Micron Expects 1GB/s SSDs Within the Year

[Rob Williams]

From our front-page news:

It's become common knowledge that upgrading your PC to "top-of-the-line" status right now will mean little six months from now. Faster components come out all the time, from CPUs to GPUs and even storage. The latter is one area where things are happening really fast, though, and proof of that is with SSD technology. We are constantly being bombarded with new releases that are not only larger in density, but faster and less expensive.

The fastest "consumer" SSD on the market right now (to my knowledge) is Intel's X25-M, which has a Read speed of 250MB/s and Write speed of 70MB/s. The company's own enterprise SSD, the X25-E, increases the Write to a staggering 170MB/s, which is all fine and good, but Micron believes that we'll be seeing SSDs within the year that will be able to withstand up to 1GB/s performance, four times what Intel's current SSDs are capable of. It's unknown whether that figure is just for Read performance, or both, but it's likely to be the former.

Tests that Micron themselves have been showing off have seen results into the 200,000 IOPS area, which is truly incredible in every regard (Intel's X25-E is 35,000). Using blocks ranging between 2KB and 2MB, the IOPS performance was still impressive, at ~160,000 IOPS. So the problem is, if we are going to be seeing performance like this within the next year, it's going to make purchasing an SSD today a difficult process. It will be nice once things settle down, but it's definitely fun to see the performance of these pushed to incredible heights. The future of storage is looking very good.

​

Klein added that Micron's SSD uses "multiple channels" and was built interleaving 64 NAND chips to achieve its high throughput. The SSD is also based on several technology advances announced by Micron this year, including its 34 nanometer NAND chip architecture announced in May and the RealSSD P200 series drives announced in August.

Source: Computerworld
​

 

[Kougar]

I wish the future was now.

Of course 1GB/s is faster than even fibre channel... nothing could use a drive that fast to it's full potential. Not to mention the next future planned SATA spec update only doubles 300MB/s to 600MB/s...

The article mentions a 16 channel flash drive. The X25-E uses 10 channels. I guess theoretically you can add flash chips (1 per hannel) as long as the controller can keep up, which only improves the performance...

 

[Rob Williams]

Of course 1GB/s is faster than even fibre channel... nothing could use a drive that fast to it's full potential. Not to mention the next future planned SATA spec update only doubles 300MB/s to 600MB/s...

The humorous thing is that it's really odd to use the word "only", because really, 600MB/s is an incredible jump. It's not as mind-bogglingly huge of a jump as say, USB 2.0 to USB 3.0, but it's far from being ho-hum. Truth is still though, we're hardly even scratching the surface of S-ATA 2.0, so I guess we're in a good spot.

Micron wasn't using a standard S-ATA though, they were using a PCI-E card that tapped into that bus for even faster performance (remember this?). I agree on the channel usage... it seems the more, the merrier. I'm wondering what kind of limitation there is there though... even 20 might be difficult to accomplish.

I think they're doing this all within a 2.5" frame as well, and if that's the case, then I think going beyond 16 channels is going to be very difficult (I think they cards inside can only hold 16 chips, but I could be wrong). I agree though, I wish the future was now, because it sure would be good to have these things common-place. Our computers would feel so much faster with these installed.

 

[Kougar]

Hehe, that was kind of my point. We weren't even remotely near the 300MB/s spec, then suddenly Intel launches X-25 and now we're pretty close to it. From what I've heard Intel plans price discounts and newer, faster models next Q1...

I don't know anything about how the channels "hook" into the controller, but the controller is the magic here. It'd need to be capable of interfacing with multiple drives. I'm guessing it acts like a RAID array in miniture, each memory channel with a chip is like an independant drive, and by in effect RAID 0'ing them you can start doubling the performance pretty quickly. This would mean the controller needs to have some complex logic handling, enough channels available, and the ability to translate typical NTFS data file structure into something designed for SSD's memory cell structure instead of perpendicular drive platters. If someone can create an open SSD file format standard that optimizes the file structure for memory chips instead of hard drive platters, the performance will only go up significantly.

More than anything I bet the limiting factor is PCB real estate, assuming there are no technical issues I'm not aware of, anything beyond 16 channels would require a PCB similar to a GPU in size, another reason I bet they used the PCIe slot.

 

[Rob Williams]

From what I've heard Intel plans price discounts and newer, faster models next Q1...

I don't know... I'm still not that excited, but it's all in thanks to the overall density. I love these Intel SSDs, and they are a blast to benchmark with, but the 80GB limit is so hard to deal with (for me, at least), so it kind of sucks the fun out of it. That's not Intel's fault... it's just how it is right now. I know I've mentioned it before, but on the benchmarking computer, I have a pretty normal collection of applications installed, and a few games, and I'm left with 15GB free. That's just not too much. Even a drive twice this size would make an immense difference.

As for the performance, I think you might be right... some sort of RAID would have to be going on. I'm not quite sure if that's what the PCI-E cards use, but it would have to be what the S-ATA would have to use. Although, I'm a little confused, because I used to think that 3.0GB/s operation was a limit applied across the entire bus, but that's not the case. I remember talking to one of Intel's benchmarking gurus at IDF, right before the launch of the X25-M, and he had four of them in RAID 0 and saw speeds close to 1GB/s Read. Aside from that though, I'm not really keen on RAID being required for such speeds, because like RAID 0 today, once there's a fault, your data is gone.

That aside, one thing I'm afraid is going to have to happen is that the SSD enclosures themselves are going to get larger. We've been sticking to 2.5" for a while, and they're great, but it seems like with an SSD the same size as a standard hard drive, it would allow a LOT more density, and also room for various memory-controller-type devices that need to be included.

As for the special "SSD file format", there exists two such file systems in Linux that are currently in the works (they are usable now, but are actively being developed), but I don't think anything like that exists for Windows, yet (nor would it likely matter, since Vista installs as NTFS). The main benefit of these FS (JFFS2, LogFS) is to increase the lifespan, by writing to the least-used blocks. NTFS, for example, writes in a rather sequential manner (there's more to it than this, but too much to type), and on an SSD, it could mean that the same blocks are going to be written over and over again, which will heighten the chance that those blocks will become corrupt.

The Linux FS' will write randomly, so as to spread the data out and increase the overall lifespan of the drive. It would be nice to see SLC go way down in price though, because it's far more durable than MLC in terms of lifespan.

More than anything I bet the limiting factor is PCB real estate, assuming there are no technical issues I'm not aware of, anything beyond 16 channels would require a PCB similar to a GPU in size, another reason I bet they used the PCIe slot.

See, that's what's weird to me though. In the earlier SSD days (I mean, just last year), there were models that were identical in shape and size to a standard hard drive. I'd imagine that they could go back to that form-factor and just stack two drives on top of each other in RAID 0. I'm not sure though... I still don't like the idea of that, heh.

 

[Kougar]

When speaking about RAID I was trying to use it as an analogy, I'm sure the actual memory chips and channel controller don't RAID things, they just use a process and physical design using the same underlying principles.

After seeing the guts of a X25-E I don't believe there would be room in a standard 3.5" drive for 16 channel / 16 chip unit, not unless they utilized both sides of the PCB? However a GPU would be perfect since it would offer twice the surface area and everything could be kept on one side of the PCB to keep manufacturing costs down. And power could be drawn from the board to suppliment a power connector.

I fully agree with ya about the HDD sizes... as ya know usually with tech a smaller formfactor means less performance, or higher cost of manufacture, so I was wondering if a 3.5" SSD might be cheaper or faster than the current 2.5" models. But if I had to guess, I'd say Intel made 1.8" and 2.5" units specifically because they fit within laptops and super-high density server racks. 2.5" size 1500RPM Savvio drives were becoming very popular, but price-wise the Intel SSD's not only clean up shop but they offer significantly better IOPs at lower latency that that segment looks for above all else.

As far as 3Gb/s aka 300MB/s goes, that is the great thing about serial connections. They are dedicated point A to point B connections, not shared.

Bit of news about the drives, price cuts and larger model info: http://en.expreview.com/2008/12/02/intel-x25-e-64gb-ssd-delayed-until-q1-next-year.html