PCIe is still too bulky. The future is mSATA. I'd rather have something more versatile. Once you upgrade, you can relegate the used SSD to a portable drive with a cheap encloser or your laptop.
To me, it looks like the same SSD as in the Macbook Air (http://www.anandtech.com/show/4527/2011-macbook-ai... And, being OWC, that makes sense. I suspect it's just a pair of MBA SSDs, running in RAID, from a PCIe controller.
Good Eye. I went and played with some pictures of this new drive and the one for the MBA and sure enough the new one seems to be a little shorter.
Ratio of the drive in the MBA (after lopping off the connector so you have a perfect rectangle) is about 4.3:1 Similar Ratio of the drive on the Accelsior is about 3.8:1
After looking at the pics a little closer, it just looks like they refined their PCB design. The Toshiba NAND packages are closer together on the new stuff than on the previous one. Perhaps they were aiming to meet the PCIe Half Length Specification so they can be validated as a Half height, Half Length card.
The Accelsior uses a MARVELL 88SE9230 Raid Controller which provides a PCIe 2.0 x2 link to the raid controller that can control up to 4 6Gb/s SATA SSDs.
Er, what? PCIe slots/devices can use different numbers of lanes, ranging from 1 to 16. These are typically labeled, sensibly enough, as "x1", "x2" etc. Graphics cards are typically x8 or x16. Other devices may need merely a single x1 or somewhere in between. Each lane in PCIe 2.0 provides 500 MB/s. For PCIe 3.0, that increases to 1 GB/s. PCIe 4.0 is slated to double that again to 2 GB/s per lane, spec is supposed to be targeted for release sometime around 2015.
Hard to tell from that pic, but I'm pretty sure the back tab is too long to fit an x1 slot. But it's short enough that it doesn't fill an entire x4 connector, so x2 seems appropriate electrically, even if it's not a standard connector.
It is the Electrical interface of the Marvell SATA-controller. It's a two lanes physical card so it does require at least x4 PCI-e slot or more. Bigger hardware raid controllers with XOR, cache and all that usually is at least x4 PCI-e 2.0 or x8. But those also has more ports. I guess it would also work in an open ended x1 slot/port at half speed. Not sure if those are around anymore though. But x1 electrical x16 PCI-e slots still are. The Marvell controller can of course support and be configured for both.
since the device can only connect at pcie 2.0 x1, it will be limited to 500mb/s. the 780mb/s numbers quoted are not plausible given this configuration.
Your not reading or understanding anything in the comments section or at the wiki, look at the Marvell link or simply count the pins and you will see that it is two lanes and that PCIe can use any number of lanes and that the hardware can negotiate any number of lanes that it supports, you can run a graphic card at x2 too for that matter. Slots and electrical are two different things here. The controller are designed to work with up to two lanes.
"A connection between any two PCIe devices is known as a link, and is built up from a collection of one or more lanes. All devices must minimally support single-lane (×1) link. Devices may optionally support wider links composed of 2, 4, 8, 12, 16, or 32 lanes. This allows for very good compatibility in two ways:"
There is no problem fitting a smaller card in a bigger slot, it's designed that way.
They would basically need a 8-port SAS-controller for x4 PCI-e support. At least as far as Marvell is concerned.
1. No PCIe x2 is not allowed. Look at recent news, Intel wanted a 2x PCIe as standard but the consortium on charge of PCIe doesn't agree with the idea. If you want tu use 2 PCIe lane with this Marvell controller you have to use a 4x PCIe controller at least. You can't aggregate 2 x1 lanes. It works only if the client chip has a internal or external PCIe switch (like those from PLX)
2. Look at the picture! It's a x1 connector on the card. How do you want the Marvell controller to use 2 PCIe lanes??? This SSD card won't exceed ~480-500MB/s.
While PCIe does not provide for a 2x physical slot, it unambiguously supports a 2x electrical interface. Plug a x2 card into a x4 slot and you'll get the much higher performance numbers you seem to want to ignore...
Slot != Card that was what my post also said, we all know that it's x1, x4, x8, x16 and x32 as far as slots go. Or mechanical interface. When it's internal on the motherboard it doesn't matter and saves lanes using only as many as you need. Using a >x4 slots leaves two or more lanes unused. There are some motherboards with electrical x2 slots though. No Mac of course. Here we simply have a card with a SATA-controller designed to be used with two lanes.
A problem with x2 is that PCIe chips aren't required to negotiate 2 lanes. For example, the ESB2 south bridge (PCIe 1.0) in the 2006, 2007, and 2008 Mac Pro's and other old PC's can only negotiate 1, 4, or 8 lanes. The Accelsior in such a slot will only use a link width of 1 lane for 195 MB/s read/writes. Even with a 2 lane link width (as provided by the 5000 series north bridge chip in the Mac Pro 2006, 2007), the Accelsior will only do 380 MB/s in a PCIe 1.0 slot.
I wonder if any PCIe 2.0 chips have the same link width limitation. The 5400 north bridge chip in the Mac Pro 2008 handles x2 fine.
I have been using the OCZ Revodrive 3 x2 240GB since November in a self-build with a i7 2700K CPU that clocks up to 3.9GHz on load and 8GB of 1833MHz RAM.
In short I got a best write of 1129.555 MB/s and a best read of 1520.693 MB/s. The best write speed is a little short of the advertised 1225 MB/s but it is in use as the boot device on machine in real use, not sitting idle waiting for test data in a lab. The best read speed is actually slightly better than the advertised 1500MB/s, so “figures meant for advertising” or not it is a lot better in real use than the OWC claim.
I expect to see storage from OCZ in the not too far distant future that saturates 4 lanes of PCIe v2 (2GB/s).
The best read speed is actually slightly better than the advertised 1500MB/s, so “figures meant for advertising” or not it is a lot better in real use than the OWC claim.
Yep. The OCZ is also about 25-30% more expensive, but it is significantly faster then the price margin. That means, by pure advertised numbers at any rate (which doesn't seem entirely unreasonable in this case since they're both SF2000 based), that it's about 50-60% more bang-for-buck. Absolute pricing will probably matter to some people, who may prefer to pocket the extra $130 rather then pay more for better performance. But it certainly looks like the primary value is the AHCI compliance.
Hopefully that becomes the standard practice for everyone going forward.
Considering there's still a significant problem with SF-controllers and dealing with uncompressible data I'd rather hold out for the Revodrive 4 which will likely feature similar controllers to the Vertex4.
In my mind I'd find it hard to believe anyone looking for a boot/software device would go PCIe at this point. Seems to me these drives would be aiming towards active workspace considering the added cost.
I doubt you could measure much of a performance increase from booting/loading software off of one of these compared to a Vertex3/4 of the same size.
They should put heatsinks on the chips. Not because it needs them, but at that price I want something neat looking other than a PCB with some boring black chips on it.
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
33 Comments
Back to Article
nathanddrews - Wednesday, April 18, 2012 - link
Native PCIe SSDs seem like the wave of the future. Eliminate the SATA controller bottleneck...http://www.anandtech.com/show/5306/ocz-marvell-cre...
havoti97 - Monday, April 23, 2012 - link
PCIe is still too bulky. The future is mSATA. I'd rather have something more versatile. Once you upgrade, you can relegate the used SSD to a portable drive with a cheap encloser or your laptop.ViperV990 - Wednesday, April 18, 2012 - link
The blades don't appear to be mSATA. Are they proprietary?eezip - Wednesday, April 18, 2012 - link
To me, it looks like the same SSD as in the Macbook Air (http://www.anandtech.com/show/4527/2011-macbook-ai... And, being OWC, that makes sense. I suspect it's just a pair of MBA SSDs, running in RAID, from a PCIe controller.Casper42 - Wednesday, April 18, 2012 - link
And whats under the Heatsink on the left side?Its not like AT to forget that kind of information.
is it a PCIe Bridge?, Is it a RAID controller? etc?
Penti - Wednesday, April 18, 2012 - link
It's a Marvell SATA-controller. As the SSD's are 6Gbps SATA SSDs it's simply a 6Gbps SATA controller on PCIe.Kevin G - Tuesday, April 24, 2012 - link
Good eye. Here is the product page for OWC's MacBook Air SSD upgrades:http://eshop.macsales.com/shop/SSD/OWC/Aura_Pro_Ex...
They likely are the same basic part, just rebranded for usage on the PCI-e adapter.
joevt - Sunday, May 27, 2012 - link
OWC says they are a different form factor than the MacBook Air SSDs.thessdreview.com says the Accelsior uses mPCIe boards.
PCTC2 - Wednesday, April 18, 2012 - link
They look like the MacBook Air sticks, but shorter.Casper42 - Wednesday, April 18, 2012 - link
Good Eye. I went and played with some pictures of this new drive and the one for the MBA and sure enough the new one seems to be a little shorter.Ratio of the drive in the MBA (after lopping off the connector so you have a perfect rectangle) is about 4.3:1
Similar Ratio of the drive on the Accelsior is about 3.8:1
After looking at the pics a little closer, it just looks like they refined their PCB design. The Toshiba NAND packages are closer together on the new stuff than on the previous one. Perhaps they were aiming to meet the PCIe Half Length Specification so they can be validated as a Half height, Half Length card.
somedude1234 - Wednesday, April 18, 2012 - link
The SandForce SF-228x are NAND->SATA controllers, which storage controller is getting you from SATA->PCIe?Some companies are using LSI, others Marvell. Does anyone know who OWC is using here?
joevt - Sunday, May 27, 2012 - link
The Accelsior uses a MARVELL 88SE9230 Raid Controller which provides a PCIe 2.0 x2 link to the raid controller that can control up to 4 6Gb/s SATA SSDs.Casper42 - Wednesday, April 18, 2012 - link
x2?Is that some newfangled interface?
zanon - Wednesday, April 18, 2012 - link
Er, what? PCIe slots/devices can use different numbers of lanes, ranging from 1 to 16. These are typically labeled, sensibly enough, as "x1", "x2" etc. Graphics cards are typically x8 or x16. Other devices may need merely a single x1 or somewhere in between. Each lane in PCIe 2.0 provides 500 MB/s. For PCIe 3.0, that increases to 1 GB/s. PCIe 4.0 is slated to double that again to 2 GB/s per lane, spec is supposed to be targeted for release sometime around 2015.Casper42 - Wednesday, April 18, 2012 - link
Except there is no such thing as x2 on the physical interface.Its x1, x4, x8 and x16
I guess they should have said something like x4 (x2 electrical)
But then again, it sure looks like an x1 from the picture.
And if it is an x1, how do you get 780MB/s through an x1 that supports only 500MB/s?
LordOfTheBoired - Wednesday, April 18, 2012 - link
Hard to tell from that pic, but I'm pretty sure the back tab is too long to fit an x1 slot. But it's short enough that it doesn't fill an entire x4 connector, so x2 seems appropriate electrically, even if it's not a standard connector.Penti - Wednesday, April 18, 2012 - link
It is the Electrical interface of the Marvell SATA-controller. It's a two lanes physical card so it does require at least x4 PCI-e slot or more. Bigger hardware raid controllers with XOR, cache and all that usually is at least x4 PCI-e 2.0 or x8. But those also has more ports. I guess it would also work in an open ended x1 slot/port at half speed. Not sure if those are around anymore though. But x1 electrical x16 PCI-e slots still are. The Marvell controller can of course support and be configured for both.SleepyFE - Thursday, April 19, 2012 - link
PCIe infact goes up to x32. graphics cards have x16 physical connectors but usualy work in x8 mode if crossFire or SLi is used.Penti - Wednesday, April 18, 2012 - link
It's the controllers interface, http://www.marvell.com/storage/system-solutions/sa...UltraTech79 - Wednesday, April 18, 2012 - link
If by 'newfangled' you mean nearly a decade old... then sure.quanstro - Thursday, April 19, 2012 - link
wiki disagrees: http://en.wikipedia.org/wiki/PCI_Express#Architect...2x is not defined.
since the device can only connect at pcie 2.0 x1, it will be limited to 500mb/s.
the 780mb/s numbers quoted are not plausible given this configuration.
Penti - Thursday, April 19, 2012 - link
Your not reading or understanding anything in the comments section or at the wiki, look at the Marvell link or simply count the pins and you will see that it is two lanes and that PCIe can use any number of lanes and that the hardware can negotiate any number of lanes that it supports, you can run a graphic card at x2 too for that matter. Slots and electrical are two different things here. The controller are designed to work with up to two lanes."A connection between any two PCIe devices is known as a link, and is built up from a collection of one or more lanes. All devices must minimally support single-lane (×1) link. Devices may optionally support wider links composed of 2, 4, 8, 12, 16, or 32 lanes. This allows for very good compatibility in two ways:"
There is no problem fitting a smaller card in a bigger slot, it's designed that way.
They would basically need a 8-port SAS-controller for x4 PCI-e support. At least as far as Marvell is concerned.
-=Hulk=- - Thursday, April 19, 2012 - link
1. No PCIe x2 is not allowed. Look at recent news, Intel wanted a 2x PCIe as standard but the consortium on charge of PCIe doesn't agree with the idea. If you want tu use 2 PCIe lane with this Marvell controller you have to use a 4x PCIe controller at least.You can't aggregate 2 x1 lanes. It works only if the client chip has a internal or external PCIe switch (like those from PLX)
2. Look at the picture! It's a x1 connector on the card. How do you want the Marvell controller to use 2 PCIe lanes???
This SSD card won't exceed ~480-500MB/s.
-=Hulk=- - Thursday, April 19, 2012 - link
Sorry it's not a 1x on the picture :D. But it's not standard.daytrippr - Thursday, April 19, 2012 - link
While PCIe does not provide for a 2x physical slot, it unambiguously supports a 2x electrical interface. Plug a x2 card into a x4 slot and you'll get the much higher performance numbers you seem to want to ignore...Penti - Friday, April 20, 2012 - link
Slot != Card that was what my post also said, we all know that it's x1, x4, x8, x16 and x32 as far as slots go. Or mechanical interface. When it's internal on the motherboard it doesn't matter and saves lanes using only as many as you need. Using a >x4 slots leaves two or more lanes unused. There are some motherboards with electrical x2 slots though. No Mac of course. Here we simply have a card with a SATA-controller designed to be used with two lanes.AnnihilatorX - Friday, April 20, 2012 - link
But did Intel mean a dedicated x2 physical slot or just a x2 electrical interface?extide - Wednesday, April 25, 2012 - link
Yes. PCIe x2 IS allowed. That card IS a 2x physical AND electrical card. PERIOD.joevt - Sunday, May 27, 2012 - link
A problem with x2 is that PCIe chips aren't required to negotiate 2 lanes. For example, the ESB2 south bridge (PCIe 1.0) in the 2006, 2007, and 2008 Mac Pro's and other old PC's can only negotiate 1, 4, or 8 lanes. The Accelsior in such a slot will only use a link width of 1 lane for 195 MB/s read/writes. Even with a 2 lane link width (as provided by the 5000 series north bridge chip in the Mac Pro 2006, 2007), the Accelsior will only do 380 MB/s in a PCIe 1.0 slot.I wonder if any PCIe 2.0 chips have the same link width limitation. The 5400 north bridge chip in the Mac Pro 2008 handles x2 fine.
Gary Byatt - Thursday, April 19, 2012 - link
I have been using the OCZ Revodrive 3 x2 240GB since November in a self-build with a i7 2700K CPU that clocks up to 3.9GHz on load and 8GB of 1833MHz RAM.I get a windows performance index of 7.9 on the OCZ as my boot drive. I tested it after a few months of real world usage and reported my findings here: http://www.linkedin.com/groups/240GB-OCZ-Revodrive...
In short I got a best write of 1129.555 MB/s and a best read of 1520.693 MB/s. The best write speed is a little short of the advertised 1225 MB/s but it is in use as the boot device on machine in real use, not sitting idle waiting for test data in a lab. The best read speed is actually slightly better than the advertised 1500MB/s, so “figures meant for advertising” or not it is a lot better in real use than the OWC claim.
I expect to see storage from OCZ in the not too far distant future that saturates 4 lanes of PCIe v2 (2GB/s).
zanon - Thursday, April 19, 2012 - link
Yep. The OCZ is also about 25-30% more expensive, but it is significantly faster then the price margin. That means, by pure advertised numbers at any rate (which doesn't seem entirely unreasonable in this case since they're both SF2000 based), that it's about 50-60% more bang-for-buck. Absolute pricing will probably matter to some people, who may prefer to pocket the extra $130 rather then pay more for better performance. But it certainly looks like the primary value is the AHCI compliance.
Hopefully that becomes the standard practice for everyone going forward.
vectorm12 - Thursday, April 19, 2012 - link
Considering there's still a significant problem with SF-controllers and dealing with uncompressible data I'd rather hold out for the Revodrive 4 which will likely feature similar controllers to the Vertex4.In my mind I'd find it hard to believe anyone looking for a boot/software device would go PCIe at this point. Seems to me these drives would be aiming towards active workspace considering the added cost.
I doubt you could measure much of a performance increase from booting/loading software off of one of these compared to a Vertex3/4 of the same size.
johnsmith9875 - Tuesday, May 8, 2012 - link
They should put heatsinks on the chips. Not because it needs them, but at that price I want something neat looking other than a PCB with some boring black chips on it.