SATA-IO Introduces the SATA v3.2 Specifications

[N.Broekhuijsen]

A new revision of the SATA specification has been announced by the SATA-IO, and it comes with some very impressive new specifications.

SATA-IO Introduces the SATA v3.2 Specifications : Read more

 

[MKBL]

- Isn't it more appropriate to put them on the same unit when they are compared? 16 Gb/s vs 6 Gb/s.

"SATA Express will also allow devices to run at speeds of up to 2.0 GB/s, as compared to just 6 Gb/s for the SATA3 standard."

- Any comment on SATA port location on motherboard? The current geometry can be inconvenient when a huge graphics card is installed.

 

[pjmelect]

"SATA Express will also allow devices to run at speeds of up to 2.0 GB/s, as compared to just 6 Gb/s for the SATA3 standard."

2.0GB/S = 20Gb/S (aprox) so it is three and a bit times faster.

 

[back_by_demand]

Agreed with using same terminology for transfer speeds, admittedly if you have an interest in this stuff and don't know you have 8 bits in a byte you should just go away and get another hobby, but the principle is sound. But disagree with the moan about SATA port locations as the assumption based on the comment is that all board vendors have ports in the same location when they clearly don't. But otherwise this is great news, all it takes is for an SSD maker to construct a 1Tb 3.5" unit which is 4 x 240Gb 2.5" units in RAID and suddenly the transfer speed goes through the roof and we need faster SATA speeds. Maybe the recent 3D storage breakthroughs are the kind of speed and storage tech that will require this enhancement. Keep up the good work, a trillion bits per minute is pretty sweet.

 

[Kewlx25]

Current SATA is rated in microsecond latency, PCIe is rated in nanosecond latency. I can't wait to see how this makes a different.

 

[universal remonster]

"2.0GB/S = 20Gb/S (aprox).

No... it doesn't. By using your math we could also say that 80% = 100% (aprox).

8bits in a Byte would be 2GB/s = 16gb/s, making it less than three times faster.

 

[milktea]

"2.0GB/S = 20Gb/S (aprox).

No... it doesn't. By using your math we could also say that 80% = 100% (aprox).
8bits in a Byte would be 2GB/s = 16gb/s, making it less than three times faster.

I believe SATA spec Gb/s is stated in 2^30 bits/s instead of 1x10^9 bits/s. For SATA3.0, 6Gb/s actually does give you 600MB/s taking into account for 8b/10b and some extra headroom. So I'm not surprise that SATAexpress 2GB/s is indeed 20Gb/s.

 

[MetzMan007]

Why data 3.2, I think data 3 has been out loud g enough to call it data 4. It's already confusing enough when some says data 3 because data 2 was also know as data 3 because it was 3Gb/sec

 

[ojas]

"SATA Express will also allow devices to run at speeds of up to 2.0 GB/s, as compared to just 6 Gb/s for the SATA3 standard."

2.0GB/S = 20Gb/S (aprox) so it is three and a bit times faster.

This might actually be correct if 2GB/s is the actual realisable throughput.

There's about 25% overhead on SATA (which is why SATA III gives you 600 MB/s instead of 750 MB/s that you should get by direct conversion of 6 Gb/s), so if you're getting 2048 MB/s, then the actual interface provides 2048 x 1.25 = 2560 MB/s = 20,480 Mb/s = 20.48 Gb/s.

HOWEVER, since PCIe 3.0 only enables up to 8 Gb/s (GT/s, to be more accurate) and PCIe 4.0 enables 16 GT/s, I think SATA 3.2 enables a 16.38 Gb/s interface (2 GB/s x 8 bits per byte) and not 20 Gb/s.

 

[ojas]

Why data 3.2, I think data 3 has been out loud g enough to call it data 4. It's already confusing enough when some says data 3 because data 2 was also know as data 3 because it was 3Gb/sec

Pretty sure it's called SATA.

 

[aluop]

"2.0GB/S = 20Gb/S (aprox).

No... it doesn't. By using your math we could also say that 80% = 100% (aprox).
8bits in a Byte would be 2GB/s = 16gb/s, making it less than three times faster.

I believe SATA spec Gb/s is stated in 2^30 bits/s instead of 1x10^9 bits/s. For SATA3.0, 6Gb/s actually does give you 600MB/s taking into account for 8b/10b and some extra headroom. So I'm not surprise that SATAexpress 2GB/s is indeed 20Gb/s.

First of all, binary units are almost never used when talking about communication speed,
The M and G here are metric not binary.
Also,

SATA 3.0 channel speed: 6Gb/s
SATA 3.0 actual data speed: 0.6GB/s not 0.75GB/s because of 8b/10b encoding

PCIe 3.0 channel speed: 8Gb/s per link
PCIe 3.0 actual data speed: 1GB/s per link because it's using the more efficient 128b/130b encoding

SATA Express = 2 lanes PCIe 3.0 = 16Gb/s = 2GB/s

 

[MKBL]

My point is that comparison should be made on the same scale. Either XX GB compared to YY GB, or xx Gb to yy Gb. It's rather about writing fundamental than on technical accuracy. Comparing on two different scales in the same sentence is awkward and can be misleading.

 

[InvalidError]

Serial link rates ALWAYS use the proper (powers of 1000) SI definition of SI prefixes. I am not aware of any exceptions to that. Ever.

The reason why the usable bandwidth is slightly lower than signaling rate is due to 64b66b encoding which expands each 64bits code word to a 66bits TMDS pattern. There are two reasons for using this encoding: prevent coupling capacitor saturation on differential links my guaranteeing an even distribution of 0s and 1s between bit groups and guaranteeing a minimum number of transitions to help receivers maintain reliable PLL lock on the bitstream.

Lower speed links often used 8b10b coding which yields a larger discrepancy between usable and line rates.

Some link-layer protocol rates are reported based on usable rate while others are reported based on signaling rate. IIRC, 100BaseTX and 1GBaseT/TX are rated based on usable link-layer bits - meaning the line signaling rate is higher - 125Mbps for 100BaseTX which uses 8b10b coding and 1031.25Mbps for 1GBase-TX (yes I know, 1GBase-T won) which uses 64b66b coding as do PCIe and most other high-speed serial links. The good old RS232 rate was based on signaling rate - every extra field you enable on that 115200 is one less bit available for actual data.

If they are going to start emulating SATA on PCIe, they should focus on native PCIe block storage and ditch SATA altogether. Similarly for USB3: if they are going to bring USB up to PCIe specs, might as well quit pussy-footing around and simply allow endpoints to negotiate native PCIe x1 on USB3 ports.

 

[back_by_demand]

Wow, some serious fucking stupidity going on here, trying to argue black is white converting GB to Gb - 8 bits in a byte, end of story

 

[gorfmaster1]

Regardless of how you convert it, It is still fast. and looking forward to see what SSD's can do with that extra bandwidth

 

[pjmelect]

Although there are 8 bits to a byte, there is overhead, so if you assume 10b to a byte you will be about right.

https://en.wikipedia.org/wiki/Serial_ATA