IN A SHOCK to no one, SandForce announced the SF-2000 family of SATA3 SSD controllers. These new parts basically double throughput, and add a few new features as well.
If you are familiar with the Sandforce SF-1000 line of controllers, you already know most of what the SF-2000 brings. If you can’t quote SSD controller specs off the top of your head, read this, and you will be up to speed. The short story is that SandForce brought two new tricks to the table, wire speed encryption and the ability to make MLC flash last as long as SLC with older controllers. Both are game changers.
SF-2000: The block diagram
There are three chips in the first batch of SF-2000 products, all of the variants are aimed at enterprise and industrial uses. The SF-2300 is targeted at industrial and embedded markets meaning it has a hardened package. This means higher temperature ratings and resistance to the environmental indignities that often come the way of the devices like this.
The SF-2500 is the enterprise version, as close to a vanilla SF-2000 chip as you are going to see for now. Moving up to the SF-2600, you get SAS-6 support and non-512 byte sector capabilities as well. This is what the trolls running your ERP system drool over, it supports all the ‘weird’ drive parameters that are so necessary in many high availability applications.
Other than those differences, the silicon is the same for all three parts, just different fuses blown and different physical packaging. All use the same 14x14mm 400 ball TFBGA pinouts as well. Any differentiation is at the firmware level, and that can be quite substantial.
So, what does the SF-2000 line bring? The first obvious one is SATA-3’s 6Gbps transfer rates. The current ‘second gen’ consumer SSDs were mostly limited by the SATA-2 transfer rate. Real world read speeds capped out at 250-280MBps, about the theoretical limit of the bus. How much more usable speed does double the transfer rate get you? Basically double.
How fast is it?
SandForce is claiming 60K IOPS (4K random R/W) and 500MBps sequential read and write performance. The important thing here is the write performance not being substantially different from the read. Most drives have a notably slower write speed, so having them equal means the controller is not sweating much, it is once again bus bound. SATA-3 is maxed out before it is widely available. SIGH.
The next big thing that the SF-2000 line brings is encryption, up to AES-256 at wire speed. To make matters more fun, SandForce has ‘double’ encryption. If you recall, the SF-1000 line could do AES-128, and the new one still has that same unit.
Due to an apparent horde of FIB machine bearing street thugs that roam around Fortune 500 company data centers, newer encryption and security standards state that the data coming off the SATA bus must be encrypted effectively immediately. SandForce’s SF-1000 line had the encryption engine as the last step before writing to flash, and that was not good enough for the newer security regulations, specifically the TCG specs. To placate the stupid, SandForce had to put another engine between the SATA circuitry and the rest of the chip.
Since the old one was there, working, and pretty small, they just left it in. That is why you have the double encryption. The people who are desperately afraid of the aforementioned FIB wielding street thugs probably still run Windows, so their machines are unsecurable regardless of hardware encryption. Why bother with a FIB when you can download any of 300 root kits and take the info you wanted, and more, remotely. Double SIGH. In any case, it is NIST XTS approved, you can tell by the little ‘Maginot’ symbol on the packaging.
Somewhat related to security is error correction. SandForce is big on error correction on the flash side, that is a large chunk of their RAISE technology. The other side of the chip has a SATA interface, and now that supports ECC, all the SAS error correction protocols, enhanced ECC with BCK and 55 bits per sector, and more. The SF-2000 is much closer to ‘good enough’ this time around.
One other bit that is quite interesting seems to be a unique feature to SandForce, performance throttling. With some non-standard SATA commands, you can set the drive to go into lower power modes. Slowing down a drive may not seem like a smart thing to do, but it has a really good reason to exist. Both Intel and AMD are pushing for servers, and to a lesser extent PCs, to cap their power use in a closed loop. You can set most modern servers to throttle themselves back once they hit certain power draw. This is useful at the data center level to prevent brownouts and hot spots. It is one of the hottest, pun intended, areas of data center management.
If you throttle the server back, is there a point to having a storage subsystem capable of sending it more data than it is able to process? The idea with Power Performance Throttling is to have the system drop hints to the drive about power states, and have the drive go into lower power modes as well. It may only save a few watts, but a few times 10,000 servers adds up. This will be commonplace in a generation or two.
SandForce is promising a wide range of flash compatibility from basically every vendor that matters. The SF-2000 line will work with SLC, MLC, and eMLC chips on a number of interfaces. Speeds of up to 166MT are now supported as well. This may not matter to the end user, but it makes a big difference to the vendors. Flexibility is a good thing here.
In the end, SandForce did what you knew they were going to do, a bit more of everything, and some new twists. Speed is basically doubled and the SATA bus is the bottleneck once again. Read and Write speeds are capped by SATA or SAS, and everything can be encrypted, twice, at wire speeds. What’s not to like in the SF-2000 line?S|A
Latest posts by Charlie Demerjian (see all)
- Intel unleashes more Kaby Lake SKUs on the yearning public - Jan 4, 2017
- Qualcomm opens up a bit more on the 10nm Snapdragon 835 SoC - Jan 3, 2017
- AMD’s Freesync 2 changes the display game - Jan 3, 2017
- Coffee Lake points to issues with Intel’s 10nm process - Dec 28, 2016
- Intel goes all Pokemon with code names – really - Dec 28, 2016