So, smaller the process, slower the SSD speed?

[techs]

Reading the reviews of the new 25nm process SSD's they say that 25nm is inherently slower than the 34nm process.

Is this the way it will continue to work as the processes get smaller and the size of drives increase? Or is it just the first implementation of the 25nm process that will show this decrease in speed?

 

[Yellowbeard]

Reading the reviews of the new 25nm process SSD's they say that 25nm is inherently slower than the 34nm process.

This to some extent but I think it will improve some as the process matures.

And, I don't study the finer points of the physics and limitations thereof regarding NAND storage. But, I don't think we'll see a lot smaller process than 25nm due to the nature of how NAND works. Eventually, we'll simply run out of room at the molecular level to have space to store data.

 

[LamTek]

I wonder how they're going to drop prices when manufacturing reaches physical limitations. Its why I'm jumping on an SSD now rather than in 2013 or so.

 

[Concillian]

NAND does not scale in the same way as CPU / GPU.
CPU and GPU get a little MHz and a bunch of extra transistors that can do more stuff. They keep the amount of silicon about the same and pack more onto it. NAND dies you want a specific amount of memory. They can't use the extra space for speed like a CPU or GPU can, it's purely shrinking the size of the features.

Look at DRAM as an example. 7 years ago PC3200 was the standard. 200 MHz and latency around 5 ns. Today we are at 800 MHz and latency around 5 ns. doubling of speed every 3 to 4 years, no real improvement in latency. That's the kind of improvement we should expect.

Also, what do you see when you compare a 2GB DIMM vs. a 4 GB DIMM...
4GB DIMMs, in general, are slower

It's easier to make fast low density memory than it is to make fast high density memory.

Unrealistic expectations are unrealistic. Initial SSD speed improvements were rapid because of immature firmware. Firmware has, for the most part, matured to a point of diminishing returns. Extra memory channels can buy some throughput, but that also costs money, expect that the SSD market will eventually be a little tiered in that regard, with product differentiation like the 320 vs. 510 coming to more vendors.

Aside from that, it's totally reasonable to expect speed improvements to plateau to about the same rate as it has been for DRAM. Expect little to no scaling on latency, and a slow, methodical speed improvement.

 

[wirednuts]

^i dont think youre thinking about it right. nand chips are going to be built on stacks of eachother next. it doesnt have to get that much faster per chip when you can just keep stacking more on there with more channels. but even so, i think this type of memory will soon enough be replaced by optical versions....

 

[HendrixFan]

^i dont think youre thinking about it right. nand chips are going to be built on stacks of eachother next. it doesnt have to get that much faster per chip when you can just keep stacking more on there with more channels. but even so, i think this type of memory will soon enough be replaced by optical versions....

This.

The biggest drop in speeds on the Vertex 2 drives switched to 25nm, for example, came from cutting the number of chips in half. All SSD drives basically do a form of RAID-0 with the chips, running them in parallel. This is why you will see speed differences with different sized drives that use the same controller.

 

[Dark Shroud]

Basically Intel is doing 25nm SSDs right while OCZ is not.

If Intel is able to ramp up their 25nm NAND production we should see some great prices by 3rd quater this year.

 

[LokutusofBorg]

I believe the NAND process is bumping up against physical limitations at the 25nm node that indicates it won't be going much smaller. I think they have one more node shrink they can do.

There are alternate technologies in the works, like Phase Change Memory. Otherwise, the developments in existing NAND will come down to DDR-type improvements and controller evolution.

 

[Dark Shroud]

Yeah I think the next shirnk from Intel is supposed to be 14nm but I kept getting conflicting stories.

My target is saving up for the Intel 300GB 320 drive. Unless I happen upon some crazy deal on the 600GB version.