Benchmark tests
Picking these drives apart on technical merits, the CrystalDiskMark results while certainly high performance, don’t show the Intel SSD 510 coming out on top. With the exception of the 33MB/s lead in the 1GB sequential read, the 510 lags behind the Crucial M4. When it comes to tests using a 32 bit queue-depth with 4KB random operations, the M4 leaves the 510 in the dust delivering read/write throughputs of 168 and 141MB/s respectively, against Intel's 81 and 50MB/s performance.
CrystalDiskMark 3 Results
Data transfer speeds in Megabytes per Second (MB/s)
Longer bars are better
In the hope that I might see Intel’s claimed performance with another benchmark, I ran AS SSD but nothing I observed was even close to the claimed 500MB/s reads or 315MB/s writes. All is not lost though. Intel does indeed deliver on IOPS. 19,514 is near enough the 20K read IOPS and when it comes to writing, the 510 blasted past the expected 8K all the way up to and beyond 12K IOPS.
ATTO test results
In the interests of completeness, I put the 510 through its paces with ATTO Disk Benchmark, which did manage to produce some rather impressive read speeds of 400MB/s. Writes remained the same at around 210MB/s though.
Next page: Compare and contrast
COMMENTS
Not that important, really
Wear leveling has been made a big issue of, because people are worried about early drive death, but it's just not a complex or expensive process. SSDs remap every page (4 or 8k) from the OS LBA to flash blocks, so they're free to save your data wherever they'd like, and ensuring that blocks are written evenly is simple (compared to other background processes likely to run on an SSD.)
I've used and benchmarked plenty of SSDs, and blocks have failed plenty of times (According to SMART stats) and I've never noticed. That's because blocks have a limited write lifetime, and when a write fails, the drive can just silently re-write the data to another block, and your OS never knows the difference, except for maybe a slightly higher latency. Drives come with varying amounts of spare blocks for re-allocation and to retain a pool of clean blocks for writing, so this block death doesn't become a problem for a while.
The main issues to worry about on an SSD is what the firmware does to maintain a pool of erased blocks, and how they protect themselves against a power failure. If you've got no erased blocks available, then each write has to wait for a block to be recycled, which takes ~50x longer than a simple write to a fresh block. As the drive fills, this is what causes the performance drop that most people worry about. TRIM can help, but the drive still has to be intelligent about how it organizes the data on flash.
Since the SSD blocks are re-mapped, the drive has to store that table somewhere. Since it's expensive to write blocks to flash (Especially blocks that don't get re-mapped, IE necessarily the ones that store the mapping table) they generally store this table in DRAM and only commit it to flash periodically. That means when you lose power (A proper full shutdown is safe,) your drive has a fair chance of losing a few of your last writes, depending on the firmware. I'd personally not consider using a drive without power fail protection (Intel SSD 320, anything with a SandForce 15xx or 25xx controller - Look for pictures of a drive taken apart, banks of SMD capacitors are easy to spot.)
Unanswered questions about wear leveling
More important than the relatively small speed differences among SSDs is the quality of the wear leveling. It seems likely to me that this varies greatly, and that the variation could be sufficient to be relevant. I don't see an easy and optimal technique, so there must be various methods, with various properties. At these prices, I'd like to know more.
Why 80% ?
If your tests show the Intel is slower and more expensive than the opposition then how does it rate 80%?
I'm not complaining, I just don't understand.
