Page 1:A Lot More SandForce
Page 2:How Can Seven SandForce-Based SSDs Differ?
Page 3:Test Setup And Firmware Notes
Page 4:What's Important: Steady-State Performance
Page 5:Benchmark Results: Storage Bench v1.0 And Real-World Analysis
Page 6:Benchmark Results: 4 KB Random Performance (Throughput)
Page 7:Benchmark Results: 4 KB Random Performance (Response Time)
Page 8:Benchmark Results: 128 KB Sequential Performance
Page 9:Sequential Performance Versus Transfer Size
Page 10:PCMark 7: Storage Suite
Page 11:Final Words
How Can Seven SandForce-Based SSDs Differ?
At first glance, there's very little physical difference between these 120 GB SandForce-based SSDs. And yet, the least- and most-expensive models are separated by $100!
When you dig deeper into these drives' internals, though, there are some significant differentiators. The SSDs employ a couple of NAND flash technologies, for example. They also incorporate firmware tweaks to optimize certain specifications.
Understanding Memory Technology
NAND flash manufacturers are tasked with addressing two issues: memory density and performance. The former gets handled almost automatically when lithography nodes advance. Obviously, smaller transistors let you put more memory cells into the same amount of space. Despite the fact that progressively smaller geometry complicates the issue of write endurance, reducing the number of program/erase cycles a given cell can tolerate, the recent shift from 3x nm to 25 nm flash is an absolutely necessary step in lowering the cost per gigabyte we all have to pay for solid-state drives.
But increased storage density is only part of the evolution happening behind the scenes. Each generational step of memory technology also sees better performance. This is masked, for the most part, by the fact that SSD controllers use multiple memory channels to aggregate throughput. But now that we're dealing with 6 Gb/s SATA drives theoretically able to move more than 500 MB/s, the data rate of each channel actually matters.
Most manufacturers are using the latest synchronous technology to facilitate the best possible performance. However, others are designing less expensive models employing older and more affordable asynchronous NAND. Take Intel's second-generation X25-M as an example. The drive employs modules compliant with the ONFi 1.0 standard operating at 50 MB/s per channel. A fully populated 10-channel controller offers enough aggregate bandwidth to more than saturate the SSD's 3 Gb/s controller. And even if Intel's proprietary controller were 6 Gb/s-capable, there's a good chance its 10 channels would be ample.
But the SandForce architecture we're evaluating today uses eight channels. And given the highest sequential numbers cited by each vendor, eight 50 MB/s channels simply wouldn't suffice in situations where SandForce's technology couldn't apply significant compression and deduplication to the data. That's why the highest-performing drives in this story use newer memory types. In this story, the fastest drives just so happen to employ Toshiba's synchronous Toggle Mode 1.0 DDR NAND. SSDs based on the second-gen synchronous ONFi standard comprise the next pack. And the least-expensive drives use asynchronous first-gen ONFi-based memory.
Although much is made of the Toggle Mode (from Samsung and Toshiba) and ONFi (Intel and Micron) interface standards, it's really only a format war, and not something end-users should have to sweat. The more telling indicator of performance is whether the memory is asynchronous or driven by a clock signal (synchronous). The following slide from Bob Pierce's 2010 Flash Memory Summit presentation helps illustrate the gains from using synchronous devices. Although Bob was talking about SLC NAND, the same gains are typical of MLC-based devices, too.
Of course, increased performance translates into higher cost. The memory itself costs more because NAND manufacturers have to synchronize it with a reference clock. Asynchronous NAND is less expensive, but you pay that penalty of lower performance.
Today's most popular SSD controllers from Intel, Marvell, and SandForce are agnostic in that they support synchronous and asynchronous memory. As you can see in the block diagram of the SF-2200 above, SandForce gives SSD vendors plenty of choice in the way they configure their drives, and that's really the most obvious way drives get set apart. These guys aren't just taking the same controller, dropping it onto a reference PCB and mating it to the least-expensive NAND available at the time.
We see vendors like Corsair and OCZ using the same controller, but then creating differentiated product lines by picking the right flash memory and tweaking their respective firmware. That's the story behind drives like the synchronous-based Vertex 3 and Force Series GT, which cost more, and the less expensive Agility 3/Force Series 3, which use asynchronous memory.
While SandForce-based drives look alike based on their spec sheets, performance understandably varies quite a bit.
|Model||S511||Force Series 3||Chronos Deluxe||Vertex 3||Agility 3||Solid 3||Wildfire|
|Number of NAND Devices||16||16||8||16||16||16||16|
|Dies Per NAND Device||1||1||2||1||1||1||1|
|Standard||ONFi 2.0||ONFi 1.0||Toggle Mode DDR||ONFi 2.0||ONFi 1.0||ONFi 1.0||Toggle Mode DDR|
- A Lot More SandForce
- How Can Seven SandForce-Based SSDs Differ?
- Test Setup And Firmware Notes
- What's Important: Steady-State Performance
- Benchmark Results: Storage Bench v1.0 And Real-World Analysis
- Benchmark Results: 4 KB Random Performance (Throughput)
- Benchmark Results: 4 KB Random Performance (Response Time)
- Benchmark Results: 128 KB Sequential Performance
- Sequential Performance Versus Transfer Size
- PCMark 7: Storage Suite
- Final Words