cool

Ouch....

So their $30 per GB comment combined with the models listed produce the following:

80GB = $2400
320GB = $9600
640GB = $19,200

So I would assume they will make the MSRP approx $2500 per 80GB such that the 640GB model will be a nice even $20K

They should be targeting companies like EMC and NetApp with this as a Read/Write Cache for their SAN solutions.

Damn you guys..... after all the money we spent on Robson!

0.00000000000000000000987654321s

So, How many times can you write to it? Without fancy tricks, the FAT sectors are going to be trashed in a few hours/days.

I'll take two. And I won't share.

They should be aiming at the consumer market, if you consider the cost saved by not having mechanical drive failure with OEMs. If Dell decided to stick one in every new PC, they'd get more revenue faster. Because if they dont, someone else will. Ramdrives have been around forever, the writing on the wall for this has been around since the price of flash per GB began to plummet. Gamers will spend the cash to Raid these together and have a totally solid state PC. Star Trek Isolinear chips here we come...

Installing an OS, some resource heavy games, or both... that could be huge. I bet a lot of people could justify $600 for that, especially gamers who routinely spend as much on graphics cards.

how reliable is this product? let's admit anything fails at least once, how does it handle data recovery?

with those.. first because of speed (light speed please) and second because of the issue with NAND having a lifetime based on the read/write of them (low level format on an old HD and it's back to life - sometimes). Even if it was $10/Gb (compared to $.5/gb-$1/gb on HDs) it would be a nice addition to an enthusiast game rig..

VooDoo and AlienWare PCs would be faster than anything we have seen up to now cuz the bottleneck of performance is the HDD, even in raid..


Guys, get one and benchmark it please.. :D
I would love to see the loading time of games and winrar/lame/video editing times.. :)

so 80 * 30 is 2400$ i wish it was affordable on the consumer end. man the performance would be great! load times! what are loadtimes!

Hehe. I'd like to see some RAID-0 benchmarks. I can only imagine the insane performance of that. Keep it coming. The prices will drop in time. The days of the hard-disk bottleneck will be over.

Most most memory lasts for 1 million writes, read don't change the lifespan. So, is this flash memory any different? Because 1 million writes isn't that many on a busy SQL server. Cool stuff though, amazing bandwidth and access time.

OH yah baby.. This my friends is CHAZZEEPPPP..
I know it is a lot per GB.. BUT MAN is that CHAZZEEPP per I/O..
To give example. a Raid Array to support 1000 Exchange users, reqires about 36x 15k Drives.
And this is a low end config..
You could destroy a 36 drive array, with 1 of these hard drives with just I/O/

Last thing, 1.5 years ago, i requested a quote from Texas memmory systems, for their I/O array, that does 100k I/O per second. for 96GB it cost 120,000 so yah,, these are chazeep..

So what's the catch 22? You know...besides the $30 per G and taking up GPU space? This is like Blu-Ray vs Betamax with a weird twist of Ferrari vs Mini van....is it really true? Would it be further boosted if tailored to PCIe 2.0...?

This cannot start in the consumer market, because they won't sell enough drives (even going through Dell). Businesses don't care if flash based drive have great performance, but they do care if flash based drives save them money in the long run. This drive would/should consume far less electricity, with the benefit of producing far less heat. For a large company that has thousands upon thousands of HDD's (that have to be kept cool and have a constant supply of power) the same amount of these would have a far lower operating cost than a traditional HDD.

The reason companies like Fusion io sell to businesses with, server farms, first is because businesses not only can afford it, but they'll buy a whole lot more than the consumer market can. The best part of this is that should big companies start buying these (and lots of them) the price/GB will drop substantially.

Funny thing is, if you compare this product to current flash based HDD's (ones that use SATA connections), the $/GB isn't that much more, and the performance is far greater.

Maybe our systems would need less RAM because file paging would be so much quicker now. So, spend less on RAM and buy a 20 gig version ... if they ever make one.

Even at $5 per GB, it would still be crazy expensive for a small OS/Programs only drive. Say around 40gb-80gb, your still talking $200-$400 on a system that will still require an additional HD for media content.

So, How many times can you write to it? Without fancy tricks, the FAT sectors are going to be trashed in a few hours/days.

It doesn't act like a normal disk drive, and since we use wear leveling and garbage collection, it will depend on the type of flash in the particular card, but it's at least on the order of 100K-1M full writes of the device. (Keep in mind that the number flash manufactureres quote is writes per page, not a total number of writes per device.)

That and who would use FAT on a device like this, sheesh! ;-) (It's like eating caviar from a dog bowl.)

how reliable is this product? let's admit anything fails at least once, how does it handle data recovery?

Of course, things fail. One of the parts of hard drive construction that alot of people miss is that you have to have ECC because bits flip, etc. We have to have the same thing going on with flash, since bits can flip in flash as well. However, most of the catastrophic failures of a hard drive come from mechanical failures, which we don't have to deal with.

Answers for some questions...

Wear-out is not a problem - wear leveling and very strong error correction make it possible to write non-stop for years (don't try that with a disk). Any degedation is caught and mapped out - no catastropic failures.

RAID - multiple cards can be RAID'ed to aggregate capacity / performance or for reliability, but do take additional slots - (time to start getting rid of the old PCIx slots wasting space!)

Intel Robson - What can I say but sorry. A tiny cache with 1/100th the performance really doesn't help with one of these. But think of all the work that Microsoft put into Ready Boost / Boot for Vista that isn't needed either. All the heuristics to guess which pages are needed just lowers performance when random access is so low latency and just as fast as streaming access.

Hey. I'm just wondering. If this thing can indeed deliver 6-800mb/s how can it fit into 'any pcie slot' ? cause far as I know a single pcie lane 'only' has a bandwidth of 500mb/s - so unless the slot has 2 or more lanes it shouldn't be possible to archieve those numbers right? at least not with 1x pcie version 1

Every one has been talking for a while now how flash will replace solid state drives but noone has done anything so far. Well these guys did and did it amazingly well. This thing is awesome. Forget about bandwidth this thing has crazy IO capabilities. You would have to spend a small fortune or raid controllers and fast drives in order to be able to rum 8 streams of data transfer like the ones in the demo. So yeah this is an expensive drive but $30/GB is not that expensive when you consider the fact that this is tiny uses hardly any power and quite huge in capacity compared to what you can get for the same price in server storage. Oh and for those complaining that it's too much for a desktop ... well it's not for the desktop. Does anyone of you have a 15000rpm scsi drive in his/her computer? I highly doubt that.

What I would really like to see in this product is replaceable flash chips. Yeah those last for a long time and stuff but come on you do not want to replace the whole board for one bad chip.

Now, they have a break through product. Let them make the most money out of it, while crushing competition, and while paying for the research they'd done. Eventually price will drop so deep you won't imagine. And this will be a breakthrough for gaming too, not only in the sense of performance. Because of the mass adoption such a thing should (and will) have, 1 - it will reside on the motherboard circuit itself; 2 - instead (or parallel with) SATA, you'll be having extension possibilities for cards, but with the 800-1600 speed instead of 150-300; and then, mass adoption again making this possible: 3 - games will reach end users directly on such cards, with all the read only that will make them play forever but without the possibility of replicating them. And you know what this means. More money to game developers for (hopefully) better games.
Now, for the author: when for a follow up article on this subject, answering asked and unasked questions. If it is as big as it sounds, it deserves it.

It was clear to me that this was the way the technology was going. And I was very disappointed with early personal computing offerings.

You're right that the card would be bandwidth limited by a single PCI-E lane, but the card pictured is definately a x4 configuration. Regardless, it is indeed electrically compatible with any PCI-E slot, one lane or 16, though it'll be bus limited in a x1 slot and won't take advantage of more than 4 lanes in a x16 slot.

Of course, the sorts of motherboards this card will live in will most surely have the x4 slots for it, and even in a x1 slot its performance would be... well, ludacris.

Cmon, with the exceptionally cheap USB Disks of 4GB or 8GB, this was a matter of time. Those people have invested some money and want to get back some, relying on performance.

Unfortunately (for them), they are not the only guys with the neat idea, and we'll just have to wait until we see some other companies producing cheap silicon-based storage cards.

I just feel pity for WD and Seagate, mechanical drives no matter if the rotate at 10k, 15k or 18k, just can't compete with this technology.

Look at all the problems that Hard disk drives do have: Heat problems, mechanical failures, high consumption, noise, high access times (6-10ms, compared to zero), non linear performance...

Hail to silicon!

im really happy to see that at last someone realized that havin 4-8 core cpu's is not the only need for high performance computing. the harddisk tech has net been changed too much since i bought my first 40MB ide harddisk to my 386DX-40;)

I cannot see anything video wise. Do I need to install something from the site or is it Apple Software I need. The old format worked for me before. Ps, Loved it when the table got smashed !

Yeah I'll take (2) 5GB harddrives ;) haha. They mention that they are obviously tweaking it for OS use. I mean come on the true meaning of this kind of power is going to hit servers first as they are consistently doing the same calculations or processes over and over again. I do like to dream though, just imagine just pressing the power button and your operating system's welcome screen is right there lol. Nice product Fusio IO.

First.this things compares to SAN solution and not your average el-cheapo desktop hard disk. In that way,30$/Gb is cheap reagrding perfomance and power its using. It almost seems too good to be true. And when things sounds like that, I am usualy sceptic. WHat's the catch. If there is no catch, well I am in the line of buying dozens of 640Gb cards.

When you get like 80-120GB for like 300-400€, i'll take one!

Looks great and can't wait to be able to afford one! ^_^
Any plan for a laptop version?

Also, one thing I would love: a hardware read-only switch (like some USB keys or the good old floppy drive). No more worries of trojans, viruses or other rootkits. A reboot and back to normal. Changing files can be written on another device or sent on the net.

Correct, it needs PCIe x 4 to get full throughput. It'll run in a slot wired x1 that's at least x4 in physical size, but will only get sub 250MB/s.

David Flynn

Many of you seem to have forgotten (or never knew) how expensive disc storage used to be. In the early 90's I paid $750 for a 1.7GB hard drive.

$30 per GB will obviously drop in price over time. You'll soon look at your 500gb hard drive in the same way you look at a floppy disk today.

Sony want around US$500 for an 8GB SxS Express Card. Thats over twice that price. Way to go Sony.

80 GB @ $30 per GB would be $2,400.00.
320 GB @ $30 per GB would be $9,600.00.
640 GB @ $30 per GB would be $19,200.00.

High end gamer? It would be a very high end gamer with the game player being the income range of Bill Gates.

The 80GB model would be great in a Oracle box. Just put the temp and undo tablespaces there and BOOM - lightning fast performance! At 2400 - that's pretty cheap nowadays. Especially seeing as it would cost you that much to group enough drives together to get that level of performance. And I think that's just it. They are targetting performance use cases, not so much capacity.

Judging from the pictures, it looks like the card is a PCI Express x4 card. Knowing this and knowing each lane is good for 250MB/s in each direction (500MB/s combined), this should allow up to 1000MB/s transfer rate in each direction (up/down), or as some people might refer to it as 2000MB/s if combining the upstream and downstream max throughputs. So them saying it will fit in any PCI Express slot is a bit misleading. It's more like any PCI Express slot that is x4 or higher, unless they plan to make slower cards that work off of x1.

They should, make this card in this way - that system should see it as standard hdd controler with out need for drivers.
in this case you can install any operating system on it...
... and your swap file.... it should speed up your machine.

expensive, I think I'll go for hybrid hard drives for the time being, hopefully will we start seeing some 3.5" drives with a couple gigs of memory storage.

I just don't know? Car or 640GB iodrive?

I think this type of tech is just for the server market for the time being.

I think it's great but I don't need 80GB on my critical high access DB. I need 20GB. Of course, the real cost isn't the GB of storage but the custom controller chip.

And I'd end up with two. It would have to at least be mirrored. Or three if I got with a Raid 5.

I'd also like to see that in a half height form factor.

oneofmany,

Does anyone of you have a 15000rpm scsi drive in his/her computer? I highly doubt that.

In my GAMING rig I have 4 with an expensive hw raid controller, a setup that costed me around 1,5k$.

And I would trade that immediately for a Fusion of 80-120GB, when the cost will be under 2k$.

80 Gigs will be enough for the OS and apps (not data).
The price needs to come way down though. Closer to $3/GB. It will happen some day. - but by then I will need much more than 80g for windows 2012.

wow.

$750 for a 1.7 GB? Wow... Go back FURTHER. Try

March 1989, Western Digital 20 MEGABytes.

Cost? $900 bucks

Thats:

$52 per MEG
$52,000 per GIG!!!!!

But, got back just a few more years:

1984: 20 MEGAbytes for $3150

Check it:
$157.40 per MEG
$157,000 per GIG!

And a quote to go with it:

"The cost for 128 kilobytes of memory
will fall below U$100 in the near future."
-- Creative Computing magazine December 1981

Oh and let's not forget paying $800 for a 9600 Baud US Robotics DS HST modem and that was WITH a SysOp discount!

Congratulations guys. I hope you do well.

I wonder how fast World of Warcraft would boot on this?

Anything this revolutionary is going to have a serious premium attached to it for the first couple of years. The primary bottleneck for computing has been the I/O operations for quite some time. This effectively removes that bottleneck. I think $30/GB is very reasonable considering the technology this delivers. As with all things in technology, the price will drop significantly over time.

Cheers,
Jason

This looks like an impressive product.

A question for the iodrive folks, if they're still lurking: I understand using the use of wear-leveling to offset the uneven degradation of the NAND sectors. But is there any non-deterministic performance degradation associated with this over time? It seems that the combination of remapping and garbage-collection of wear-leveling might impact random-access performance at some point. I'm just wondering if that happens after a month, year, 3 years, or not at all in the reasonable expected lifespan of the product.

Guys,
Where can I buy an 80GB version - now. The time for discussion is over. GIMME ONE!
Sweeeeeeeeeeeeettttttt
want want want want want

$30/1GB isn't that bad considering *slower* flash devices are still around $10/1GB.

"You'll soon look at your 500gb hard drive in the same way you look at a floppy disk today." -Bob

While this is a nice thought, I still think spin drives will still have their place, since they are at about $0.25/1GB right now (or 120x cheaper for equal capacity), and should hold a steady parallel decrease in price for a long time.
What I would really like to see is new file-systems and RAID standards that are structured to take these different storage types into account. High IO meta-data, databases, small files, and cache would be optimized to move to the flash part of the RAID. Large, low IO, streamy data would move to the fast spinning disks (with more redundancy for fail-safety). Backups and snapshots would all move to the slower, less reliable disks, with overkill redundancy (because they are so darn throw-away cheap).

Tape and floppy are "dead" now because they never kept up with hard disk cost efficiency improvements. I suppose that could happen to any mechanical storage, but it will take a lot of cost drops for any SSD tech to reach cost/capacity parity.

With 4 GB you can do two things. First you can mirror RAM. Instead of shutting down dump all the contents of RAM onto the flash memory. When you resume your computer comes back in a few seconds even if you unplug the power to the computer.

Also you could use the flash memory as a RAM drive and have a cold boot that is very fast (under 10 seconds?). You can also put certain big programs in this RAM drive and these programs start instantly. Perhaps software could be written to turn the flash drive into a hard drive cache so that you never have to wait more than a few seconds to load any program.

This two areas are where you could get most of your bang for the buck. Having a 20 G or even an 80 G flash drive would make the computing experience even faster but would be very expensive.

Would be an IPOD ON FLASH MEMORY CRACK!!!
Drop this puppy down to $400, put it in an ipod
and watch the RIAA/MPAA crawl into the fetal position and cry like a little girl!

This is truly inspiring. AMD's fusion, Intel's...whatever....THEY HAVE NOTHING. This tech coming to the standard data center server and soon enough, I'm sure, to the desktop scene is astounding. This will surely be the biggest performance boost I believe the IT world has ever seen.

As someone has already mentioned, who needs PCIe??? Don't think that motherboard manufactures aren't drooling over the possibilities this opens the door for. This tech will come to the consumer market faster than we'll realize. Then it's going to be a whole new ballgame.

Who needs OS CDs and DVDs anymore.... Windows 20xx (pre-configured....basically) on a read only, properly sized, flash card. Simply install the card into (insert new slot with direct CPU access here) and boot your machine to desktop. Any configuration data gets saved to the motherboard array.

Can you say bye bye RAM if this is integrated into the mobo and given direct CPU access? (To the poster above, you will still need RAM if it will only be PCIe, as these drives have DMA (Direct Memory Access), not direct CPU access).

The possibilities are truly endless.

Good luck Mr. Flynn. You have certainly peaked the interest of the Daily Tech community. However, I do cheer your competition in this venture, since they will be the one's driving your prices right into the mainstream.

Impressive.

I like to picture this as sort of the "ultimate cache." I believe companies with high-volume, critical data centers are going to be a bit skeptical at first. After all, hot-swappable disk arrays are tried and proven. So, because the iopts/sec is so great on these, you could potentially use them as your primary units of storage, and have several streams constantly synchronizing the new data back to the old-school disk arrays as a backup.

"I wonder how fast World of Warcraft would boot on this?"

Relative to the price, not very fast at all. Most game map loads also require a lot of CPU power rather than storage bandwidth because they are compressed and designed for systems with slow, mechanical hard drives.

However, if you were booting thousands of copies WoW every minute, or more importantly, if you were running one of the databases on the backend of Blizzard's game servers, one of these would be amazing. Considering that one of these fits in one slot and delivers the performance of a RAID 5 array of a few dozen ultra-high performance SAS 15,000 RPM drives with a number of very expensive controllers.

I totally agree with you. It should be treated like a regular hard drive. I would try my favorite operating system; eComStation http://www.ecomstation.com It would be neat to see how fast it would boot up (blink it is up already :) ). I expect the cost will come down, eventually. It usually does. It might be awhile before I get a chance to try it. :(

fix your problem, I am not going to install firefox.

[Brian: Pls install the latest version of Flash and you should be able to see the video. -ed]

... because it's MUCH SLOWER THAN YOUR RAM.

Go and freakin' check the facts before you start talking out of your bottom part.

Yeah, sure. What a great way to use it a completely useless, idiotic way...

Why on Earth would put anybody a 800MB/s read storage into a USB2-connected iPod w/ QVGA screen?

Utterly clueless.

$30 per gig is very reasonable imo, 80 gig drive for $2400. They'll get a bunch of corporations buying it to trial... and within 6 months the price will drop by 50% to $1200, and a bunch more corporations and a few rich crazies will buy it... a few months later it'll drop again to $600 and volumes will start to seriously ramp.

By 2010 we might all have one to speed up "Windows 2010" boot times.

what a sucker... OS/2 rocks, but eCom sucks! I miss my OS/2

The blink and it's up. It's always the same, tech gone up, but the startups are always slow. The same point the guy taht posted before you about World of Warcraft applies to booting an OS

Seems hard to believe a reader of TG Daily wouldn't have FF installed.

And ya, this kind of advance is the kind of once every 3-4 year thing that will give us all super computer performance on the desktop. I sure appreciated the move where he put the drive in his shirt pocket. Sweet..

I guess the wait for I-RAM 2 won't be necessary, still... I-ram was using volatile DDR which is faster than flash.
If they came out with a pci-e interface version of I-ram 2 (instead of sata 1) using DDR2 and multiple channels.. it would be a few generations of performance over this thing. still, volatility aside we are going to see some neat stuff evolve if this works out, for people (pro.'s) that know what this thing can do it will be worth the price if they need multiple io'ps over regular raid abilities. this is 1,000 times more i'os than a "fast HD" I run dual 74gb raptors, for enthusiast gamers this has been typically the best performance/price combo for the time being.

I love the I-rams but if they don't kick it into gear soon they will lose market share in this the future of storage, and yes.. with drives using memory it will eliminate the need for ram, but raise the need for bigger l1 and l2 caches on die

This would be insane on any system. If only I had 20k to put towards this. Ill wait for sure obviously unless I win the lottery. Btw they were founded in June of 06 so if this thing is really as bada** as it looks they will be one of the quickest most successful startups this industry has ever seen. Good luck Fusionio

Recall that hard drives first cost $30 per Gig about six years ago, and nobody was bitching about them being too expensive back then. They were totally amazingly cheap.

Now, you're not going to use this to back up your porn collection -- your going to use it for your databases and transaction processing machines, where it will absolutely rule -- using disks for that has always hard as you were dominated by seek time.

Of course, Google keeps their whole database in RAM, and that's viable too.

Thad

http://www.tgdaily.com/index2....Itemid=118

WinXP and IE

got it from the 'print this page' icon.

Like where is the preorder button :) Getting 2-4 of those and ...wow...thats all i have to say

So...you think a few milliseconds is really going to make that much of a difference if one of these arrays could have a huge bandwidth interface directly with the CPU?? It's one less step that the system would have to worry about. No more HD>FSB>RAM>CPU. Also, by the time this would hit desktops, I'm SURE there will be speed improvements, but at the seek rates they are talking about, I don't really think you would see a terrible difference.

These guys are going to sell a lot to the NSA.

"Tape and floppy are "dead" now because they never kept up with hard disk cost efficiency improvements."

Heh. Um, no. Tape is certainly not dead. Current state of the art is LTO-4, which stores up to 1.6TB per cartridge and realistically runs at >150MB/sec continuous transfer. Single FC/SCSI/SAS disk drives can't sustain that kind of rate. Hence, why most tape drives back up from arrays.

I would include DLT-S4, but SDLT drives are extremely sensitive to heat and just not as reliable as LTO.

$30 a gig is actually quite nice. A couple of years ago, a 36GB solid state drive ran about $30K. Not your plain old flash.

For speeds approaching 1 GB/sec, it's like a drive made out of ram... except that you get persistence as well.

My only worry would be chip failure. With 640 GB of capacity, and such an aggressive transfer bandwidth, I'd expect to see flash errors pop up quite quickly, unless the device can silently swap out bad sectors with a reserve cache.

Either people keep forgetting this or they just don't know, but flash-based memory (rewriteable NAND memory) isn't very durable. You can write to a given section about 100,000 times before it fails. These drives may be ok in applications that require alot of reading, but they will not be ok for write-intensive use.

RE: Would be an IPOD ON FLASH MEMORY CRACK!!! Sep 28, 2007 10:07
Yeah, sure. What a great way to use it a completely useless, idiotic way...

Why on Earth would put anybody a 800MB/s read storage into a USB2-connected iPod w/ QVGA screen?

Utterly clueless.

__________________
Did I mention by the time they got the shrink down to the ipod form factor USB 3.0 would be the standard of 2009... dummy!
4.8gb of bandwidth! Yeah, boo hoo, only 600MB of
bandwidth but there could be a slightly crippled version just to max-out usb 3.0, but you probably could squeeze more like 10gb out of the optical cable interface (why stop @ 4.8?)

Compare what we have today the 1980s when system memory was counted in kilobytes for most of the 80's. Cutting edge NV-Flash can and SHOULD do more than be server fodder.. it should be made for the masses! If not ipods, than UMPC's, cell phones, etc.

I get so tired of the "flash memory has limited writes" comment. Disks fail as a whole ALL the time in our datacenter and they are 15k enterprise drives. A million writes with flash is fantastic. Since there is no rotational delay, you can write evenly without penalty. So, a million writes means on a 100 gig drive (or card) you can write 100 petabytes before the flash starts to fail while still having MANY times the performance. Sounds good to anyone in a datacenter.

The lifetime is 100,000 writes per sector (or maybe more, depending on the chips). If there are 160 million sectors, the total number of writes (assuming perfect wear-leveling) is 16 trillion writes. At a 100,000 writes per second, that is 800,000 sectors per second, or about 6 months. But I doubt many applications would be that intense, and it is perfectly conceivable that the expected life of a flash drive is many times that of a SCSI drive, measured in writes or in years.

Back in 1992 or thereabouts I paid nearly $175 for a 30MB (that's MEGAbytes) Seagate hard drive. As I recall it had about 30MS access time and that was considered reasonably fast at the time.

This is the SEXIEST THING i have EVER SEEN. I've been wondering why nobody has built such a thing, and HERE IT IS!!! ! !! My prayers have been answered!! ! !!

Obviously, there are advantages to this solid-state solution, but at $30/GB I would much rather get more RAM. First of all, DDR2-800 is more than 10 times faster, and value modules come at the same price. Most delays in my system are small enough for me not to notice, however, when the RAM fills up, and the paging file is used intensely, I can definitely see a huge performance drop.
At least for me, spending my money on more memory seems more logical at the moment. Getting more memory will increase performance more than getting a faster paging file.
As far as application startup time is concerned, the SuperFetch service in Vista will keep the common used apps in memory, so they won't need to be accessed of the HDD (assuming you have suffiecient memory).
Of course, of the price dropped to $2 or $3/GB, I would definitely get this drive.

Doesn't NAND Flash wear out?
ioMemory™ uses sophisticated error correction techniques and wear-leveling to minimize fatigue and extend the service life of the NAND components within the ioDrive™, which results in a service life of eight years compared to the five year service life of mechanical hard drives. When an ioDrive's™ ioMemory™ does finally start to fatigue, the result is a reduction in available free space - no data is lost and the system remains operational.

You know you're a geek when you are drooling over the thought of having one of these in your computer.

From Fusion io's official site:

KEY APPLICATIONS
• Web Servers & Web Services
• Data Acquisition & Warehousing
• Simulation & Visualization

There's nowhere "home users". Although, in the FAQ section it says the product will be available to consumers market through some resellers.

Also, from the FAQ:
"ioMemory™ uses sophisticated error correction techniques and wear-leveling to minimize fatigue and extend the service life of the NAND components within the ioDrive™, which results in a service life of eight years compared to the five year service life of mechanical hard drives. When an ioDrive's™ ioMemory™ does finally start to fatigue, the result is a reduction in available free space - no data is lost and the system remains operational."

And, finally, yes data storage had times that 1 GB costed even more than 30$. But you cold buy a 40MB hard drive. This product is offered at 80GB minimum, so it's not the same.

If this product find its way in the corporations and sales go nice, I also believe that in a few years we may see smaller (and maybe slightly slower) solutions for the cosumers, at around 200-500$.

I'd really like to see some tests on solaris or freebsd with ZFS - that filesystem, with its built-in raid. As it uses copy-on-write, it's good
on wear as well.

This team could fly...

This is a bleeding edge technology with a bleeding edge price. As the tech develops and the price plummets, we will see the HD go the way of the slide rule.

Jesus, stop comparing current products with things of the past, unless you have a point to make!

In the '80s, I bought an 4.77MHz Amstrad CPC with 64KB of RAM, is there any Spectrum users, or any ENIAC operators who can beat that? :P

And just in case you haven't noticed, the bottleneck of current PCs, is not the processor(s), it's not the RAM, yep, it's that bloody mechanical noisy thing, that consumes 8ms of your time, every time it tries to access a distant sector! :D

Would someone please check the following math to see if I have this right? Based on what I read here, plus a few assumptions, it seems that this ioDrive has far more write cycles than a hard disk drive.

If we presume that one of these ioDrives has 100,000 (1e5) writes per byte, and the drive has 320GB (3.2E11)capacity, then the number of writes per drive (in bytes) is 3.2E16. If we presume that they will speed this thing up in the future, and obtain a write rate of 1 GB/s (1e9; this assumption reduces the effective life), then we come up with 3.2e7 seconds or 1.01 years before wear out. If they double the capacity, then 2.02 years. If half the rate, then 2.02 years. Etc.

But, let's use the same formulas, reversed, with a hypothetical hard disk drive. Let us assume that the drive can be hammered continuously with random writes for 10 years without before failing (is that sufficiently optimistic?). This would be 3.16e8 seconds. Presuming that this hypothetical HDD could do 100 MB/s random write (yea, right!), then we would have 3.16e16 writes per drive (in bytes). If this HDD had 320GB in capacity, then the number of writes cycles per byte would be 9.86e4, which is effectively the same as the 100,000 write cycles for the flash.

Now, of course, I don't believe that said hypothetical HDD is even possible to build, and any existing drive would have less bandwidth (say less than 5 MB per second random write), shorter life (say 2.5 years being constantly hammered, on the average), yielding slightly more than 1,000 write cycles per byte.

So, if this is all correct (and yes, this presumes that wear leveling is being done; anyone smart enough to build this ioDrive is smart enough to take care of that), this thing has WAY more life in it than an HDD, and only gets better with increased capacity. So, what am I missing? Or, are the comments about wear out simply FUD being put out by HDD manufacturer marketing weasels?

$30 a gig seems unreasonable given the cost per gig of normal HD, I don't see why cost per gig is even referenced, because its not really built for that, which is talked about repeatedly through the presentation. Its about performance, we know about capacity, we are all buzzwordage on capacity, how many library of congresses it happens to hold yada yada yada. But this is for performance which no mechanical moving part can really get close to. Its not RAM, its not meant to replace RAM, its bringing Drive storage to a better performance level, a step closer to RAM speeds, but without the volitility of RAM, I don't see this being useful so much for your average desktop user, or even your hardcore gamer even after the prices drop. Unless your seriously into dropping your pants and sizing yourself up against someone else, its meant for the server market. What do servers do? Access the same damn data over and over again, what does this do, increase the performance of accessing that same damn data over and over again, even if that data happens to be quite a few gigs in size. This is perfect for that specific purpose and for that, this just sings.

A bit expensive, isn't it?

I have two Gigabyte iRAMs in RAID0 for a total of 8GB of "Hard Drive" space @ 280MB/sec. They cost me: (2 iRAMs @ $135 / each = $270) ((8) 1GB PC3200 DDR DIMMs @ ~$50 /each = $400) = $670 total, or $83.75 / GB.

The Fusion io is almost 3 times faster, almost three times the storage space... AND it retains data even after the power goes out (the iRAM has a battery but only lasts about 8 hours).

I will seriously consider getting one for my development work!

Any one thats failing to realise why this thing cost to much should rip 8 DVD's to their hard drive and copy them to another place on their hard drive, come back and ten minutes. while this device did it in 5 SECONDS

would it be possible to pair up with a motherboard manufacturer and have 20GB (or more/less..) directly wired to the motherboard for the OS primary files, i mean you could make that even faster than PCI-e, and then you really could have no loading times at all for the OS..
i really think that could be a viable market..

i-ram drive did this with "DDR RAM" like 2 years ago just on a smaller scale. You can get up to like 8gb for less than this and is still crazy fast.

Ram drive = shits weak

grate idea but at 19,200 bucks for 640gigs this isn't happening for me or my company.

Doesn't NAND Flash wear out?
ioMemory™ uses sophisticated error correction techniques and wear-leveling to minimize fatigue and extend the service life of the NAND components within the ioDrive™, which results in a service life of eight years compared to the five year service life of mechanical hard drives. When an ioDrive's™ ioMemory™ does finally start to fatigue, the result is a reduction in available free space - no data is lost and the system remains operational.

You can't boot from ram... FTW.

I think that 600 and 800 Mb/s is running under linux, but windows can't manipulate such volume of information. Any modern processor can, but no window$.

Because they don't use the SATA controller, they plug in using PCI-E. You cannot run these in RAID any more than you can run 2 GPUs in raid, or 2 HDD in SLI. There will, no doubt, be mechanisms in mobo's if this catches on, but at the moment, you can have 2, you can have 3, but in no way can you run them in co-operation to make them go faster.

That drive is lightning fast compare to sustain transfer rate of a scsi hard drive which is max out at 100MB, but are those number sustain transfer rate or just burst rates?

Nobody mentioned it yet, but it will be dead-silent too. Only an artifact compared to its other properties, but nevertheless very welcome too :-)

In the end...

The world will use a open source OS and it will most likely be Linux.

No one mentioned Video Cameras. I think in 10 years we will all similar technology in our camcorders only with much better size/storage ratio's. I would love to shoot 5 hours of HD video and then transfer it to my computer in under a minute!

The video editing capabilities of this technology have been overlooked. I know many of you are SysAdmin's etc. but the video possibilities are endless for this.

Video Editing is some slow and cumbersome today and the quality is poor to. Currently to get the best quality video you still need to record onto tape. To get this on the computer you need to run the tape at real time and capture it onto the computer. The HDD camcorders use non-ideal compression and still offer relatively low transfer rates.

Having these in camcorders would also significantly improve battery life as well.

I enjoyed all of the comments tremendously!! This is very exciting.

Actually, another thought is a hybrid with IBM's nano-store technology (not the exact name), but a trillion bits per square inch is not to far away as well!!

That will be the video camera storage of the future!

to:Daniel Feenberg
you says that there is a 6 month life at continuous writes. I say you need to read the sector after you write to verify each write cycle, so that you know when a sector fails. which means 6 months of writing plus 6 months of reading, 12 months total.

Software RAID.

I don't understand the 10k ft max altitude limitation of this device?? Can anyone shed some light on why you think that is listed? This would be a great alternative to SSD for high altitude applications (Airborne..etc). Where building a pressurized encloser can be prohibitive for a list of reasons.

I wrote their tech department...but I haven't heard back.

If your OS (Windows XP for instance) see's them as disk drives...can't you use "dynamic disk" to create a software RAID?

Never underestimate the bandwidth of a station wagon full
of mag tape tooling down the freeway.

The price of such a popular chip should go down rapidly, my estimate based on other pop-tech devices; $30/GB 2007-9 then $15/GB 2009-12 then $5/GB 2012-14 with some 50% discounts from overshoot production - $2.5/GB 2014 or 40GB = $100 - - following a 50% over 2yr / 33% over 3yr / 50% over 2yr pattern - - but Harddrives should scale-up much cheaper still...

Thanks for the estimates! I think it will be much much faster though. Technology is screaming and there is always someone at your back. There will be competition in no time and it will force prices to drop very quickly.

I am not sure if any of you have seen the linux distro timeline (http://futurist.se/gldt/gldt76.png) but I would love to see a similar timeline for price per gig since the 80's.

Gamma rays (and other types of radiation) come in from space and those that get through the atmosphere cause "soft errors" when they pass through semi-conductor devices. Higher altitudes result in less atmospheric protection, to stop the rays from space, so that is the likely reason for the 10,000Ft. rated operational limit. Any unshielded RAM (or other types of solid-state memory) used for NASA projects is also subject to this soft-error problem.

FusionIo has stated an eight year service life for all their products. But that is under their as-yet unstated assumptions about duty cycle, average loads, and percent of writes versus reads. The worst case is maybe 10x shorter for a write-mostly application.

As Feenberg calculated, an 80GB unit could accumulate more than 100K erase cycles across all its NAND erase blocks in 6 months. (Read-after-write verification steps are not commonly done on Flash media.)

At that 100K point, up to 2% of all equally-worn blocks have already permanently failed to erase or rewrite, but those early failures have been transparently hidden by the 2% extra blocks that all NAND controllers must hold in reserve. By reserving more than 2% for spares, any well-designed NAND controller can go well beyond the 100K guideline specified by the NAND chip makers. But how much further one can go, depends on chips' bitcell voltage statistics that are not promised by the chip makers. This can be determined only by chip-level and drive-level testing. Most SSD vendors are not yet experienced enough to say, so they conservatively echo the chip makers' 100K component figure.

STEC has that testing experience, and they promise drive-level wear endurance that is 5x-10x the endurance of the parts they build from. Samsung tested 10 drives and found that all passed the 3 million cycle mark without losing data, but they do not yet promise that. Most NAND products built with M-System's TrueFFS firmware have quoted 1 million erase cycles per logical block at the filesystem level, when assembled from generic chips with a 100K spec for physical blocks.

The practical lifetime for IoDrive is likely several times longer than the (100Kerase * capacity / writerate) formula would suggest.

The service lifetime is longer for higher-capacity drives (because each chip and physical block sees less write wear). So a 640GB IoDrive would last 8 times as long as 80GB.

The lifetime is longer when write rates are capped by Sata links or by having few NAND chip interface channels, eg most SSD drives. FusionIo's very high write bandwidth will require them to specify their endurance limit more exactly than for their slow competitors.

Most SSDs other than IoDrive are very inefficient at small random writes, because they copy dozens of existing pages when writing one new page. This greatly shortens their service lifetime, when measured in random-write GBytes rather than years.

I don't think chip sockets or RAID would help with endurance limits. If one chip on a board has failed from wear, all the other equally-worn chips are similarly close to the point of ultimate failure.

Similarly, if one SSD volume in a RAID has failed due to erase-cycle wear, then all the other volumes have accumulated similar erase cycles and are similarly close to failure. There are no mechanical failures to survive. RAID can help with random bit errors beyond the drive's own ECC repair logic, and with other weird glitches. The benefit of doing that with inherently reliable SSDs hasn't been measured and reported by anyone yet.

They should change how the Operating system manage Fusion IO and fuse it with regular heavy read write duty medium to make hybird drives. Also make it cheaper. I dub android.. I mean RAID 16.

Store files that have Read only characteristics increase the read speed to Gigabytes/sec instead of 600mb/sec.
Then use regular raid 0 drive to store the heavy duty write files.
Or add i-ram III to boost the read write for gamers and have some O/S parts on it should help with consistant file changing.

This should give us benefit of insane reads/writes and longer life. Use like to store Media files, Program/ game libraries, O/S libraries files that won't get any updates often.

(1) Claim 1 : Disks today give 100 IOPS, is wrong. Its easy to get 200 IOPS. So claim #s are off by 2x.

(2) Claim 2: 1000 disks = 1 IODrive card is wrong. 1000 disks = 1TB * 1000 ~= 1Peta byte. The IODrive can store up to 320GB today.

(3) Claim 3: Throughput tests. Random and sequential access is very different. Its easy to get sequential throughput from a disk at around 90 MBps. So one needs about 9 disks to match an IODrive card (not 1000 disks).

Even if you look through the misleading marketing fluff, the basic performance on this device looks good. I am not sure what happens to this device if it is written to a lot. Disks are hot swappable, is this card hot-swappable?

based of the Data Sheet on Fusions site...

Price/gigabyte for networked storage averages $30-50/Gb, while consumer electronics enjoy approximately $.50/Gb.

Sounds like consumer side is gonna run about $40 for an 80gig drive.....

Marco VanBasten. It seems you dont know how raid works. You wont get 900MB/s read with 9 disks... only about 300MB/s and that is if you have large files. When it comes to small files like 4KB then the speed is much slower, around 5MB/s and still you cant get more than 200 IOPS. This is because raid doesn't double your hdd speed. 2 disks on raid 0 will give you about 50% increace and then 4 disks on raid 0 will give you about 25% increase compared to 2 disks on raid 0. And think about the access times... fast 15.7k hdd has an avarage 4.6-6ms access time. This drive has access time measured in nanoseconds, not microseconds because you don't need to spin the magnetic platter to find something.

Directly from the data sheets ...
ioMemory™ uses sophisticated error correction techniques and wear-leveling to minimize fatigue and extend the service life of the NAND components within the ioDrive™, which results in a service life of eight years compared to the five year service life of mechanical hard drives. When an ioDrive's™ ioMemory™ does finally start to fatigue, the result is a reduction in available free space - no data is lost and the system remains operational.

This ioDrive is one hack of a gadget... can´s wait until the technology is advanced enough to carry 640GB on a USB-thumb-drive around in my pocket...

The data sheet says
"When an IoMemory does finally start to fatigue, the result is a reduction in available free space - no data is lost and the system remains operational."

But this is only true for the NAND controller's pool of initially reserved physical space. It does not apply to a file system's pool of logical block addresses of deleted file chunks. The controller cannot subsequently expand its pool by borrowing from the file system's pool. It has no way to tell the various file systems that some logical block address and its former contents have now been made permanently inaccessible.

When retiring of worn blocks has eventually consumed all controller-reserved block capacity, the next block retirement must trigger permanent, visible drive problems. Either a persistent write error on some LBAs, or a moving pattern of write errors on many LBAs, or a catastrophic refusal to do any more write ops anywhere. This applies to all NAND SSDs.

Most SSDs act fine without complaint during their final months of using up all their spares, and then they fail catastrophically "without warning". Some SSDs now report their diminishing sparing capacity via product-specific S.M.A.R.T. attributes. I hope this becomes standardized.

It's certainly possible to go way back, to the original IBM hard drives. The IBM 305 RAMAC cost $160,000 in 1957 dollars, for 5mb. That's $32k/mb, or $32 MILLION dollars per GB, in 1957 dollars, which is probably like $400 million today. Personally my first was at my highschool- a Corvus 5mb drive with 8 controllers, that cost about $10k in 1980. That's $2k/mb, or $2 million/GB. So $30/GB is cheap historically speaking. But with today's cheapest disks going for $.20/gb, it's expensive. Why not make this thing into a physical hard drive controller too. It would be perfect with about 16gb of NAND, caching a couple 1 TB drives in RAID 0. It could be a lot cheaper then too.

Here we are talking about PCIE 1.0

With PCIE2.0 and more impoverishment in technology , i am sure we can replace Ram with this one. Seeems impossible ?

We can do it now;
Get a motherboard with DDR3 1800 Mhz Ram with least possible MB of ram (to save costs). AND set the SWAP Files or SWAP partition to this drive.

If we get this Drive just for SWAP Partitions, i think we will get a big boost in performance TODAY.

Cant wait for PCIE 2.0 versions.

OK Since the flash game is on the map now how about other companies like (NVIDIA,ATI/AMD,INTEL,OCZ)ect think about using flash in there products CPU GPU and MEMORY to boost the competition all the way around in which game programmers could really use the reality of physics in games that will blow all minds!
GAME TIME

Contract price for 1GB MLC NAND flash is only $5.50 now. For SLC it's only $12.50. For 2GB it's a bit more expensive per bit, $27 for 2GB. So their guideline of $30/GB is anywhere from 5.5x inflated to only 2x inflated. When they decide to charge a fair price for their base controller electronics, and not gouge on a per-GB basis, then it'll be worth talking about.

I've read a few comments like "can't wait for pcie 2.0 versions" what on earth are you going to do with 5GB/s bandwidth for a 800MB/s device??? use a hammer to fit 6 cards in the one slot?

For the gaming market we have IRAM, an affordable sata 1 implementation. Your product needs to copy the HyperOS idea, use a 5 1/4 inch form factor, a DVD drive case, and put your drive into that. But don't price yourselves out of the market like HyperOS. They market gamers but the cost is rediculus.

Why SATA2, well its got decent speed for consumers, up to 300MB/s, and everyone has them. For a Gamer who WILL have SLI/Crossfire, won't have a spare 16x PCIe slot and not likely have 4x slots either. Gamers do have 1x slots, but no RAID option with just 1 slot. With SATA 2 the gamer will likely buy 2 or 3 drives and raid them together.

Size, keep it useful, 4GB is enough but only just, 8GB is sweet. Price, in England £200 is good, but £250 is getting dangerously high.

Advantages of SATA2 = no drivers, everyone has one, easy to raid.

Hear about AuzenTech and their X-Fi clone? Over priced to Creative but still affordable to consumers, used better components and offered digital live encoding. Sold out in 2 months, awaiting a new batch!

1 more thing, contact Seagate and tell them to donate to your project to SLOW DOWN its development!

For 100,000 IOPs? Or 4 in parallel for 400,000 IOPs? $30/G is ridiculously cheap. This will change the ways enterprises think about databases and servers.

I paid $2000 in 1981 for a 20MB tandon hard drive, and in 1996 paid $600 for a 300MB ESDI hard drive. $2400 for primary storage is a decent price. Storage should be MOST of the cost of a computer, not the other way arround. I'll start saving my pennies now.

I know exactly what you mean! This company really needs to work on it's IT Department, with the way they developed it, it's almost like they are targeting SERVERS instead of GAMERS!

Hahaha, funny how they actually ARE targetting servers eh?
dont for the love of everythign good, become another clone of HyperOS, their solution is priced rediculously high.
last i heard these guys were releasing q1, next year. product does look interesting, and id honestly want a product with at least 80 gig ram myself, you dont use these things for swap files, you want your whole Operating system in it, if its vista. that means minimum 8 gig, so basically dedicated like 20gig to the OS, and the apps, then have 60 gig left over for moving games into. or if your like me and have multiOS setups, you might want something like a 160 gig drive ;)

Yes this drive is amazingly fast but it was built for companys that spend thousands of dollars on hds for critical components. these type of companys can afford this drive due to the ammount of money spent on keeping the hard drives working. these drives would really reduce heat levels greatly for servers. 30 dollars is reasonable. but in the end you have to ask your self one thing as a consumer, is this drive really worth 20k just so i can shave minutes off my time on the computer? o and you guys are complaining about the life span of the drives take into account that nothing is ever guarenteed to last a certain ammount of time, hds can go at any time. but what the heck thats just my two cents.

There hasn't been any news on the fusion-io site; does anybody know if they are still on schedule for Q1 2008?

[TG Daily has been in intermittent contact with the Fusion IO makers. Currently they are working on a revision to the previous card. Cards should be available any day now. -Rick]

Well, I've spent a time to see comments.
50% of readers doesn't get simple thing - you just don't need that thing. Because HDD is not main bottleneck for you. If you want to go faster - place more RAM. That's it. Because file cache created by OS is much faster than 100K IO/sec. And don't forget that RAM is minimum 5Gb/second. The only problem here is that we don't have desktop boards with more than 4 slots. That's limiting to 8GB.

Others who are talking about using it instead of RAM. What are you talking about? RAM is much faster. And well it's cheaper.

The only ones who will buy it is a users of really large databases. Because you can buy a server mobo with 24 slots and that's limiting to 96Gb of memory. That's not those 640Gb. And that costs a lot, because 4Gb ECC memory is not that cheap as 1GB non ECC for now. Also even if you have that a lot of memory and DB is fitting - you just need to spend a lot of money to make safe Flush. Because critical DB needs to hard-write data on every transaction. That's what limiting TPS.

So 30$/Gb is not something like 6$/Gb NAND 300$/controller, but for many companies that price will be acceptable and they need to make a profit. And in terms of corporate sector - 1800$ is not a price at all for making their sub 60Gb DB work close to in-memory DBs (but with safe commits).

We all would like to spend less and I hope that this thing will go and will see a competitors. That will move a prices. I believe something like 3-5$/Gb can be achievable within 18 months.

And last thing I want to say - SATA II will quickly become obsolete, because not only Fusion thinks about placing several NAND chips and make parallel access to each. China developers also knows about that ability. And when you have such nice possibility you can quickly overcome any bandwidth limit. The only limit will be your controller. Because it needs to be very fast. And if price on NAND will continue to drop - we'll see

Hell I'd buy one for my gaming rig. But gamers do need to see what they can upgrade first because its a lot cheaper to buy more ram right now(gamers don't usually sit on reg/ecc memory).

I would not be surprised to see these drives(40GB, 60GB or even a smaller special made one) as optional equipment in Alienware gaming rigs when drivers are ready...

@ whoever wrote this

The lifetime is 100,000 writes per sector (or maybe more, depending on the chips). If there are 160 million sectors, the total number of writes (assuming perfect wear-leveling) is 16 trillion writes. At a 100,000 writes per second, that is 800,000 sectors per second, or about 6 months.

isnt this complete rubbish?
If 1 sector fails after 100k writes, and we have 160million sectors, then guess what, we need 160million seconds to destroy all sectors!
Resulting in a freaking lifespan... of 1851 days - 5 years !
At full speed...