SATA Express aims to unite the worlds of Serial ATA and PCI Express. The standard was introduced in 2011 and ratified two years later. Now, at last, the first compatible devices are nearly upon us. Select motherboards based on Intel's
8+1 Series "next-gen" chipset have SATA Express connectors onboard. Asus is also working on a Hyper Express storage device that uses multiple SSDs to exploit the faster interface.
The Hyper Express drive is due this quarter, and I've been using a prototype to test upcoming motherboards. Expect full coverage of those boards soon. In the meantime, you can whet your appetite with a quick look at the drive.
Before we delve into specifics, it's worth clarifying a few things about the SATA Express standard. According to the SATA-IO group behind the spec, "The SATA Express environment is pure PCIe." It comprises dual lanes with support for Gen3 speeds. The fastest implementations can offer up to 2GB/s of bandwidth, while those based on Gen2 PCIe are capped at 1GB/s.
Although the standard is based on PCIe, the SATAe physical interface also accommodates Serial ATA devices. On a motherboard, the SATAe port is essentially two SATA ports plus a smaller, third connector to the left. Existing SATA drives should work in the port's, er, ports, without issue.
SATA Express hosts must support both SATA and PCIe drives. SATA drives will be limited to the AHCI interface, while PCIe implementations can choose between AHCI and NVM Express. AHCI was conceived when mechanical hard drives were the norm, so it's poorly suited to the radically different performance characteristics of solid-state storage. NVMe, on the other hand, was architected specifically for non-volatile memory like NAND flash. The interface promises lower latency than AHCI in addition to a simpler command set and reduced software overhead.
Windows 8.1 includes native drivers for both AHCI and NVMe, so SATA Express devices of either persuasion should work with the OS sans additional software. Device makers are free to provide their own drivers, of course. We didn't get any with Asus' Hyper Express prototype, which appears in the Windows Device Manager hanging off a "standard SATA AHCI controller" connected to the chipset's PCIe interface.
While the Hyper Express appears as a single drive, it's actually fueled by two. The 2.5" prototype stripes a pair of mSATA SSDs in a RAID 0 array powered by an ASMedia ASM1062R controller chip. The chip isn't listed online, but it seems to be an AHCI variant with dual Gen2 PCIe lanes and dual 6Gbps SATA ports. As for the drives, the unit we've been testing came loaded with a pair of Kingston SSDNow mS200 120GB SSDs. These SandForce-based offerings aren't the fastest mSATA parts around, but users should be able to spec whatever they want with the finished product.
That final product will be housed in a larger 3.5" case. Two configurations are slated for production: one that accommodates a pair of 2.5" SATA drives, and another with pairs of M.2 and mSATA slots for mini SSDs. Both variants will apparently rely on the same RAID setup, but they'll be sold as barebones kits, without drives installed. Pricing hasn't been set.
Striping drives is a great way to improve performance, but there's a gotcha for SSD arrays. RAID controllers don't always know what to do with the TRIM command, which is used to combat flash memory's block-rewrite penalty. The Hyper Express currently lacks TRIM support, so it's forced to rely on the SSDs' internal garbage collection routines to clean up unused flash pages. Asus concedes that TRIM is vital, and it sounds like the final versions of the Hyper Express will implement the feature.
Interestingly, the Hyper Express does employ SRIS, a PCI Express provision otherwise known as Separate Refclock with Independent SSC (Spread-Spectrum Clocking). PCIe devices normally sync with a 100MHz reference clock that's passed over the interface by the host. That's fine for slot-based solutions, but it's trickier for cabling, which requires costly shielding to handle that sort of signal. SRIS employs a separate clock generator on the device, removing the need to pass the reference signal over the interface.
In addition to supporting SRIS at the device level, Asus has integrated the feature into its motherboard firmware. SRIS was implemented with cooperation from Intel as part of an effort to validate SATA Express. "Our SATA Express solution is the only one to have undergone full SRIS testing and validation to ensure the true potential of the interface is realized," Asus says.
SATA Express devices will ship with their own cables, so there's an incentive for other drive makers to support SRIS. We're waiting to hear back from other motherboard makers to see if they're following suit.
By the way, the cable that came with the prototype is pliable, neatly sheathed, and equipped with its own little power dongle. The PSU's power connector plugs into the dongle rather than the drive.
Before I share a quick benchmark result, keep in mind that this is pre-production hardware. The drives are also in a used state. My usual secure-erase methods don't recognize the Hyper Express, though Asus says the firmware in its newer ROG motherboards has an integrated tool that does. Either way, the numbers are quasi-meaningless. Make of them what you will.
Yep, the Hyper Express prototype is faster than a typical 6Gbps SATA drive. It will be interesting to see how the final version stacks up, especially versus a comparable RAID config tied to the Intel chipset's RAID controller. Intel's RST drivers have supported TRIM for RAID 0 arrays since 2012, so the built-in alternative is already one step ahead.
Still, Asus deserves credit for being the first to show off SATA Express hardware—and for implementing SRIS. We'll soon see whether the company's focus on SATAe translates to a superior implementation on next-gen motherboards. Stay tuned.
46 comments — Last by TwistedKestrel at 11:18 PM on 05/06/14
|OCZ's Vector 180 solid-state drive reviewedBarefoot goes bigger||33|
|The SSD Endurance Experiment: They're all deadThis is the end, beautiful friend||179|
|A first look at USB 3.1 performanceWith bonus Type-C connector glamor shots||51|
|Crucial's BX100 and MX200 solid-state drives reviewedBrothers from different mothers||50|
|Asus' Crossblade Ranger motherboard reviewedSocket FM2+ joins the Republic of Gamers||37|
|MSI's X99S MPower motherboard reviewedHaswell-E noir||5|
|Some 840 EVOs still vulnerable to read speed slowdownsPatched drives exhibit problems with old data||115|
|Samsung's Portable SSD T1 reviewedA pocketable 850 EVO||34|
|The TR Podcast is live, so come ask us stuff!||0|
|AMD shows off DirectX 12 performance with new 3DMark benchmark||35|
|Intel and Micron sampling 3D NAND based on floating gates||16|
|Report: Microsoft to build an Intel-powered, non-Pro Surface tablet||47|
|Toshiba's 3D flash spreads 16GB over 48 layers||6|
|Cougar's 300M gaming mouse looks awfully familiar||14|
|Wednesday Night Shortbread||69|
|EVGA Hybrid liquid-cools the GeForce GTX 980||20|
|Rowhammer attack exploits shrinking process size in DRAM||40|