Previewing Silicon Motion SM2260 NVMe Controller With 3D M…

Silicon Motion’s SM2260 is their first NVMe PCIe SSD controller. We’ve already reviewed one SSD using this controller: the Intel SSD 600p. That drive used Intel’s 3D TLC NAND and set a new low for NVMe SSD prices while also offering performance that is mostly beyond the reach of any SATA SSD. Today we have an engineering sample from Silicon Motion that pairs the SM2260 controller with Micron’s 3D MLC NAND—our first look at the performance of the Intel/Micron 3D MLC and a preview of what to expect from products like the ADATA XPG SX8000.

One of the first announced design wins for the SM2260 controller was Micron’s Ballistix TX3 SSD, which was intended to be their first consumer PCIe SSD and a flagship product for their overhaul of the Ballistix brand. The drive was originally intended to hit the market in September 2016 but was canceled before launch. Since then, Micron has not introduced any consumer products using the MLC version of their 3D NAND, leaving the 3D TLC-based Crucial MX300 as their top consumer SSD. Intel shipped their 3D TLC-based 600p with a customized SM2260 controller, but the aging SSD 750 remains their high-end consumer NVMe offering.

With a controller that was spurned by Micron and 3D MLC NAND that its manufacturers have not tried to introduce to the consumer market, this engineering sample offers us some insight into Intel and Micron’s consumer SSD strategies and the opportunity they passed up. It also offers us a preview of ADATA’s XPG SX8000, their first NVMe SSD and one that is based on this controller and NAND flash combination. ADATA is the first brand to announce products with Micron’s 3D MLC for the consumer market, with the XPG SX8000 NVMe SSD and the Ultimate SU900 and XPG 950 SATA SSDs, all using Silicon Motion controllers.

ADATA usually has a broader range of controller and NAND combinations than any other brand, but so far Micron is their only choice for 3D NAND for SSDs and Silicon Motion is their controller vendor. This is due in large part to Silicon Motion being the first controller vendor ready with a complete hardware and firmware reference design tuned for Micron’s 3D NAND. At CES earlier this year, we also saw Mushkin announce SATA and NVMe SSDs based on the combination of Micron 3D NAND and Silicon Motion controllers.

The SM2260 controller supports a PCIe 3.0 x4 interface and implements the NVMe 1.2 protocol. The controller has a dual-core ARM Cortex-R processor and is manufactured on TSMC’s 40nm LP process, an interesting juxtaposition when you consider that most of its competition is on 28nm. The controller is packaged with a thin copper heatspreader that we first saw relatively late in the chip’s design cycle. LDPC error correction and TCG Opal encryption are supported.

The sample we are reviewing today is a M.2 2280 SSD with a total of 16 of Micron’s 256Gb 3D MLC dies spread across four packages and the controller’s 8 channels. This amount of flash would typically be used for a drive with an advertised capacity of 480GB, 500GB or 512GB, but this sample has a usable capacity closer to 515GB. This means its spare area is a little smaller than is normal—about 6.31% rather than the more common 6.85%. The firmware implements a dynamically-sized SLC cache whereas the Intel SSD 600p used a fixed-size cache.

Silicon Motion will surely face more competition this year for design wins for 3D NAND SSDs. Micron used a Marvell controller for their own Crucial MX300 SATA SSD, the only shipping SSD with Micron’s 3D NAND that doesn’t use a Silicon Motion controller. We expect Phison’s controllers to be a popular choice once Toshiba has 3D NAND for the SSD market and Plextor is likely to continue using a mix of Marvell and Silicon Motion controllers when they begin transitioning to 3D NAND. This year we will probably also see Maxiotek score a design win for their MK8115 DRAM-less SATA SSD controller that is primarily intended for use with 3D TLC NAND. But in the meantime, Silicon Motion’s SM2260 is the NVMe controller that’s already shipping paired with 3D NAND.

The main competition we’ll be comparing this sample against includes:

• The Patriot Hellfire 480GB, using the Phison PS5007-E7 controller and Toshiba 15nm MLC
• The Intel SSD 600p 512GB, using the SM2260 controller and Intel 3D TLC
• The Plextor M8PeG(N) 512GB, using the Marvell 88SS1093 controller and Toshiba 15nm MLC
• The Samsung 960 EVO, using Samsung 3D TLC, tested in 250GB and 1TB capacities

• Thanks to Intel for the Core i7-4770K CPU
• Thanks to ASUS for the Z97 Deluxe motherboard
• Thanks to Corsair for the Vengeance 16GB DDR3-1866 DRAM kit, RM750 power supply, Carbide 200R case, and Hydro H60 CPU cooler

Our performance consistency test explores the extent to which a drive can reliably sustain performance during a long-duration random write test. Specifications for consumer drives typically list peak performance numbers only attainable in ideal conditions. The performance in a worst-case scenario can be drastically different as over the course of a long test drives can run out of spare area, have to start performing garbage collection, and sometimes even reach power or thermal limits.

In addition to an overall decline in performance, a long test can show patterns in how performance varies on shorter timescales. Some drives will exhibit very little variance in performance from second to second, while others will show massive drops in performance during each garbage collection cycle but otherwise maintain good performance, and others show constantly wide variance. If a drive periodically slows to hard drive levels of performance, it may feel slow to use even if its overall average performance is very high.

To maximally stress the drive’s controller and force it to perform garbage collection and wear leveling, this test conducts 4kB random writes with a queue depth of 32. The drive is filled before the start of the test, and the test duration is one hour. Any spare area will be exhausted early in the test and by the end of the hour even the largest drives with the most overprovisioning will have reached a steady state. We use the last 400 seconds of the test to score the drive both on steady-state average writes per second and on its performance divided by the standard deviation.

The SM2260 sample has relatively poor steady-state random write performance given that it uses NVMe and MLC NAND. The use of SLC caching and the lower than normal spare area of this drive both contribute to poor steady-state performance but may not significantly impair short-term performance.

The SM2260 sample has an even lower consistency score than the Intel SSD 600p, which uses basically the same controller and TLC NAND, but has substantially more spare area.

Default
25% Over-Provisioning

After the very short initial burst of great performance around 140k IOPS, the SM2260 sample transitions abruptly to a steady state that it maintains throughout the rest of the test with no long-term shifts in behavior. With extra overprovisioning reserved, there’s an intermediate phase consisting of mostly performance around 80k IOPS and second burst at 140k IOPS before a higher performance but no more consistent steady state is reached.

Default
25% Over-Provisioning

Looking more closely at the steady state, the SM2260 sample is mostly shifting between four performance levels, with the most common being around 5k IOPS. However, when it goes through periodic phases of lower performance, it is stuttering hard and will often go for an entire second without completing any I/O. This is clearly poorly-managed garbage collection, possibly exacerbated by thermal throttling and definitely suffering from insufficient spare area.

With more overprovisioning, the severe stuttering is all but eliminated and the normal performance range jumps to around 24k IOPS with periods where it drops to around 6k IOPS.

The Destroyer is an extremely long test replicating the access patterns of very IO-intensive desktop usage. A detailed breakdown can be found in this article. Like real-world usage and unlike our Iometer tests, the drives do get the occasional break that allows for some background garbage collection and flushing caches, but those idle times are limited to 25ms so that it doesn’t take all week to run the test.

We quantify performance on this test by reporting the drive’s average data throughput, a few data points about its latency, and the total energy used by the drive over the course of the test.

The SM2260 sample’s average data rate on The Destroyer is just a hair slower than the Phison-based Patriot Hellfire. This makes the SM2260 sample the slowest NVMe SSD using MLC NAND, but it’s still faster than any SATA SSD.

The SM2260 sample’s average service time during The Destroyer is again in last place for NVMe/MLC SSDs, but the SATA SSDs and some of the TLC-based NVMe SSDs are trailing behind by a wide margin.

The SM2260 sample is not great at avoiding high-latency outliers above 100ms and ranks behind the Samsung 850 PRO. At the 10ms threshold, the SM2260 sample performs quite well with big advantage over the Patriot Hellfire, the Intel SSD 600p and the SATA SSDs.

The SM2260 sample with 3D MLC improves significantly over the poor power efficiency of the TLC-based Intel SSD 600p, but it still ranks poorly overall. It is tied with the Phison E7-based Patriot Hellfire.

Our Heavy storage benchmark is proportionally more write-heavy than The Destroyer, but much shorter…