I'd be happier with an announcement that a firmware upgrade is available to fully implement TRIM on a Raid setup. As much as I'm always ready to upgrade to the biggest, latest and greatest; no TRIM support in Raid is a potential huge show stopper to upgrade from current Intel 34nm G2 SSDs on my annual next build. Which ever manufacturer comes out with a proper/equivalent solution to keep SSDs in multiple Raid configurations operating at their peak efficiency is going to get my money.
TRIM is a horribly implemented command in SATA drives, as TRIM is a synchronous command. All the read-ahead and write-behind operations block waiting for TRIM to finish. One proposal from the Linux community was to allocate blocks from a "already written" pool instead of using TRIM. This would allow the never-written area of the drive unwritten as long as possible.
Yeah but this will prevent wear leveling system to do its work.
For RAID with TRIM, on OCZ forum, someone pointed out the RAID capability of Win7 that would probably allow TRIM and RAID together. But, of course, software RAID isn't as good as hardware RAID.
So the 25nm process will be rolled out at all their plants? I was wondering what IM Flash Technologies was, since they built a new plant near my house.
So I guess ~5nm will be the limit before you reach atomic levels, aye? Might be good news or bad news, I'm not sure.
Immersion lithography is only a stopgap solution because it relies on the greater index of refraction of water. Since we really don't have many fluids with an even larger index of refraction, I'm highly curious how low 2x and 1x nm will be attempted - double patterning? nanofabrication?
Personally, I think true nanomachines (as in bottom-up assembly?) will appear at 11 nm or below, based on my rudimentary understanding of chip fabrication and reading on wikipedia. You simply can't make silicon much smaller than that.
With a 1,414x shrink between each full process node, you have 4 shrinks left after 25nm until you reach 5nm. With 15 months between each process node that gives 5 years... At that point you need to start 3D processes to scale further, or start using nano-technology.
ONFI 2.2 specifies 200MB/s interface bandwidth max for each flash channel, you may have multiple ICs (packages), each with multiple dies, pr flash channel. It will essentially enable higher bandwidth at fewer channels, wich means lower controller design costs and less complexity. You also run into a push to larger page sizes and block sizes in order to scale the bandwidth effectively from a single IC. 4KB random read bandwidth will not take a huge hit (rather stay about the same as today), as partial page reads are easily done, it will however not scale from todays 32nm chips (pr channel).
I am wondering if the other memory vendors will have to do the same? Also what will the perf impact be as we are already dealing with filesystems designed for spinning disks? They have added trim/gc items but wondering if the whole filesystem is next thing to really need a overhaul.
Since mine is an early-2008 model, it can't take full advantage of SATA-connected SSDs, being limited to 150 MB/sec on the SATA interface (and Apple shows no signs of upgrading the firmware). By contrast, PCI Express Cards are 2.5 Gb/sec, which is about 310 MB/sec. This would also effectively give me two drives, leaving the HD for mass storage.
FileMate does make some ExpressCard SSDs. However, while faster than a conventional HD, they are significantly slower than the current Intel SSDs.
just swap out your DVD drive for another hard drive. I got an Intel 80GB in place of my DVD drive and a 500GB Seagate in the main drive bay. Though, I hardly ever used my DVD drive.
Thought of that, but it voids the remaining 2 years on my extended warranty (yes, it's an early 2008 model, but I purchased it in Feb. 2009 on clearance). And I do actually use the DVD drive with some frequency.
Do you know how many NAND die can be assembled in each package?
I remember seeing the circuit board of the 160GB G2, and it looked liked it only had 10 NAND packages (unless they were double-stacked?) So that would be 16GB per package, or 4 of the 32 Gbit die per package.
But the 80GB G2 also has 10 packages, I think (at least, the controller is 10 channel). So is it using only 2 die per package?
Can more than 4 die be assembled in each package? Having, say, 8 of the upcoming 64 Gbit die in a package blows my mind.
I thought a floating gate device was technically a charge trapping device since it involves some kind of charge transfer and holding in the electrically isolated floating gate. Is the only difference between a "charge trapping" device and a floating gate device something like the floating gate being made of a nitride? Or is all this just bad nomenclature? Thanks!
They are the same thing. All charge trapping devices like EEPROM, NOR, and NAND uses floating gate and it is how you arrange them that makes the difference.
Since I can get an X-25M G2 80GB for $150, and 160GB for $300, should I just take the leap and get one of those, or should I wait for something better? I really wish
raid had trim support.
Intel's Employee Purchase Program is for employees only, and not for resale. If your friend gets caught selling the SSDs bought through EPP, he can get fired.
Also, SSDs are out-of-stock at EPP right now, due to high channel demand.
Change that "can get fired" to "will be fired". For each purchase, employees acknowledge its for their own personal use only. Companies are pretty strict about this. Not smart to risk your job over saving a few bucks for a "friend".
I think maybe he was being sarcastic?
The logic vexed in the article mentioned twice the capacity at the same production cost. This would be 'wonderful' if this formula funneled down to the consumer. Kingston announced the 40gv now for 85.00 back in October ,09. that never was really available. Now its 30g for 90.00, supposedly coming ? Could the consumer get a 30g drive for 45.00 when this new tech is in full operation ? And with 230/170 r/w speeds with the new controllers. I doubt it, when the WD velociraptor 150g is still a 160.00. Maybe industry wide competition could make it happen.
well, they didn't "announce" the 40GB at $85....that's where it wound up with rebates + promo pricing. You always have to keep in mind that there is overhead for the controller IC, casing, marketing, manufacturing, ect. So prices will not scale 2:1..
All that said, I'd fricken love to find a 32GB SSD for $45...hell, even $75. 1.8" would be even better!
For $300, I would grab the 160GB G2 in a heartbeat. I paid $450 for mine and it's worth every penny.
Yes, newer technology is going to come out that will make these feel slow, small, and generally outdated. For the most part, I don't play the waiting game unless something is right around the corner (for example, I wouldn't advise anyone to make a major graphics card purchase right now until we see what happens in March).
I dont think you'll see SSDs anytime soon (read: years) that make current (intel) SSDs feel slow unless you like to play benchmarks all day long or are exclusiveley interested in sequential read/write performance.
You will almost certainly find an SSD that will make current SSDs feel *small*. New SSDs could easily feel faster than non-TRIM current SSDs. You might need a later motherboard with some yet-to-be-speced SATA3 system to make a state-of-the-art (1/31/2010) seem slow, and then only being used for virtual memory swap.
You should buy products when you need them. Current guesses on Fermi is that it should be a bit faster than a 5870, but it is going to cost a ton. There will also not be anything until mid-summer to fill the needs of consumers that like to game, but don't have a trust fund. And soon after Fermi the new generation of AMD/ATI will be rolling out (5XXX was a die shrink and DX11 only, not a new design). Just buy when you need it and stop trying to time the market.
Anand was right about the smaller drives in that article back in 2009. The supposed price drops for the OCZ Colossus line did not materialize with current low street prices being about 50% higher than MSRP shown in the article. Too bad ....
Most of the benchmarks are not even reaching the practical / theoretical limits of SATA 3Gbps speed. So why would a fatter pipe , SATA 6Gbps allow much higher speed with the same controller?
And have you figured out why Kingston V+ is so fast on benchmarks yet? I was asking last time and no one provide an solid answer. Like you have mentioned, the 4 corners of SSD performance, Seq R/W, Random R/W. Kingston perform very well in Seq R/W and only the low end for Random R/W. However it was the fastest in multiple Anand / Synthetic benchmarks.
P.S - VR-Zone just posted Next Gen Intel SATA 6Gps SSD will be out on 1st of March. Will we get a preview before that?
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
39 Comments
Back to Article
kunedog - Thursday, February 18, 2010 - link
As of yesterday, Newegg has the OEM G2 X-25M 80GB for $220 shipped and the 160GB for $430 shipped:http://www.newegg.com/Product/Product.aspx?Item=N8...">http://www.newegg.com/Product/Product.aspx?Item=N8...
http://www.newegg.com/Product/Product.aspx?Item=N8...">http://www.newegg.com/Product/Product.aspx?Item=N8... (comes with a free USB enclosure)
This is the FIRST time they have been in stock for more than a day at anywhere near the retail prices that Anand predicted over SIX MONTHS ago:
http://www.anandtech.com/storage/showdoc.aspx?i=36...">http://www.anandtech.com/storage/showdoc.aspx?i=36...
Butch306 - Monday, February 1, 2010 - link
I'd be happier with an announcement that a firmware upgrade is available to fully implement TRIM on a Raid setup. As much as I'm always ready to upgrade to the biggest, latest and greatest; no TRIM support in Raid is a potential huge show stopper to upgrade from current Intel 34nm G2 SSDs on my annual next build. Which ever manufacturer comes out with a proper/equivalent solution to keep SSDs in multiple Raid configurations operating at their peak efficiency is going to get my money.jhh - Tuesday, February 2, 2010 - link
TRIM is a horribly implemented command in SATA drives, as TRIM is a synchronous command. All the read-ahead and write-behind operations block waiting for TRIM to finish. One proposal from the Linux community was to allocate blocks from a "already written" pool instead of using TRIM. This would allow the never-written area of the drive unwritten as long as possible.kensiko - Tuesday, February 2, 2010 - link
Yeah but this will prevent wear leveling system to do its work.For RAID with TRIM, on OCZ forum, someone pointed out the RAID capability of Win7 that would probably allow TRIM and RAID together. But, of course, software RAID isn't as good as hardware RAID.
AtwaterFS - Sunday, January 31, 2010 - link
since SSD's are the cause for a fatter SATA pipe...Xyn - Sunday, January 31, 2010 - link
So the 25nm process will be rolled out at all their plants? I was wondering what IM Flash Technologies was, since they built a new plant near my house.So I guess ~5nm will be the limit before you reach atomic levels, aye? Might be good news or bad news, I'm not sure.
jimhsu - Sunday, January 31, 2010 - link
Immersion lithography is only a stopgap solution because it relies on the greater index of refraction of water. Since we really don't have many fluids with an even larger index of refraction, I'm highly curious how low 2x and 1x nm will be attempted - double patterning? nanofabrication?jimhsu - Sunday, January 31, 2010 - link
Personally, I think true nanomachines (as in bottom-up assembly?) will appear at 11 nm or below, based on my rudimentary understanding of chip fabrication and reading on wikipedia. You simply can't make silicon much smaller than that.GullLars - Sunday, January 31, 2010 - link
With a 1,414x shrink between each full process node, you have 4 shrinks left after 25nm until you reach 5nm. With 15 months between each process node that gives 5 years... At that point you need to start 3D processes to scale further, or start using nano-technology.ONFI 2.2 specifies 200MB/s interface bandwidth max for each flash channel, you may have multiple ICs (packages), each with multiple dies, pr flash channel. It will essentially enable higher bandwidth at fewer channels, wich means lower controller design costs and less complexity. You also run into a push to larger page sizes and block sizes in order to scale the bandwidth effectively from a single IC. 4KB random read bandwidth will not take a huge hit (rather stay about the same as today), as partial page reads are easily done, it will however not scale from todays 32nm chips (pr channel).
Turas - Sunday, January 31, 2010 - link
I am wondering if the other memory vendors will have to do the same? Also what will the perf impact be as we are already dealing with filesystems designed for spinning disks? They have added trim/gc items but wondering if the whole filesystem is next thing to really need a overhaul.semo - Monday, February 1, 2010 - link
of course it needs an overhaul but it won't happen. we'll be talking about LBAs in 20 years time. Just like x86chemist1 - Saturday, January 30, 2010 - link
I'd like to see someone manufacture one of these super-fast SSDs in a PCI ExpressCard/34 format, which would fit flush into the PCI slot on my MacBook Pro. [See http://www.macworld.com/article/145185/2009/12/mbp...">http://www.macworld.com/article/145185/2009/12/mbp...]Since mine is an early-2008 model, it can't take full advantage of SATA-connected SSDs, being limited to 150 MB/sec on the SATA interface (and Apple shows no signs of upgrading the firmware). By contrast, PCI Express Cards are 2.5 Gb/sec, which is about 310 MB/sec. This would also effectively give me two drives, leaving the HD for mass storage.
FileMate does make some ExpressCard SSDs. However, while faster than a conventional HD, they are significantly slower than the current Intel SSDs.
ksherman - Sunday, January 31, 2010 - link
just swap out your DVD drive for another hard drive. I got an Intel 80GB in place of my DVD drive and a 500GB Seagate in the main drive bay. Though, I hardly ever used my DVD drive.chemist1 - Sunday, January 31, 2010 - link
Thought of that, but it voids the remaining 2 years on my extended warranty (yes, it's an early 2008 model, but I purchased it in Feb. 2009 on clearance). And I do actually use the DVD drive with some frequency.milli - Saturday, January 30, 2010 - link
Toshiba already has 3-bits-per-cell & 4-bits-per-cell MLC NAND.ioannis - Saturday, January 30, 2010 - link
that is all nice and good, but is there any news on memristors? I know it's HP and not Micon/Intel working on those, but just wondering...jwilliams4200 - Saturday, January 30, 2010 - link
Anand:Do you know how many NAND die can be assembled in each package?
I remember seeing the circuit board of the 160GB G2, and it looked liked it only had 10 NAND packages (unless they were double-stacked?) So that would be 16GB per package, or 4 of the 32 Gbit die per package.
But the 80GB G2 also has 10 packages, I think (at least, the controller is 10 channel). So is it using only 2 die per package?
Can more than 4 die be assembled in each package? Having, say, 8 of the upcoming 64 Gbit die in a package blows my mind.
PandaBear - Saturday, January 30, 2010 - link
Toshiba have done 16 dies, but the yield is very low, the economy of scale stop at 4 die per chip.AnandMeat - Saturday, January 30, 2010 - link
Anand,I thought a floating gate device was technically a charge trapping device since it involves some kind of charge transfer and holding in the electrically isolated floating gate. Is the only difference between a "charge trapping" device and a floating gate device something like the floating gate being made of a nitride? Or is all this just bad nomenclature? Thanks!
Meat
PandaBear - Saturday, January 30, 2010 - link
They are the same thing. All charge trapping devices like EEPROM, NOR, and NAND uses floating gate and it is how you arrange them that makes the difference.ajhix36 - Saturday, January 30, 2010 - link
Since I can get an X-25M G2 80GB for $150, and 160GB for $300, should I just take the leap and get one of those, or should I wait for something better? I really wishraid had trim support.
ajhix36 - Saturday, January 30, 2010 - link
No its 100% true! I know someone who works for Intel, its through their employee pricing program! I can also get an i7 920 for $142...hyvonen - Sunday, January 31, 2010 - link
Intel's Employee Purchase Program is for employees only, and not for resale. If your friend gets caught selling the SSDs bought through EPP, he can get fired.Also, SSDs are out-of-stock at EPP right now, due to high channel demand.
Hector1 - Sunday, January 31, 2010 - link
Change that "can get fired" to "will be fired". For each purchase, employees acknowledge its for their own personal use only. Companies are pretty strict about this. Not smart to risk your job over saving a few bucks for a "friend".Friends don't let friends risk their jobs.
mrd0 - Saturday, January 30, 2010 - link
Where do you see those prices?kunedog - Monday, February 1, 2010 - link
"Where do you see those prices?"Good question. Even Anand's initial predictions were higher than that:
http://www.anandtech.com/storage/showdoc.aspx?i=36...">http://www.anandtech.com/storage/showdoc.aspx?i=36...
And even most of *those* prices were never seen again.
notty22 - Saturday, January 30, 2010 - link
I think maybe he was being sarcastic?The logic vexed in the article mentioned twice the capacity at the same production cost. This would be 'wonderful' if this formula funneled down to the consumer. Kingston announced the 40gv now for 85.00 back in October ,09. that never was really available. Now its 30g for 90.00, supposedly coming ? Could the consumer get a 30g drive for 45.00 when this new tech is in full operation ? And with 230/170 r/w speeds with the new controllers. I doubt it, when the WD velociraptor 150g is still a 160.00. Maybe industry wide competition could make it happen.
RU482 - Saturday, January 30, 2010 - link
well, they didn't "announce" the 40GB at $85....that's where it wound up with rebates + promo pricing. You always have to keep in mind that there is overhead for the controller IC, casing, marketing, manufacturing, ect. So prices will not scale 2:1..All that said, I'd fricken love to find a 32GB SSD for $45...hell, even $75. 1.8" would be even better!
RU482 - Saturday, January 30, 2010 - link
I call shens. I'm in the process of quoting 5k pcs of the 80gb drive, and have not yet broke the $200 barriervol7ron - Saturday, January 30, 2010 - link
I agree.I think he might see a 40GB for $150 and an 80GB for $300. Market prices are just under those.
CurseTheSky - Saturday, January 30, 2010 - link
For $300, I would grab the 160GB G2 in a heartbeat. I paid $450 for mine and it's worth every penny.Yes, newer technology is going to come out that will make these feel slow, small, and generally outdated. For the most part, I don't play the waiting game unless something is right around the corner (for example, I wouldn't advise anyone to make a major graphics card purchase right now until we see what happens in March).
Griswold - Sunday, January 31, 2010 - link
I dont think you'll see SSDs anytime soon (read: years) that make current (intel) SSDs feel slow unless you like to play benchmarks all day long or are exclusiveley interested in sequential read/write performance.wumpus - Sunday, January 31, 2010 - link
You will almost certainly find an SSD that will make current SSDs feel *small*. New SSDs could easily feel faster than non-TRIM current SSDs. You might need a later motherboard with some yet-to-be-speced SATA3 system to make a state-of-the-art (1/31/2010) seem slow, and then only being used for virtual memory swap.mcnabney - Sunday, January 31, 2010 - link
Why wait?You should buy products when you need them. Current guesses on Fermi is that it should be a bit faster than a 5870, but it is going to cost a ton. There will also not be anything until mid-summer to fill the needs of consumers that like to game, but don't have a trust fund. And soon after Fermi the new generation of AMD/ATI will be rolling out (5XXX was a die shrink and DX11 only, not a new design). Just buy when you need it and stop trying to time the market.
dagamer34 - Saturday, January 30, 2010 - link
As exciting as this is, we need more manufacturers to develop similar technology so that prices can get cheaper!vol7ron - Saturday, January 30, 2010 - link
Anand,Could you discuss a little more about the ONFi? Also, you said the interface speeds can reach 200MB/s, how does this compare to the previous speeds?
Thank you,
vol7ron
murphyslabrat - Saturday, January 30, 2010 - link
Yessir, I know you think this every time a new flash technique is developed, but I can't wait for Christmas.jcollett - Wednesday, May 5, 2010 - link
Anand was right about the smaller drives in that article back in 2009. The supposed price drops for the OCZ Colossus line did not materialize with current low street prices being about 50% higher than MSRP shown in the article. Too bad ....iwod - Thursday, February 17, 2011 - link
Most of the benchmarks are not even reaching the practical / theoretical limits of SATA 3Gbps speed. So why would a fatter pipe , SATA 6Gbps allow much higher speed with the same controller?And have you figured out why Kingston V+ is so fast on benchmarks yet? I was asking last time and no one provide an solid answer. Like you have mentioned, the 4 corners of SSD performance, Seq R/W, Random R/W. Kingston perform very well in Seq R/W and only the low end for Random R/W. However it was the fastest in multiple Anand / Synthetic benchmarks.
P.S - VR-Zone just posted Next Gen Intel SATA 6Gps SSD will be out on 1st of March. Will we get a preview before that?