What's this? The long awaited specs for Intel's third generation SSD? Indeed.

Internally it’s called the Postville Refresh (the X25-M G2 carried the Postville codename), but externally it carries the same X25-M brand we’ve seen since 2008. The new drive uses 25nm IMFT Flash, which means we should get roughly twice the capacity at the same price. While Intel is sampling 25nm MLC NAND today it's unclear whether or not we'll see drives available this year. I've heard that there's still a lot of tuning that needs to be done on the 25nm process before we get to production quality NAND. The third generation drives will be available somewhere in the Q4 2010 - Q1 2011 timeframe in capacities ranging from 40GB (X25-V) all the way up to 600GB.

Despite the Q1 release of Intel’s 6-series chipsets, Intel is listing the new X25-M as being 3Gbps SATA only. The SATA implementation has been updated to support ATA8-ACS so it’s possible we may see official 6Gbps support once Intel has a chipset with native support.

The new drive’s performance specs are much improved. The comparison between old and new is below:

Intel Consumer SSD Comparison
  Intel X25-M G2 (34nm) Intel X25-M G3 (25nm)
Codename Postville Postville Refresh
Capacities 80/160GB 80/160/300/600GB
NAND IMFT 34nm MLC IMFT 25nm MLC
Sequential Performance Read/Write Up to 250/100 MB/s Up to 250/170 MB/s
Random 4KB Performance Read/Write Up to 35K/8.6K IOPS Up to 50K/40K IOPS
Max Power Consumption Active/Idle 3.0/0.06W 6.0/0.075W
Total 4KB Random Writes (Drive Lifespan) 7.5TB - 15TB 30TB - 60TB
Power Safe Write Cache No Yes
Form Factors 1.8" & 2.5" 1.8" & 2.5"
Security ATA Password ATA Password + AES-128

If these numbers are accurate, the new Intel drive should be roughly equal to Crucial’s RealSSD C300 and SandForce SF-1200 based drives. There are many different ways to measure this data however so the numbers may be higher or lower in our tests. Note that performance could also go up by the time drives are available as there's still a lot of tuning going on right now. I'd say that at these performance levels Intel had better be very aggressive with pricing because I'm expecting much better from the next-generation SandForce drives.

Write amplification appears to be more under control with the third gen X25-M. Intel upgraded the total 4KB random writes spec from 7.5TB - 15TB on the G2 to a much higher (and wider) range of 30TB to 60TB depending on drive and spare area.

Intel hasn’t disclosed any information about spare area, but given the huge increase in longevity of the drives I suspect that spare area has gone up as well (at least on the larger drives).

The G1 and G2 drives didn’t store any user data in the off-controller DRAM, the third gen drive changes that. A large part of why the C300 is so quick has to do with its large external DRAM, something Intel has avoided implementing in the past due to the associated risk of data loss. Intel refers to the 3rd gen X25-M has having a power safe write cache, which sounds to me like it has an external DRAM paired with a big enough capacitor to flush the cache in the case of sudden power loss.

Full disk encryption is the next big feature on the Postville Refresh. You get AES-128 support on the consumer drives. I’m guessing there’s a new version of the SSD Toolbox in the works as Intel is also promising Windows based firmware updates.

The new X25-M will be available in both 1.8” and 2.5” versions. The 1.8” drive tops out at 300GB, you’ll need the 2.5” form factor for 600GB.

In addition to the new X25-M there’s a new X25-E due out in Q1 2011. Codenamed Lyndonville, this will be the first Intel Enterprise SSD to use MLC flash. It’s not quite the same MLC used on the consumer drives but rather a modification of the 25nm process that trades data retention for longevity.

Standard MLC will last for 12 months after all erase/program cycles have been consumed. Enterprise grade MLC will last only 3 months after exhausting all erase/program cycles but will instead support many more cycles per cell.

The X25-E improves specs compared to its predecessor:

Intel Enterprise SSD Comparison
  Intel X25-E (50nm) Intel X25-E (25nm)
Codename Ephraim Lyndonville
Capacities 32/64GB 100/200/400GB
NAND IMFT 50nm SLC IMFT 25nm eMLC
Sequential Performance Read/Write Up to 250/170 MB/s Up to 250/200 MB/s
Random 4KB Performance Read/Write Up to 35K/3.3K IOPS Up to 50K/5K IOPS
Max Power Consumption Active/Idle 3.0/0.06W 5.0/0.095W
Total 4KB Random Writes (Drive Lifespan) 32GB: 1PB
64GB: 2PB
100GB: 900TB - 1PB
200GB: 1PB - 2PB
400GB: 1.4PB
Power Safe Write Cache No Yes
Form Factors 2.5" 2.5"
Security ATA Password ATA Password + AES-128

Larger capacities, higher performance, AES-128 support and comparable lifespans to the old X25-Es are all in store early next year. Note that Intel tests 4KB random write performance differently on enterprise vs. consumer drivers so you can’t directly compare the numbers between the X25-M and X25-E. The X25-E will be 2.5” only.

Intel isn’t the only one working on a controller update. SandForce and Indilinx are both heading towards production versions of their next-generation controllers. I expect we’ll see preview class hardware before the end of the year, with mainstream availability in Q1 2011.

POST A COMMENT

97 Comments

View All Comments

  • apriest - Tuesday, October 05, 2010 - link

    Yay!!! Now, how about trim with RAID 0? I'm still waiting for that with my G2 SSDs... Reply
  • HachavBanav - Tuesday, October 05, 2010 - link

    Some facts:
    -Any OS uses one (or more) page size for all its io
    -A RAID array is build using a mix of a "stripe size" on each of the drives participating in the array
    -TRIMing less than a stripe is a very difficult task even for RAID 0

    Example with any Windows 7 installed on a RAID 0 array build with 2x HDD and a 64KB stripe
    ==> The OS cluster size is usually 4KB, so you may (simplified process):
    -Write a small file on the first clusters (#0 = 4KB)
    -The RAID controller read the first 64KB stripe of the first HDD and rewrite it with 4KB updated
    -Write another small file on the 2d clusters (#1 = 4KB)
    -The RAID controller read the first 64KB stripe of the first HDD and rewrite it with 4KB updated
    -Delete the first file will send a TRIM command to #0
    ==> Now, if the SSD TRIM this #0, what would be the 4KB values read from the first 64KB stripe for my 3rd small file to be written ?

    This implies that the controller should:
    -be able to read a dedicated special value for a TRIMmed sector
    -be able to ignore writing those sectors while updating a RAID 0/1 stripe
    -be able to compute parity on those sectors

    ==> And please keep in mind that TRIMing is not a mandatory feature, it just helps the SSD Garbage collector in its erase work...most SSD has some reserved FREE space that makes this TRIM feature as an obsolete one
    Reply
  • apriest - Wednesday, October 06, 2010 - link

    It's Intel... they're smart... :-)

    Seriously, nice explanation of the complexity of trim on RAID.
    Reply
  • MrBrownSound - Tuesday, October 05, 2010 - link

    I'm Running two individual X25-M's G2 versions separately; no raid 0 :( . I really am looking forward to trim adaptation in all the popular Operating Systems. Reply
  • JHX - Tuesday, October 05, 2010 - link

    Anand, for your next SSD tests, please do some full-disk encryption test and see how it affects performance.

    I have an Intel G2, and using TrueCrypt full disk encryption, I only get 40-45% of the read and write performance.

    Is that something you've looked into?
    Reply
  • DesktopMan - Tuesday, October 05, 2010 - link

    I'm also a little disappointed encryption hasn't been tested at all. SSDs make a lot of sense in laptops, and it's in those setups you're likely to see full disk encryption.

    SandForce has built in encryption but I haven't seen anything about that anywhere.
    Reply
  • eanazag - Tuesday, October 05, 2010 - link

    OCZ's pro models which cost about twice as much as the regular Vertex 2 drives support it with their own software. Sandforce's implementation is not OPAL spec. Reply
  • eanazag - Tuesday, October 05, 2010 - link

    Bitlocker will give you 60+% performance. Hardware encrypted drives perform at full speed. The encryption runs all the time. Only the authentication gets turned on or off for supporting it. Reply
  • JHX - Tuesday, October 05, 2010 - link

    Nope, I tried Bitlocker, their claims are wrong. The SSD takes just as big of a performance hit as it does with TrueCrypt. Reply
  • ssj4Gogeta - Wednesday, October 06, 2010 - link

    Won't encryption be processor limited as opposed to depending on the SSD?
    Maybe you should try using Westmere (Core i5) as they have hardware encryption support and will be much faster. (I think Windows 7 Bitlocker has support for it? Not sure.)
    The desktop Core i5 6xx and mobile Core i5 (except 450 and 430) and i7 980X have AES-ni support.
    Reply

Log in

Don't have an account? Sign up now