A Closer Look And Inital Performance Testing
The BP5e employs a 7mm form factor, so it fits in newer platforms that require the thinner z-height. We recently tested a full-sized notebook that weighs close to eight pounds. Inside, it took 7mm-tall SSDs, and wouldn't accommodate standard 9.5mm drives.
Inside, we find the standard PS3110-S10 layout for Toshiba flash. The 960GB model uses eight flash packages, each of which connects to the Phison S10 through a dedicated channel. Two DDR3 DRAM packages serve as page table buffers to increase random I/O performance.
Data Type Performance Comparison
As we've seen before, Phison's S10 controller gives up small-block throughput when it reads back incompressible data. Using 4KB blocks at a queue depth of one, you're left with one-fifth of the performance versus the same test run using compressible information.
Most of our tests do, in fact, use incompressible data, but your PC won't necessarily be the same. This approach was hotly debated during the SandForce era, before studies on the entropy levels of common file types were published. That research shows most data is compressed to some degree, though there are a few file types that aren't.
Native TLC Sequential Performance
OCZ's Trion 100, the company's first TLC-based SSD, uses Toshiba 19nm TLC and sustains a 64KB-block sequential write speed of 115 MB/s. The drive uses a controller very similar to Phison's S10. The BP5e with Toshiba 15nm TLC and newer firmware than the Trion 100 increases that number to roughly 240 MB/s in the same test. Switching to 128KB blocks results in a performance boost to just under 300 MB/s.
Toshiba's manufacturing node isn't what improves performance. Rather, Phison's powerful ECC engine (enabled by the quad-core processor) with 120-bit error in 2KB data correction allows the drive to write directly to the TLC flash. That might sound like the standard operating procedure, but it isn't on most TLC-based SSDs. To preserve data-write cycles and extend endurance, most other drives with TLC flash first write to the emulated SLC layer, and then send it to the TLC. This is referred to as folding the data.
In the extended write test pictured above, we see the SLC area soaking up incoming data like a sponge. When that's full, new incoming data goes directly to the TLC, rather than waiting on the SLC and a flush before accepting more. We've said before that, for TLC to be a successful mainstream technology, drive vendors would have to hide its native write performance. It appears that Phison and MyDigitalSSD are doing this with the BP5e.