That connector in the middle of the board and the mounting holes makes it look like they could stack a daughterboard with additional controllers and flash onto this unit.
no, i agree with zdw, it indeed appears to be a daughter-board interface. why would they have such a wide mounting area for a firmware update board and who would you need a firmware update board in the first place? it's PCIe, plug it in and the firmware update tool will find the device id, much like it would search the SATA channels. it is doubtful its a diagnostics port. however, if it isn't an expansion interface, it might not exist on the final product, either.
I think he is correct. It is a battery port. If your remember anything about the sandforce drives one of their main features is that they cache more data to prevent excessive writing to the drive and get around some of the problems that were had in the past.
Here is a quote taken from a blurb about a missing supercap in a consumer sandforce OCZ drive: "...Its only enough for a short burst, but it's enough to get the job done. This is important on SSDs more so than HDDs in enterprise. On an HDD when write caching is turned off, the last write command is lost, but the rest of your data is secure. On standard SSDs because of the nature of Flash and Flash management it is not typically possible to guarantee data integrity on sudden power loss..."
whats more likely its for is so an second bank of flash as there is 4 screw holes there to hold an second bank of flash, making it into an 4x raid card for flash the slot most likely is connected to the intel raid chip
Where are these speed estimates from? SF1200 could very well be limited by SATA 3gbps, so two of them in raid 0 might give you better than 2x the performance of a single controller in current SSDs.
They could implement trim as a OS driver if they wanted to. No idea if they do.
DesktopMan, you appear to be confusing SATA with PCIe. PCIe 4x ought to yield, correct me if I'm wrong, 2GB/s bandwidth. [http://en.wikipedia.org/wiki/PCIe]
One thing in the article that bugs me is "Obviously you lose TRIM support". I assume RAID 0 is responsible for this, but why? Are all SSD RAID drives incapable of TRIM? Is there a work-around?
TRIM is still NOT possible with SSD's in RAID. Garbage Collection would be the alternative but I do not believe garbage collection works with the Sandforce controllers.
No garbage collection? I don't know where you people get these ideas. ALL SSDs have garbage collection, otherwise it would NOT BE POSSIBLE to have an SSD without the TRIM command.
No I think he might be right. SATA-II only has about 300 Mbyte/s compared to 2 GB/s for pcie x4 if wiki is correct on this. Therefore if SF-1200 is limited by SATA-II, this drive could be more than twice as fast. Let's assume 1 SF-1200 could read at 400 Mbyte/s. On SATA-II this will be something close to 300 minus overhead. (which it is)
On pice x4 it's not bandwith limited so you could get 800 Mbyte/s. hypothetical.
Are there any downsides to pice drives? (beside losing trim in this one)
The downside would be that your using your PCIe Bus :)
If you already have other things like maybe two graphics cards in Crossfire or even just one graphics card, you don't want to max out your bus.
Ultimately I think it would be fine but just keep in mind that for example your PCI bus only has 133mb/s transfer rate no matter how many PCI SATA controllers you install into it. They all share that bus.
There is a 32pin EEPROM in the upper left of the card, so it is most likely bootable
Also right above both of the SandForce controller sits a quite large rectangular white thing maked "X1 and X2" on the PCB, I'll take a guess those are x-caps
And lastly there are many unpopulated headers on the board, for example J3 and J5 between the SandForce controllers, can most likely be used for diagnostics and firmware flashing (disaster recovery) There are also many contacts points named "TP" this is most likely a development board with extra stuff for diagnostics
The firmware is probably flashed to the card via the PCIe bridge of course, like you do with any other RAID controller
And the big connector in the middle with 4 holes surrounding it is most likely for capacity expansion via a daugtherboard
I think there is confusion between PCI and PCIe going on. Old PCI shared a single bus for all traffic, PCIe has dedicated lanes for each slot back to the southbridge (or such).
Also, "raw" PCIe SSDs cannot be booted, but this SSD and any other with RAID functionality can be a boot device -- it's just a RAID card as far as the system is concerned, it doesn't care what is beyond the Intel RAID controller.
There are a whole lot more pci express lanes on some controllers than most people need though. Even graphics cards only lose a small percentage of performance running at 8x 2.0 instead of 16x 2.0 so even if you used some of those lanes (iirc you have to go from 16 to 8) it would be fine.
Yes, I ment what I said about Sata II. If the current gen SSDs are bandwidth limited the actual performance (at least read performance) of SF1200 might be higher than we are currently seeing, this this card having a potantial to be faster than 2x of current SSDs.
As for trim, I'm not sure there even is such a thing as trim for pci express devices. Without looking into the driver model of these cards I see no reason for the driver not to allow trim messages to pass down though, by telling the OS that yes, this is a drive that supports trim. After all, this is not a software raid device that is detected as two drives in the OS so the problems with raid 0 with two SSDs don't really apply here. Just issues with trim + pci express in general.
The way the screw holes are mounted behind the center port there it looks like it is designed to hold something heavy and mounted towards that end. Expansion sounds right, but it may be a diagnostic port as well.
Who knows if this will stay when it comes to popular distribution or if this was the tinker-toy of the developers......
If the daughter card has 2 additional SSD's and a supercap to prevent data loss during power outage, then the ability to plug 4 such cards into my motherboard would make an excellent boot drive with read speeds around 4.4 GB/sec in Raid (16 SSD's X 275 MB/sec read per SSD average)
So far so good! If I am then booting up in, for example 0.63 seconds but want to shave a few more milliseconds from my boot time, I don't think I can push the buttons on my stopwatch fast enough to verify if my Boot Tweeks are working or not
Serious Gamers really need to be ready to play by the time they release the power button on their computer after all
Wink Wink
I just hope Intel has a 16 core CPU to handle the overhead by the time I can afford these and 2 Dual GPU Fermi Cards
The SF-1200 is a SATA-II to flash controller chip. You're still fundamentally limited by the SATA-II interface. Whatever RAID controller they put on the card, it still talks to the two SF-1200's via SATA-II.
PCI-e 1.1 x 4 gives you 1000 MBps in each direction, which is more than enough for two SATA-II interfaces, which can provide at most 600 MBps.
The Fusion-IO cards use a purpose-built PCI-e to flash controller, which eliminates the SATA speed bottleneck but also substantially increases costs. By pairing a COTS flash controller with a COTS RAID controller, OCZ is able to offer a PCI-e SSD at a much lower price point, but at the expense of performance.
This is a cool development, as it brings moderate-performance PCI-e SSD's at a much more accessible price point.
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
32 Comments
Back to Article
zdw - Monday, May 31, 2010 - link
That connector in the middle of the board and the mounting holes makes it look like they could stack a daughterboard with additional controllers and flash onto this unit.devene - Monday, May 31, 2010 - link
pci-e x4 is still mostly used in servers... And that connector in the middle of the board is probably used for flashing the firmware.Samus - Monday, May 31, 2010 - link
no, i agree with zdw, it indeed appears to be a daughter-board interface. why would they have such a wide mounting area for a firmware update board and who would you need a firmware update board in the first place? it's PCIe, plug it in and the firmware update tool will find the device id, much like it would search the SATA channels. it is doubtful its a diagnostics port. however, if it isn't an expansion interface, it might not exist on the final product, either.ratbert1 - Monday, May 31, 2010 - link
I am thinking connector for battery/supercapacitor memory failsafe.johnsonx - Monday, May 31, 2010 - link
why would it need that? it's a FLASH drive.PedroDaGr8 - Monday, May 31, 2010 - link
I think he is correct. It is a battery port. If your remember anything about the sandforce drives one of their main features is that they cache more data to prevent excessive writing to the drive and get around some of the problems that were had in the past.Here is a quote taken from a blurb about a missing supercap in a consumer sandforce OCZ drive:
"...Its only enough for a short burst, but it's enough to get the job done. This is important on SSDs more so than HDDs in enterprise. On an HDD when write caching is turned off, the last write command is lost, but the rest of your data is secure. On standard SSDs because of the nature of Flash and Flash management it is not typically possible to guarantee data integrity on sudden power loss..."
bwj - Monday, May 31, 2010 - link
It makes no sense to use a 60-pin connector for a supercap.mindless1 - Monday, May 31, 2010 - link
Exactly, it is definitely not "only" a battery or supercap connector. Definitely a data bus with that many pins.A battery or supercap connector would just be unused solder pads or holes, or a standard low cost SIP connector already soldered on.
bwj - Monday, May 31, 2010 - link
Why would you need a 60-pin interface for a battery? It is almost certainly an expansion port for daughtercard. See also the Fusion-io ioDrive Duo.pkoi - Monday, May 31, 2010 - link
pci-e 4x being for server ? any 16x will do,almost all board have multiple pci-e slot or Have a look at old EP45-DQ6
leexgx - Monday, May 31, 2010 - link
whats more likely its for is so an second bank of flash as there is 4 screw holes there to hold an second bank of flash, making it into an 4x raid card for flash the slot most likely is connected to the intel raid chipSlash3 - Tuesday, June 1, 2010 - link
Countdown to the RevoDrive Pro, now with double the capacity and performance!Mr Perfect - Monday, May 31, 2010 - link
What is it all the cool kids are saying nowadays? Oh, right. U MAD BRO?DesktopMan - Monday, May 31, 2010 - link
Where are these speed estimates from? SF1200 could very well be limited by SATA 3gbps, so two of them in raid 0 might give you better than 2x the performance of a single controller in current SSDs.They could implement trim as a OS driver if they wanted to. No idea if they do.
yyrkoon - Monday, May 31, 2010 - link
I think you're a bit confused.ekv - Tuesday, June 1, 2010 - link
DesktopMan, you appear to be confusing SATA with PCIe. PCIe 4x ought to yield, correct me if I'm wrong, 2GB/s bandwidth. [http://en.wikipedia.org/wiki/PCIe]One thing in the article that bugs me is "Obviously you lose TRIM support". I assume RAID 0 is responsible for this, but why? Are all SSD RAID drives incapable of TRIM? Is there a work-around?
DEVAST8 - Tuesday, June 1, 2010 - link
TRIM is still NOT possible with SSD's in RAID. Garbage Collection would be the alternative but I do not believe garbage collection works with the Sandforce controllers.bwj - Tuesday, June 1, 2010 - link
No garbage collection? I don't know where you people get these ideas. ALL SSDs have garbage collection, otherwise it would NOT BE POSSIBLE to have an SSD without the TRIM command.beginner99 - Tuesday, June 1, 2010 - link
No I think he might be right. SATA-II only has about 300 Mbyte/s compared to 2 GB/s for pcie x4 if wiki is correct on this. Therefore if SF-1200 is limited by SATA-II, this drive could be more than twice as fast.Let's assume 1 SF-1200 could read at 400 Mbyte/s. On SATA-II this will be something close to 300 minus overhead. (which it is)
On pice x4 it's not bandwith limited so you could get 800 Mbyte/s. hypothetical.
Are there any downsides to pice drives? (beside losing trim in this one)
xeopherith - Tuesday, June 1, 2010 - link
The downside would be that your using your PCIe Bus :)If you already have other things like maybe two graphics cards in Crossfire or even just one graphics card, you don't want to max out your bus.
Ultimately I think it would be fine but just keep in mind that for example your PCI bus only has 133mb/s transfer rate no matter how many PCI SATA controllers you install into it. They all share that bus.
softdrinkviking - Tuesday, June 1, 2010 - link
also, i think you can't use it to boot from,or at least PCIe SSDs couldn't do that the last time i checked.
Per Hansson - Tuesday, June 1, 2010 - link
There is a 32pin EEPROM in the upper left of the card, so it is most likely bootableAlso right above both of the SandForce controller sits a quite large rectangular white thing maked "X1 and X2" on the PCB, I'll take a guess those are x-caps
And lastly there are many unpopulated headers on the board, for example J3 and J5 between the SandForce controllers, can most likely be used for diagnostics and firmware flashing (disaster recovery)
There are also many contacts points named "TP" this is most likely a development board with extra stuff for diagnostics
The firmware is probably flashed to the card via the PCIe bridge of course, like you do with any other RAID controller
And the big connector in the middle with 4 holes surrounding it is most likely for capacity expansion via a daugtherboard
e_sandrs - Tuesday, June 1, 2010 - link
I think there is confusion between PCI and PCIe going on. Old PCI shared a single bus for all traffic, PCIe has dedicated lanes for each slot back to the southbridge (or such).Also, "raw" PCIe SSDs cannot be booted, but this SSD and any other with RAID functionality can be a boot device -- it's just a RAID card as far as the system is concerned, it doesn't care what is beyond the Intel RAID controller.
DesktopMan - Tuesday, June 1, 2010 - link
There are a whole lot more pci express lanes on some controllers than most people need though. Even graphics cards only lose a small percentage of performance running at 8x 2.0 instead of 16x 2.0 so even if you used some of those lanes (iirc you have to go from 16 to 8) it would be fine.Yes, I ment what I said about Sata II. If the current gen SSDs are bandwidth limited the actual performance (at least read performance) of SF1200 might be higher than we are currently seeing, this this card having a potantial to be faster than 2x of current SSDs.
As for trim, I'm not sure there even is such a thing as trim for pci express devices. Without looking into the driver model of these cards I see no reason for the driver not to allow trim messages to pass down though, by telling the OS that yes, this is a drive that supports trim. After all, this is not a software raid device that is detected as two drives in the OS so the problems with raid 0 with two SSDs don't really apply here. Just issues with trim + pci express in general.
bwj - Tuesday, June 1, 2010 - link
TRIM, obviously, is a feature of SATA. If this thing doesn't look like a SATA disk then TRIM is beside the point.Simen1 - Tuesday, June 1, 2010 - link
How can they possibly reduce the price of a drive with 80-90% by adding a Raid chip?Does Anand mean 10-20% _MORE_ then a single drive?
overzealot - Tuesday, June 1, 2010 - link
Either it's been updated, or that's exactly what he said.Ninjahedge - Tuesday, June 1, 2010 - link
The way the screw holes are mounted behind the center port there it looks like it is designed to hold something heavy and mounted towards that end. Expansion sounds right, but it may be a diagnostic port as well.Who knows if this will stay when it comes to popular distribution or if this was the tinker-toy of the developers......
Out of Box Experience - Tuesday, June 1, 2010 - link
If the daughter card has 2 additional SSD's and a supercap to prevent data loss during power outage, then the ability to plug 4 such cards into my motherboard would make an excellent boot drive with read speeds around 4.4 GB/sec in Raid (16 SSD's X 275 MB/sec read per SSD average)
So far so good! If I am then booting up in, for example 0.63 seconds but want to shave a few more milliseconds from my boot time, I don't think I can push the buttons on my stopwatch fast enough to verify if my Boot Tweeks are working or not
Serious Gamers really need to be ready to play by the time they release the power button on their computer after all
Wink Wink
I just hope Intel has a 16 core CPU to handle the overhead by the time I can afford these and 2 Dual GPU Fermi Cards
newrigel - Saturday, June 19, 2010 - link
When are these coming to the US?newrigel - Sunday, June 20, 2010 - link
Mac??somedude1234 - Monday, June 21, 2010 - link
I'm a bit late on the comment, but...The SF-1200 is a SATA-II to flash controller chip. You're still fundamentally limited by the SATA-II interface. Whatever RAID controller they put on the card, it still talks to the two SF-1200's via SATA-II.
PCI-e 1.1 x 4 gives you 1000 MBps in each direction, which is more than enough for two SATA-II interfaces, which can provide at most 600 MBps.
The Fusion-IO cards use a purpose-built PCI-e to flash controller, which eliminates the SATA speed bottleneck but also substantially increases costs. By pairing a COTS flash controller with a COTS RAID controller, OCZ is able to offer a PCI-e SSD at a much lower price point, but at the expense of performance.
This is a cool development, as it brings moderate-performance PCI-e SSD's at a much more accessible price point.