OCZ'z Vertex 3 Pro Demo: World's First SandForce SF-2000
by Anand Lal Shimpi on January 5, 2011 9:37 PM ESTA few months ago SandForce announced its second generation SSD controller: the SF-2000 series. The specs SandForce released at the time were almost too good to be true. As a refresher here’s what we saw back in October of last year:
SandForce is promising a single enterprise level drive that can deliver 500MB/s sequential reads and writes (for highly compressible data), and up to 60K IOPS for 4KB random reads and writes. That’s not an evolutionary improvement, that’s more than a doubling of what most of the competition can do today. Even compared to existing SandForce drives it’s a huge increase in performance. But as I’ve heard many times before, anyone can put out a promising PDF.
Today at CES, OCZ previewed its first SF-2000 based drives: the Vertex 3 Pro and Vertex 3 EX. Both are based on SandForce’s SF-2582 controller, the highest end offering in the SF-2000 family. The drives won’t see the light of day for months (sometime in Q2) and what OCZ is showing today is very, *very* early silicon and hardware. The drives are using 32nm Toshiba toggle-mode NAND (effectively DDR NAND), however OCZ will go to market with 25nm Intel NAND when the drive is ready.
First let’s look at the specs OCZ is promising for these drives:
OCZ’s specs are even higher than SandForce’s. This is getting ridiculous. Thankfully, OCZ let me run some of my own Iometer tests on the drives to verify the claims. Surprisingly enough, the Vertex 3 Pro looks like it’s really as fast as OCZ and SandForce are claiming. When running highly compressible data (pseudo random in Iometer) at low queue depths, I get 518MB/s sequential write speed and nearly 500MB/s for sequential read speed. Remember this is the very first version of the drive and there’s months of tweaking ahead to get it ready for production. Performance may even increase by the time OCZ actually ships the drive. Furthermore, this is the performance of a single drive with a single controller - there’s no funny on-board RAID going on, we’re just talking about the performance of a single drive.
| OCZ Vertex 3 Pro Preliminary Test Data | ||||
| Iometer 2010 Test | Incompressible Data | Compressible Data | ||
| 128KB Sequential Write (QD=3) | 262.MB/s | 518.2MB/s | ||
| 128KB Sequential Read (QD=3) | 493.4MB/s | 492.3MB/s | ||
| 4KB Random Read (QD=3) | 186.5MB/s | N/A | ||
| 4KB Random Write (QD=3) | 162.7MB/s | 227.0MB/s | ||
Even if we look at incompressible data (fully random), the performance is unbelievable. You get better minimum performance on the SF-2582 than peak performance on the SF-1200/SF-1500. Note that we couldn't run all of our tests given the very early nature of the hardware sample. The fact that we could get these numbers at all on the first beta of the drive was beyond impressive.
Obviously to hit these speeds you need a 6Gbps controller. Thankfully there are at least a few ways to get those ports.
It’s looking like SandForce will be last to bring out their next-generation drive in the first half of the year with both Micron and Intel beating it to the punch, but if we can get this sort of performance, and have it be reliable, it may be worth the wait.
Facebook
Twitter
RSS
62 Comments
View All Comments
evilspoons - Thursday, January 6, 2011 - link
The Vertex 2 / Agility 2 dissipate like ONE WATT while being tortured (http://www.anandtech.com/show/3681/oczs-vertex-2-s...I hardly think it's a problem.
ppokorny - Thursday, January 6, 2011 - link
I recently had a chance to use those new 3.5" drives, and they mix plastic and metal mounting holes in a way that makes the mechanical fit not be flush on the sides.That caused problems when I had to mount the drives in a hot-swap drive tray of a server.
marraco - Thursday, January 6, 2011 - link
Of course, the Vertex 3 should saturate SATA 2 on sequential speed, but I wish to know how much slower performs on 4 Kb tests when connected to SATA 2.Marc HFR - Thursday, January 6, 2011 - link
Dear Anand,4K Random read is more than 3 times faster than SF-1200 one.
Are you sure you 8GB LBA space restriction on this test ?
H8ff0000 - Thursday, January 6, 2011 - link
A lot of us want to know this. I know there's probably no info on this, but some sort of ballpark figures would help OCZ's business. If people heard a ballpark figure that didn't scare the piss out of them, they very well might actually wait for this to come to market instead of buying a C400. If I remember correctly, they said somewhere around $200 for a 120GB C400. If this were $250ish ballpark I'd wait.strikeback03 - Thursday, January 6, 2011 - link
Problem is that if they threw out a number now for a product they are hoping to launch in 4-5 months, even saying it is "ballpark" people will complain if they don't hit it. Say they said $250 right now, and you wait, and they launch in June at $300 for 120GB. Some users would be extremely upset by the wait and the "price increase".tjoynt - Thursday, January 6, 2011 - link
Why the heck did Intel decide to name their "some % random" IOMeter write strategy "pseudo random"? "pseudo random" already means something and it is not that. :(vol7ron - Thursday, January 6, 2011 - link
Page 1:"I get 518MB/s sequential write speed and nearly 500MB/s for sequential read"
- Are you sure that's accurate, or is it natural to have sequential write faster than sequential read?
Page 2:
'5.25” drive by'
- Probably meant drive bay
glugglug - Thursday, January 6, 2011 - link
On normally HDDs, random write is almost always faster than random read because the buffer cache used by the controller has a far more significant impact on writes.On an SSD maybe this applies to even sequential I/O as the wear leveling algorithm/ logical-to-physical block mapping may make your sequential operations not really 100% sequential.
FilipK959 - Thursday, January 6, 2011 - link
Well, my old Pentium 3 PC the PC 100 sdram had a transfer rate 420Mb per second according to SiSoft Sandra if I remember correctly so this is just amazing that a mass storage device can pull this off. :))