled-sot-mram-devices-show-record-low-switching-energy-and-virtually -
dimar
Right, but I meant max power, not idling.bit_user said:A lot of drives will go even lower, if you have ASPM enabled. For something like a laptop, idling at even 0.5 W is actually rather a lot.
This might be informative, but beware that it's provided by a SSD manufacturer:
https://sabrent.com/blogs/storage/ssd-power-consumption
The reason desktop users might not want to enable ASPM is that it measurably hurts responsiveness. It takes time for the drive to climb back up to higher power states.