Slashdot Mirror
Intel, Micron Boost Flash Memory Speed by Five Times
Lucas123 writes "IM Flash Technologies, a joint venture between Intel and Micron, announced they've been able to improve NAND memory and its circuitry in order to boost read/write speeds by five times their current ability. The new 8Gbit single-level cell, high-speed NAND chip will offer 200MB/sec read speeds and write speeds of up to 100MB/sec, which means faster data transfer between devices like solid-state drives and video cards. IM Flash Technologies plans to begin shipping the new chip later this year."
Will it finally make sense for USB 3 flash drives? ;)
This will be useful for solid state disk drives. Unfortunately USB is stuck at an effective throughput of about 28MB/s so it won't help for cheap external drives.
Excuse me, but please get off my Pennisetum Clandestinum, eh!
However if you consider all the other places flash is being used. It wil be a big boon. Now all we need is for MRAM to come online.
So what are the read/write cycles, how much will they cost, and when can I get 200GB of them all in a nice pretty box? Even 10GB would be good for a nice little web server. Near zero latency would mean slashdotting is reduced to network bandwidth.
Wtf does this have to do with video cards?
Wonder when AMD will buy a memory company or team up with them. It's pretty clear the knowledge of compressing a CPU down to size is being used in memory here.
As usual - the lifetime of a product also requires the consumers to buy a new hot version.
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
It could significantly increase the usefulness of suspend/resume at the OS level. The limits on writes is a headache, but it would be possible to treat flash devices as additional swap space, making it theoretically possible to have hot-swappable swap devices as per some rather ancient mainframes. (Virtual swap space can be larger than the physical space directly available to a machine.)
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
...the chips cost six times as much as normal ones.
Perhaps flash drives will not have the sheer storage volume of current magnetic disks in near future, but even if they can be used widely as a primary OS + Application installation drive, it will still benefit a great majority of people on desktop.
Servers (particularly the database which is a frequent bottleneck) and notebook computer of course will also reap the benefit.
This is great, but we realy miss filesystem for such big NAND.
Either we use FTL [1](flash translation layer) to put FAT, but that that's quite ugly (FAT is not aware of flash and not robust to power lost, FTL is optimized for FAT).
Either we put flash filesystem like jffs2 or yaffs2, but they will eat lot's of RAM and take lot's of time on such big flash.
I wonder what are the performance with a filesystem.
PS : there is logfs or ubifs that should be better flash fs, but there are not ready.
[1] BTW FTL is patented.
Flash has always been relatively fast (slowed by USB 1.1 "interface". Now maybe USB 3.0 is needed or PATA/SATA internal boxes.
One place this may really help is cameras. The shutter lag is still bad, and this might help.
Flash cards don't have a seek time at all... How could they? It doesn't have any moving parts.
Yes, I know fast, big ssds are coming and they'll be getting faster, but they'll be expensive for a while, but still slower than an iRAM and much slower than memory on a pcie16x slot, should such a thing ever happen.
Try do random access on the next USB-flash stick you have access too.
That's a USB issue, not a flash issue.
The reason is that USB does things in transactions, and has to schedule all the transactions with priority. This is because of USB's fundamental flaw - it requires the host to poll devices. So a host will poll interrupt devices first, then handle isochronous transfers (bandwidth and time dependent traffic). Leftover bandwidth is then allocated to control and then bulk traffic. A USB host can do this once every millisecond, but most OSes break it out into more coarse granularity to avoid overloading the CPU when doing USB transfers. 10ms is about average for Windows, Linux is around 4ms. Basically, Windows will schedule all traffic on 10ms boundaries, so every part of a transaction will take place every 10ms. (10ms is a nice number because it means Windows can do the scheduling every timer tick).
If you do a USB disk request (read block N), the USB Mass storage driver will make a transaction to read a block. It will then issue the request to Windows' USB stack, which then add it with all the requests. If there's sufficient bandwidth in the next 10ms frame, it'll add the request to that frame. In the meantime, it's handling the current frame. When the next frame goes through, it sends the request, and if your USB stick is fast enough (usually is, but hard disks, it isn't) it responds immediately. If your USB stick isn't fast enough, then it will accept the request and wait for Windows to poll it again to see if there's any data, at which point the data will be transferred.
It's not the flash memory doing the seek (in fact, every time you access it, you "seek" it, it's part of the normal behavior) it's USB.
Fucking Finally. Thank God.
SSD's have been the most overhyped and overpriced pieces of shit since day one. We can't go a few weeks without some magazine/online publication spewing the propaganda to people, and especially gamers, that SSD's are the Holy Grail of performance.
I felt like coming downstairs on Christmas day to see my parents brutally butchered, the dog raped, and my presents smashed right before the jerk leans down and tells me that there is no Santa Claus... and he killed the Tooth Fairy and shoved her up the Easter Bunny's ass.
The read speeds were great, but the write speeds were horrible. I could post quite a bit of independent tests showing the SSD's had serious problems when used in a RAID configuration and that due to the write speeds, were not actually a good solution to increase OS performance and responsiveness.
1st Gen SSD is a total ripoff, but I can't wait for this to be applied to the next generation of SSD's. Don't let me down this time guys! Please.
(begin obligatory esoteric Jab),, Hey!,, Steve Case!,, I bet you didn't expect to see something like this come from Boise,,,, did ya!! (End obligatory esoteric jab)
Thanks for the info, it was very helpfull. I have used CF cards on IDE, but never managed to have one that is fast enough to even get close to 10ms "seek" time, and recently I read some specs about a new 333x CF and it said quite clearly 8ms seek time...
I do understand why there shouldn't be any seek times on random access memory, but really there is a big difference between RAM and Flash. I mean AFAIK most flash drives are made from blocks of memory and I can understand if you would have to make the controller read that whole block to get the info I want.. That's my thought, it might be wrong...