The Samsung SM961 will be Samsung’s new top-of-the-range M.2 SSD line for OEMs, which will be offered in 128 GB, 256 GB, 512 GB and 1 TB configurations (by contrast, the SM951 family did not include a 1 TB option). The drive will be based on Samsung’s MLC V-NAND as well as the company’s Polaris controller. Samsung is specing the SM961 at up to 3200 MB/s for sequential reads and up to 1800 MB/s for sequential writes, but does not specify which models will boast with such numbers. The new SSDs can perform up to 450K random read IOPS as well as up to 400K random write IOPS, which looks more like performance of server-grade SSDs.
Originally 6-7 years ago when SSDs were offered up in very small sizes say 32GB or 40GB, it was painfully obvious they were a niche product. If you could get WindowsXP to run in 32GB and save all your data to a USB Thumb drive you might be able to downgrade a laptop and gain some speed. Ditching a spinning magnetic hard drive was a guarantee you might get a boost (even 10% might be worth the trouble of swapping to a smaller SSD). Data rates too improved each time a new product was announced. Independent hardware designers like Crucial and OCZ were putting together various suppliers chipsets and NAND controllers and getting reviews in all the hardware comparison sites. Everyone was waiting for the next gen SandForce NAND controller to see how much better it would be at Random Reads/Writes. Progress was slow but steady, prices were fairly steady too, no big drops as new gen hardware hit the shelves of online store fronts.
Eventually bigger names got in the game like longtime stalwart SanDisk, but then too Intel and Samsung. Top performers were going after the high, high end with PCI based Flash drives running on PCI Express and getting double to triple the speed of a SATA based SSD. Eventually SATA became SATA2, and PCI Express became PCIe 2.0. All these entered into the mix of higher performance and still very steady prices especially on the high-end. You almost always had to pay $1200 to get a halfway speedy PCIe SSD. But the true benefit was the amount of performance you got over the ACHI/SATA interface built into the motherboard. I wrote, linked and commented a lot on the way forward, urging anyone reading this to keep looking ahead to the performance end of the spectrum.
By the time Samsung and Intel really started ramping up product the middle ground and bottom feeders of the SSD market were starting to compete on price alone. SATA based SSDs hit hard limit of ~250MB/sec up to 500MB/sec random read and write. And more likely it was below 500MB/sec as that was the physical upper limit of the speed of a SATA hard drive (due in no small part to the design being for spinning magnetized platters). Suffice to say Samsung and Intel could bide their time making new NAND chips and memory controllers and plopping them in new products for the high/middle markets for SSD. Eventually they could license these same technologies to the bottom end of the market after the price premiums were collected on sales of newly announced products. It became clear Intel and Samsung were owning SSDs but eventually too they would dominate PCIe SSDs and a nascent M.2 form factor for small light laptops.
All that is what has proceeded today’s press release by Samsung. This product announcement (quoted at the top of this article) follows fast on the heels of yesterday’s (Tuesday March 22) announcementof another SSD. That one was an SSD on a chip in essence. By itself that announcement was reason enough to share it with anyone I knew who follows technology news. But today’s announcement (the one which is more interesting) further emphasizes Samsung is on a tear and is single-handed pushing the technology to it’s physical limits. What does that mean?
The original SSDs we knew from 6-7 years ago ran in the range of 150MB/sec random read/write. That eventually would top out and has stayed at ~500MB/sec. A SATA based SSD will never exceed that limit. PCIe 3.0 devices with 4 data lanes have a top limit too. And the M.2 form factor can use that very interface as it’s bridge into the motherboard. Fastest consumer level PCIe SSDs I remember seeing were always in the 1,000MB/sec range as high as 1,200MB/sec and again they were expensive, always around $1,200 USD. But now as of today Samsung has designed and produced a device that does 3,000MB/sec and has utterly saturated the design spec for PCIe 3.0 at 4X. The IOPs, the benchmark measure of any storage based chip, hard drive technology is also massive with the new devices. Again 6-7 years ago there were shootout/benchmark competitions to see who could build the first 1Million IOPs storage array. This was for enterprise architects who had FiberChannel switches, and SAS spinning disks running in parallel with all kinds of RAID storage controllers, and file servers doing the tests.
As a first, 1Million IOPs (input-output operations) was a big deal and millions of dollars were spent trying to hit that mark or show that it could be hit with a some mix of a vendor’s product. Today, with Samsung’s announcement of the SM961 M.2 drive you’ve got in the size of a stick of gum a device that will theoretically do 1/2 of 1Million IOPs. Two devices paired up in a RAID 0 config WILL do ~1Million IOPs in the space of “2 sticks of gum”. That is a giant leap in the 6-7 years in which NAND based memory controllers and chip packaging and memory cells have been designed. All one can say is now theoretically speaking hard drives are no longer the performance choke point for a desktop or even a laptop computer. We have storage that can move data in and out of memory at 3,000MB/sec. What’s the next hurdle? What possible hard limit in computer technology is the next one that needs to be addressed? Is it power? Does all this need to run off of a thin/light Lithium Ion battery? I don’t know. But I can imagine there’s some design team working on it right now. I just feel we’re coming to the end of a long evolution in the fundamental building blocks of desktop PCs. It’s all small refinements and polishing now that the heavy lifting is done.