Discussion in 'Article Discussion' started by Gareth Halfacree, 13 Aug 2013.
SATA Express to boost peak SSD speeds.
Couldn't they have just stepped it down to about 14Gb/s and not use 2 lanes? Also, does that mean it requires 2 lanes per SATA port, or 2 lanes for an entire set of ports (presumably 4 or 6 per set)? If it's the latter, I think it's dumb they use 2 lanes because that means you can't buy a 1x slot card, and it's a real waste of a 16x slot.
There are no ports: SATA Express connects via a PCI Express slot, not a SATA port. This isn't an HBA they're talking about: a SATA Express drive would be a bunch of NAND flash strapped to a PCI Express card. Basically, exactly what we've got now in PCIe SSDs but 'talking' SATA so the OS doesn't have to give a monkey's how the drive it's communicating with is connected.
Ah ok. I understood the idea of the PCIe card but I guess I was picturing more that there would also eventually be standard SATA ports that operate at 16Gb/s.
One thing to point out on the article, there are plenty of SSDs that saturate the bus. SATA 3.0 is theoretically capable of 6Gbps, or 750MB/sec. However, SATA overhead imposes a limit closer to 550-575MB/sec. There are a number of SSDs that are capable of this today.
It is "easy enough" with a moderately large SSD to easily exceed the limits on SATA3.0, all you have to do is expand how many channels the controller uses. There are no hard an fast limits on this, other than most SSDs use 8 channels because they are already saturating the SATA3.0 link...so why use more?
Moving to SATA3.2 and SATA Express with a 16Gbps limit means that manufacturers can easily move to 12 or 16 channel controller designs for their SSDs and bump the speeds by a fair margin. I doubt we'll have to wait more than 2-3 years before SATA Express is maxed out by the fastest SSDs.
The other thing this helps with is random performance and access times. SATA has a good sized overhead added on top of what the controller and NAND itself is actually capable of. I'd be suprised if this didn't increase performance by a good percentage in random access times and things like random 4k I/O...which are what needs improvement more than high sequential transfer speeds.
Keep in mind, most PCI-e SSDs currently work using a PCI-e bridge chip. It is SSD controller that either "speaks" SATA on the output side, or else it works with an SATA controller and then is connected to a PCI-e bridge chip and then connected to the machines PCI-e controller. So it is STILL working through SATA. SATA3.2 however removes that. You are connecting directly to the PCI-e controller without a bridge chip and the controller on the SSD is speaking PCI-e effectively (or at least it is extending SATA over PCI-e from what I understand of it).
Since it is all working native PCI-e...even though it is speaking SATA over it, it removes a HUGE amount of the delays and overhead of the SATA bus.
This is a lot more than simply increasing max theoretical bandwidth to 16Gbps. It is also about reducing I/O bottlenecks, random performance, new form factors, reducing power consumption and heat.
That's actually a rather impressive idea. SataE probably won't take off that quickly. But the PCI-e solution is now much more elegant than beofre.
The problem i see, is that by the term the industry moves forward with SATA-Express 16Gbps, the speed of NAND with 8 Channel would have surprised the Interface Speed again. Since PCI-E 4.0 isn't even ready yet why cant we move straight to 4x PCI-E interface?
Likely the next spec will. Since this is basically SATA extended over the PCI-e interface, it is maxed out at 4xPCI-e 2.0. I am a little befuddled as to why it doesn't work over PCI-e 3.0 and uses the slower standard, but it is what it is.
I'd imagine it would be a (relatively) minor process to revise/update the spec at a future date to have it take advantage of PCI-e 3.0 and a future 4.0 for large speed boosts.
It might just be that PCIe 3.0 is much harder/more expensive to implement than PCIe 2.0.
PCIe Gen 2 was basically just a speed bump up from PCIe 1, and not much else changed. PCIe 3.0 has a completely different physical layer and the speed bump up to 8GT/sec pushes it over the edge of what you can do "easily" in digital silicon. PCIe 2.0 at 5GT/sec is 2.5GHz, which we're used to seeing in chips the whole time. To get your PHY running at the 4GHz you need for 8GT/sec gen 3 takes much more effort, and probably adds costs that you don't want in an SSD controller that costs a few $.
I would love to see sata ports, both on the mb as well as esata, deliver power and not just data so i don't need to plug in extra power lines to SSD's, HDD's and ODD's.
Separate names with a comma.