I would really like to see a performance comparison between all the SSD from SATA 3Gbps to latest PCI-E x4 with SF3700. Not just benchmarks but startup time etc.
I wonder on the difference if we are reaching to the point where CPU is back to the bottleneck. Or would we have to wait for MRAM for another leap in difference.
As I thought in the original SF-3700 article, SATA Express is DOA.
There's a reason why Intel isn't integrateing it into its 9 series chipsets and ASUS delayed the only motherboard scheduled to be released with SATA Express. The SF-3700 needs PCIe 2.0 x4 for full bandwidth, but SATA Express only provides half that. M.2 has its purpose for small form factor systems, but there's no significant advantage for SATA Express instead of PCIe other than backwards compatibility with SATA3.
Except that PCI-e drives aren't something that is easily accomodated in all systems.
SATA Express is not dead. Or at least I have heard no rumors or rumblings that it is. When did Intel ever say it would be integrated in 9 series chipsets for Broadwell? I've seen it listed on Intel roadmaps for Skylake, along with DDR4, for at LEAST the last 6 months.
I only would have been suprised if they had moved it up.
SATA Express can eventually accomodated PCI-e 3.0 (later maybe 4.0) as far as I know. Current is 2x PCI-e 2.0, which is 1,000MB/sec, minus very small overhead. That is roughly double what SATA3 is capable of. The other optional SATA Express, which I think is more enterprise targeted, is 4x PCI-e 2.0, or 2,000MB/sec.
Extending it to PCI-e 3.0, which I assume wouldn't take long doubles those speeds again.
There gets to be a point where form factor and system integration are massively more important than shaving a few hundred miliseconds off your boot time or an application launch time.
By the end of the year or early 2015 I am sure we'll see SATA Express boards with Skylake offering native support.
PCI Express 2.0 (January 2007) has up to to 500 MB/s with 20% overhead for the protocol per line so that's PCI Express 2.0 x4 lines = 15.62Gb/ 1.95 GB/s -20% overhead.
PCI Express 3.0 (November 2010) has up to to 985 MB/s with 1.54% overhead for the protocol per line so that's PCI Express 3.0 x4 lines = 488.28Gb/ ,61.03 GB/s -1.54% overhead.
so any of the large foundries could shink Mram to something like 40nm and improve its throughput past the fastest DDR3 and potential DDR4 on current 32bit/64bit memory controllers.
not to forget that "Wide I/O" four 128-bit DDR3 memory channels, providing a 512-bit wide interface to memory can currently a total bandwidth of 17GBps at 266Mhz, and Wide I/O 2 has the same four-channel architecture, but the channels have a higher bandwidth of 6.4GBps and have 64 I/Os per channel, yielding an overall 25.6GBps data rate on 421 package balls to CPU.
Later Wide I/O 2 devices will have up to 128 I/Os per channel and a data rate of 51.2GBps. Wide I/O 2 also opens the possibility of a 1,024-bit interface to memory. then there's its sibling Hybrid Memory Cube that can do 320GBp to start, with 70% less energy per bit than DDR3 and 90% less board space than today's RDIMMs? so MRAM if EverSpin finally joined the 120+ member consortium today would be as close as we have got to universal non volatile memory to date.
the simple fact is the PCI-e but needs a massive update if we are to get anything close to what even just a few wide IO configured non volatile Mram packages could provide next year or to...
You are reporting on old information. SATA express disappeared from the 9 series chipset roadmap in the late summer in both the LGA2011 r3 and LGA1150 chipsets. I doubt it means that SATA express will never be implemented, but you shouldn't count on it in 2014. Because of this, PCIe card SSDs will be the form factor of choice.
Regarding your other argument. The SATA (and IDE and SCSI) cabled interface is only necessary when you have mechanical hard drives and optical disks. They are too big to be mounted on PCIe cards, so it makes the most sense to mount them somewhere else in the case and run data cables to them. In contrast, you can put all the chips of a SSD onto a very tiny PCB. They are so small that they fit within the motherboard's space constraints and no longer need to be mounted externally. Look inside modern 2.5" SSDs; their cases are vastly oversized for the PCBs. The only reason 2.5" SSDs exist is to maintain backwards compatibility with the existing SATA form factor that was developed for laptop hard disks. It's time to move on to a form factor more space optimized for small SSDs.
In addition, SSDs are high bandwidth devices, and interface speeds are limited by the physics of cables, specifically that interference in long cables limits bandwidth. The closer the SSD is to the host machine, the faster its data bus can be run. Why does it make any sense to have to mount a tiny SSD in a remote location and run a cable to it when you could make an identical SSD on a small PCIe card and connect it directly to a faster slot with a wider bus?
SATA Express will come at some point, but I only see it being used for hard drives. Current hard drive controllers can only transfer ~200MB/s at peak, with some higher bandwidth bursts, so it will be a while before the 1GB/s bandwidth of SATA Express is needed. However, SSDs can already max out the interface; it's saturated before it even came to market. PCIe is the fastest expandable bus in a PC and doesn't suffer the cable bandwidth constraints of SATA. Why not use that for fast SSDs?
To expand upon that, it is still easier to make big capacity SSDs in larger formats and easier to make them fast. So 2.5" form factor is still resonably important for a lot of laptops.
Enter SATA Express.
And in servers, and in...the list goes on.
Its pretty hard to hot plug a PCI-e card from a server, compared to hot plugging a 2.5" SSD.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
8 Comments
Back to Article
LordConrad - Wednesday, January 8, 2014 - link
Can't wait! I have two Chronos Deluxe drives and they work great.iwod - Wednesday, January 8, 2014 - link
I would really like to see a performance comparison between all the SSD from SATA 3Gbps to latest PCI-E x4 with SF3700. Not just benchmarks but startup time etc.I wonder on the difference if we are reaching to the point where CPU is back to the bottleneck.
Or would we have to wait for MRAM for another leap in difference.
The Von Matrices - Thursday, January 9, 2014 - link
As I thought in the original SF-3700 article, SATA Express is DOA.There's a reason why Intel isn't integrateing it into its 9 series chipsets and ASUS delayed the only motherboard scheduled to be released with SATA Express. The SF-3700 needs PCIe 2.0 x4 for full bandwidth, but SATA Express only provides half that. M.2 has its purpose for small form factor systems, but there's no significant advantage for SATA Express instead of PCIe other than backwards compatibility with SATA3.
Movieman420 - Thursday, January 9, 2014 - link
"The PCIe-based drives should be faster, particularly in sequential transfer rates..."The 3700 is also said to not have the speed drop with incompressible data like the 2281s do.
azazel1024 - Thursday, January 9, 2014 - link
Except that PCI-e drives aren't something that is easily accomodated in all systems.SATA Express is not dead. Or at least I have heard no rumors or rumblings that it is. When did Intel ever say it would be integrated in 9 series chipsets for Broadwell? I've seen it listed on Intel roadmaps for Skylake, along with DDR4, for at LEAST the last 6 months.
I only would have been suprised if they had moved it up.
SATA Express can eventually accomodated PCI-e 3.0 (later maybe 4.0) as far as I know. Current is 2x PCI-e 2.0, which is 1,000MB/sec, minus very small overhead. That is roughly double what SATA3 is capable of. The other optional SATA Express, which I think is more enterprise targeted, is 4x PCI-e 2.0, or 2,000MB/sec.
Extending it to PCI-e 3.0, which I assume wouldn't take long doubles those speeds again.
There gets to be a point where form factor and system integration are massively more important than shaving a few hundred miliseconds off your boot time or an application launch time.
By the end of the year or early 2015 I am sure we'll see SATA Express boards with Skylake offering native support.
BMNify - Thursday, January 9, 2014 - link
PCI Express 2.0 (January 2007)has up to to 500 MB/s with 20% overhead for the protocol per line
so that's PCI Express 2.0 x4 lines = 15.62Gb/ 1.95 GB/s -20% overhead.
PCI Express 3.0 (November 2010)
has up to to 985 MB/s with 1.54% overhead for the protocol per line
so that's PCI Express 3.0 x4 lines = 488.28Gb/ ,61.03 GB/s -1.54% overhead.
DDR3-800 PC3-6400 6400 MB/s 800
DDR3-1066 PC3-8500 8533 MB/s 1066
DDR3-1333 PC3-10600 10667 MB/s 1333
DDR3-1600 PC3-12800 12800 MB/s 1600
and EverSpin/Freescale http://www.dailytech.com/Freescale+Semiconductor+A...
current 64 Mb (8 MB) ST-MRAM chip that performs at DDR3 speeds on an old 90nm process.
so any of the large foundries could shink Mram to something like 40nm and improve its throughput past the fastest DDR3 and potential DDR4 on current 32bit/64bit memory controllers.
not to forget that "Wide I/O" four 128-bit DDR3 memory channels, providing a 512-bit wide interface to memory can currently a total bandwidth of 17GBps at 266Mhz, and Wide I/O 2 has the same four-channel architecture, but the channels have a higher bandwidth of 6.4GBps and have 64 I/Os per channel, yielding an overall 25.6GBps data rate on 421 package balls to CPU.
Later Wide I/O 2 devices will have up to 128 I/Os per channel and a data rate of 51.2GBps. Wide I/O 2 also opens the possibility of a 1,024-bit interface to memory.
then there's its sibling Hybrid Memory Cube that can do 320GBp to start, with 70% less energy per bit than DDR3 and 90% less board space than today's RDIMMs? so MRAM if EverSpin finally joined the 120+ member consortium today would be as close as we have got to universal non volatile memory to date.
the simple fact is the PCI-e but needs a massive update if we are to get anything close to what even just a few wide IO configured non volatile Mram packages could provide next year or to...
The Von Matrices - Friday, January 10, 2014 - link
You are reporting on old information. SATA express disappeared from the 9 series chipset roadmap in the late summer in both the LGA2011 r3 and LGA1150 chipsets. I doubt it means that SATA express will never be implemented, but you shouldn't count on it in 2014. Because of this, PCIe card SSDs will be the form factor of choice.http://vr-zone.com/articles/sata-express-intel-9-s...
http://vr-zone.com/articles/intel-core-i7-ivy-brid...
Regarding your other argument. The SATA (and IDE and SCSI) cabled interface is only necessary when you have mechanical hard drives and optical disks. They are too big to be mounted on PCIe cards, so it makes the most sense to mount them somewhere else in the case and run data cables to them. In contrast, you can put all the chips of a SSD onto a very tiny PCB. They are so small that they fit within the motherboard's space constraints and no longer need to be mounted externally. Look inside modern 2.5" SSDs; their cases are vastly oversized for the PCBs. The only reason 2.5" SSDs exist is to maintain backwards compatibility with the existing SATA form factor that was developed for laptop hard disks. It's time to move on to a form factor more space optimized for small SSDs.
In addition, SSDs are high bandwidth devices, and interface speeds are limited by the physics of cables, specifically that interference in long cables limits bandwidth. The closer the SSD is to the host machine, the faster its data bus can be run. Why does it make any sense to have to mount a tiny SSD in a remote location and run a cable to it when you could make an identical SSD on a small PCIe card and connect it directly to a faster slot with a wider bus?
SATA Express will come at some point, but I only see it being used for hard drives. Current hard drive controllers can only transfer ~200MB/s at peak, with some higher bandwidth bursts, so it will be a while before the 1GB/s bandwidth of SATA Express is needed. However, SSDs can already max out the interface; it's saturated before it even came to market. PCIe is the fastest expandable bus in a PC and doesn't suffer the cable bandwidth constraints of SATA. Why not use that for fast SSDs?
azazel1024 - Thursday, January 9, 2014 - link
To expand upon that, it is still easier to make big capacity SSDs in larger formats and easier to make them fast. So 2.5" form factor is still resonably important for a lot of laptops.Enter SATA Express.
And in servers, and in...the list goes on.
Its pretty hard to hot plug a PCI-e card from a server, compared to hot plugging a 2.5" SSD.