Intel's 34nm SSD Preview: Cheaper and Faster?
by Anand Lal Shimpi on July 21, 2009 11:00 AM EST- Posted in
- Storage
The rumors are true, Intel’s 2nd generation SSDs are available starting today.
The high level details are pretty interesting:
- The new drives will be available in 80GB and 160GB sizes and are still called the X25-M and X18-M. The X18-M will start shipping later this quarter.
- 34nm flash (down from 50nm in the original X25-M), allows Intel to include roughly twice the flash in the same size die.
- The enterprise SLC version doesn’t get the 34nm treatment at this point.
- The smaller flash die results in lower prices, the 80GB model will sell for $225 while the 160GB version should sell for $440.
- Best case read/write latency has been improved (more details below).
- The 34nm drives have a new controller and new firmware, also contributing to better performance (2 - 2.5x more 4KB random write IOPS than the old drive!). Enterprise level workstation/database apps should see an immediate performance benefit, client desktop performance is unknown. Don't expect a significant increase in PCMark or SYSMark scores, but in real world usage the new drives could feel faster.
- The new controller is Halogen-free (the old one wasn’t) so Apple could theoretically use the new drives in their systems without being un-green.
- TRIM isn’t yet supported, but the 34nm drives will get a firmware update when Windows 7 launches enabling TRIM. XP and Vista users will get a performance enhancing utility (read: manual TRIM utility). It seems that 50nm users are SOL with regards to TRIM support. Bad form Intel, very bad form.
- I get my drive this week, so expect a review to follow.
Overall it’s an evolution of the X25-M, and not a revolutionary new design. The focus of the evolution is definitely price. Intel wants the X25-M to be used, not only at the high end, but even in mainstream PCs. At $225 for an 80GB drive, the new X25-M is currently cheaper than most Indilinx based drives on the market:
Drive | NAND Capacity | Cost per GB | Price |
Intel X25-M (34nm) | 80GB | $2.81 | $225 |
Intel X25-M (34nm) | 160GB | $2.75 | $440 |
OCZ Vertex (Indilinx) | 64GB | $3.41 | $218 |
OCZ Vertex (Indilinx) | 128GB | $3.00 | $385 |
Patriot Torqx (Indilinx) | 64GB | $3.48 | $223 |
Patriot Torqx (Indilinx) | 128GB | $2.85 | $365 |
OCZ Agility (Indilinx, non-Samsung Flash) | 64GB | $2.77 | $177 |
OCZ Agility (Indilinx, non-Samsung Flash) | 128GB | $2.57 | $329 |
OCZ Summit (Samsung) | 128GB | $3.04 | $389 |
There’s going to have to be a price correction from the competition. We may also see more manufacturers branching out to different flash memory vendors to remain price competitive (similar to what OCZ did with the Agility line).
The Specs Breakdown
The major differences between the 1st and 2nd gen X25-M are highlighted in the table below:
X25-M Gen 1 | X25-M Gen 2 | |
Flash Manufacturing Process | 50nm | 34nm |
Flash Read Latency | 85 µs | 65 µs |
Flash Write Latency | 115 µs | 85 µs |
Random 4KB Reads | Up to 35K IOPS | Up to 35K IOPS |
Random 4KB Writes | Up to 3.3K IOPS | Up to 6.6K IOPS (80GB) Up to 8.6K IOPS (160GB) |
Sequential Read | Up to 250MB/s | Up to 250MB/s |
Sequential Write | Up to 70MB/s | Up to 70MB/s |
Halogen-free | No | Yes |
Price | $345 (80GB) $600 - $700 (160GB) | $225 (80GB) $440 (160GB) |
The move to 34nm gives Intel the ability to both decrease costs and increase capacity. It now costs Intel the same to make a 160GB drive as it used to cost to make an 80GB drive, and about half to make an 80GB drive. Given the current cost structure, I’d say there’s still more room for Intel to drop prices but there’s just no need to given the competitive landscape.
Better Performance
The performance of the new drives is improved. Read and write latency are both improved by around 30%. These figures are for a single sequential operation, so you're looking at best case performance improving on the drive. Intel also tweaked the controller and its firmware to further improve performance; the result is much faster 4KB random writes on the new drives.
The real world benefits are difficult to predict. If you're running an I/O intensive app on the desktop then you'll see a definite improvement from the new drive. More typical desktop/notebook workloads probably won't see a tremendous difference between the 1st and 2nd gen drives. Large file sequential read/write speed remains unchanged and although there have been tweaks to the controller’s algorithms, the overall architecture hasn’t changed either.
I’ll be able to confirm for sure when I get a drive this week.
Other Spec Changes
There are two other specifications that have changed, one for the better and one for the worse with the new drive:
X25-M Gen 1 | X25-M Gen 2 | |
Idle Power (MobileMark 2007) | 65mW | 75mW |
Operational Shock | 1,000G for 0.5 msec | 1,500G for 0.5 msec |
Load power hasn't changed between the drives, they are both at 150mW. Idle power went up by 10mW thanks to the new controller. These are "typical" values while running MobileMark 2007, so I'm not sure if things change at all if you're looking at real world workloads. Intel claims that the new drive doesn't deliver worse battery life than the old one, but I'll have to verify.
The maximum amount of operational shock that the 34nm drive can take went up by 50%; 1500Gs for 0.5 msec sure seems like a lot. I'm not sure if anything changed internally to allow for this higher rating or if it's simply the result of better manufacturing/improving production.
Halogen Free, Apple Friendly
A little known fact about the original X25-M was that its controller wasn’t Halogen-free. Because Intel used Halogens in the first controller, companies that had strict environmental restrictions (e.g. Apple) wouldn’t touch the drives.
Apple couldn't claim BFR-free on its new MacBook Pro if it used Intel's 1st gen X25-M. Bromine is a Halogen.
The new drive has a new controller and it is Halogen free. For Apple to glance over the X25-M in its mobile lineup now would be a serious mistake.
How to Tell the Drives Apart
Despite the price drop and internal changes, Intel is still calling these things the X25-M and X18-M. So how do you tell the new drives apart from the old ones? It all comes down to the part number; if the last two digits are a G1 then it’s the old drive, if they are a G2 it’s the new one. If you have them in hand, the new drives are silver, the old ones are black.
X25-M Gen 1 | X25-M Gen 2 | |
80GB 2.5" | SSDSA2MH080G1 | SSDSA2MH080G2 |
160GB 2.5" | SSDSA2MH160G1 | SSDSA2MH160G2 |
80GB 1.8" | SSDSA1MH080G1 | SSDSA1MH080G2 |
160GB 1.8" | SSDSA1MH160G1 | SSDSA1MH160G2 |
TRIM Support: Not For 50nm Drives
Part of today’s announcement is the fact that Intel will enable TRIM on these 34nm drives when Windows 7 ships. Intel is planning on releasing a user downloadable firmware update that will enable TRIM support. Windows Vista and XP users will get a performance enhancement tool that presumably will just manually invoke the TRIM command. I suspect that Intel is waiting until Windows 7 to enable TRIM support is to make sure that everything is thoroughly tested. As we’ve seen with other attempts to enable TRIM, it’s a tricky thing to do.
The disappointing part of the announcement is that there’s no TRIM support for the first gen 50nm drives. As far as I can tell, this isn’t a technical limitation of the drives, but rather something Intel is choosing to enable only on the 34nm products.
Final Words...for now
I'm still waiting on my 34nm review sample, as soon as I get it I'll start working on a full review. I've already started work on the newest SSDs from the competition, so expect something soon.
73 Comments
View All Comments
has407 - Tuesday, July 21, 2009 - link
From the Intel product page (http://www.intel.com/design/flash/nand/mainstream/...">http://www.intel.com/design/flash/nand/mainstream/...:"10 Parallel Channel Architecture with 34nm MLC ONFI 1.0 NAND (34nm)"
[NOTE: same for 50nm devices]
Based on numbers from Intel/ONFI presentations (http://onfi.org/presentations/)">http://onfi.org/presentations/)...
1. Read and write performance is in line with what you'd expect from MLC with 4KB pages and 10 channels, and 1x 2l-plane die/channel: 10x 30MBs = 300MBs sustained read; 10x 7MBs = 70MBs sustained write; read is obviously SATA limited. (I assume they're actually using 2x 32Gb/4GB 34nm stacked die/channel x10 channels for the 80GB drive. Not sure about the 160GB drive.)
2. It appears Intel has gone with more channels and fewer die/channel and maybe smaller page sizes, whereas others have gone with fewer channels and more die/channel and maybe larger page sizes. Intels approach is more expensive, and provides better random write performance at the expense of sequential write performance.
That said, it's curious that the higher write IOPs and reduced write latency, especially on the 160GB drive, don't translate to higher sequential write performance. Maybe a controller limitation? Or maybe the 34nm parts are actually bit slower write than the 50nm parts, but effective throughput is the same due to reduced latency? Or maybe they don't want to risk cannibalizing the E drive market (yet)?
has407 - Tuesday, July 21, 2009 - link
Couple other thoughts... Maybe Intel is being very conservative with these drives and doing read-verify for writes, or intentionally throttling write rates to avoid potential wear problem? I can only imagine the sh*tstorm if drives started failing after a couple years, even if it was a minority of drives/users.Looking at the SLC E drives, they spec the 32GB drive at 1PB (10^15) lifetime random write, and the 64GB drive at 2PB random write. That yields a write endurance of ~32K cycles/block. MLC drives will be lower, and smaller 34nm geometry will likely be lower still(?) without something else such as additional ECC/spares to compensate. Unfortunately they don't spec the M drives other than a "Minimum of five years of useful life". (Interestingly, both E and M drives are spec'd with the same BER of 1 sector per 10^15 bits read.)
p.s. Sorry the ")" got included in the hyperlink in my previous post; proper URL's are:
http://www.intel.com/design/flash/nand/mainstream/...">http://www.intel.com/design/flash/nand/mainstream/...
http://onfi.org/presentations/">http://onfi.org/presentations/
Kurotetsu - Tuesday, July 21, 2009 - link
I always thought it was because they wanted the M-series to focus on consumer desktop performance. Which means maximizing random performance. It seems to have worked too, as Intel drives are still the SSDs to beat right now.stevty2889 - Tuesday, July 21, 2009 - link
I don't see any reason why Intel can't reach the same write speeds, I would imagine this 256GB crucial drive is using the same flash, and has similar read/write speeds to the newer samsung's.http://www.crucial.com/store/partspecs.aspx?IMODUL...">http://www.crucial.com/store/partspecs.aspx?IMODUL...
NeBlackCat - Tuesday, July 21, 2009 - link
I've seen speculation that the new drives would have built-in encryption, no sign of that?Anand Lal Shimpi - Tuesday, July 21, 2009 - link
No encryption for this controller :)Take care,
Anand
Soldier1969 - Tuesday, July 21, 2009 - link
It's nice to see ssd drives dropping in price but the capacity still isn't there for the price. I'll stick with my 600gb velociraptors in raid 0. When ssd drives bring out 500 gb and 1tb drives for around $200 or less I'll get a set but I'm not paying over $400 for less than 200gb no.benrico - Tuesday, July 21, 2009 - link
I think a lot of us are jumping on the Patriot torqx drives because they are hitting a great $/gb (see table). Any info on the new crop of drives would be immensely useful.Also, Intel is totally missing my price point here, while it is definately a no brainer to get an g2 intel ssd, I dont want a 80gig drive, and 400+ for the 160gig still seems too much for storage.
My guess is that those prices will be fixed for some while....we shall see.
semo - Friday, July 24, 2009 - link
Still don't get SSDs do you? They're not for storage only but for everyday performance. If you do CAD work or movie editing then money in memory+CPU is well spent but not so much for everyday computing.HDDs are for storage people!
AnnonymousCoward - Friday, July 24, 2009 - link
Intel is missing MY price point by not offering a 40GB model! WinXP and my collection of programs don't exceed 10GB, which leaves 30GB for 5-8 games. 40GB isn't by any means a small amount of flash for a single unit, when you consider that USB drives have been coming in 16MB-4GB for years. I'm not willing to blow $225, but I might spend $115.On another point, let's look at Intel's price margin. $20 8GB USB drives are everywhere, and they are at least 50nm tech. Buying 10 for $200 would give 80GB, and that's with a loss of efficiency with 10 form factors and controllers. For at least the same price margin, Intel could sell their 34nm 80GB drive for $100.