Hard Drives And SSDs: Capacity Vs. Performance
Before we dive into SSD performance, you’re probably asking why performance improves as capacity goes up, right? After all, that’s certainly not the case with hard drives. In a mechanical drive, capacity is determined by the form-factor, number of platters, and areal density. Often, however, hard drives have a lower shipping capacity than what you would get if you added up the drive's relevant specs.
For example, a 160 GB Barracuda 7200.12 uses a single 500 GB platter. However, Seagate only uses the outer-most sectors, because this allows for higher performance. So whether you buy a dual-platter 1 TB drive or the 160 GB Barracuda 7200.12, you’re getting a maximum outside-diameter data rate of 125 MB/s.
Of course, SSDs are different. Instead of a spinning magnetic media, you have solid-state memory packages attached to a piece of controller logic. Those NAND-based devices communicate over multiple channels, and it's up to the SSD vendor to populate them in a way that yields the desired capacity, performance level, and cost.
Intel's own proprietary controller is a 10-channel design, for instance. In our example on the previous page, the 80 GB X25-M achieves 70 MB/s because all 10 of its channels are populated. The 40 GB X25-V employs the same controller, but a shift down to five channels correspondingly cuts write speed to 35 MB/s.
The Marvell 88SS9174 controller used by Crucial in its m4 SSDs is an eight-channel design. All of the m4s fully populate the controller's available channels, and yet there are still significant spec sheet-level differences between the four family members. This is because simply exploiting every channel isn't enough to saturate them. The number of packages residing on each channel matter. The number of memory dies in each package matter. The density of each die matters. And the firmware-level modifications a company like Crucial implements to help control performance scaling up and down the stack matter.
Crucial m4 | 64 GB | 128 GB | 256 GB | 512 GB |
---|---|---|---|---|
Channels Used | 8 | 8 | 8 | 8 |
Memory Packages | 8 | 16 | 16 | 16 |
Memory Packages Per Channel | 1 | 2 | 2 | 2 |
Die Density | 32 Gb | 32 Gb | 64 Gb | 64 Gb |
Dies Per Package | 2 | 2 | 2 | 4 |
Dies Per Channel | 1 | 4 | 4 | 8 |
As you can see, all four SSDs look quite similar from the top. We've clearly labeled a memory package on the 64 GB model, in case you're not familiar with the terminology in the chart above.
Within each one of those packages, you can have one, two, or four physical NAND flash dies. The shot above is an 8 GB (64 Gb) die manufactured on IMFT's 25 nm process. It measures about 167 square millimeters on its own.
Standardized interfaces like ONFi ensure that any compatible NAND device employs the same pin-out, a unified command set, improved data integrity, and a host of other benefits. When a controller vendor like Marvell or SandForce incorporates support for those interface standards, it allows drive partners to switch between NAND chips manufactured by several possible suppliers without worrying about some sort of compatibility issue. It just so happens that all four Crucial m4 drives employ 25 nm ONFi 2.2-compliant NAND manufactured by Micron (hardly a surprise there, right?).
See how the 128, 256, and 512 all have 16 NAND packages (eight up front and eight around back)? Similar though they look, they're not all the same. The 128 GB model has two 4 GB die per package, the 256 GB has two 8 GB dies per package, and the 512 GB has four 8 GB dies per package. That’s how SSD manufacturers end up with higher-capacity SSDs, even though the number of chips you see stays the same.