It looks like you're new here. If you want to get involved, click one of these buttons!
Intel recently launched the SSD DC S7300, an enterprise-focused SSD. The price tag is over $2/GB, which makes it vastly more expensive than consumer SSDs. It offers HET MLC NAND flash, for better write endurance.
But Intel was also keen to claim that it offered more consistent write performance than competitors. A 4 KB random write test might show 40K IOPS on average, or equivalently, taking 0.025 ms for an average write. But for most SSDs, some of those writes will take several milliseconds, or in some cases, even tens of milliseconds. Obviously, it can't happen very often if the average is only 0.025 ms. But it can still be disruptive.
To put this in video card terms, 50 frames per second is great if it's a new frame every 20 ms with no exceptions. It's not so good if it's only 50 frames per second on average, but once per minute, a frame takes an entire second. That one frame gets overwhelmed in the averages, but it's still really disruptive to gameplay.
For SSDs, this is especially critical if you're using RAID. Some people around here are big fans of RAIDing SSDs. I say it's a bad idea for consumer use, but there are good reasons to put SSDs in RAID. The trouble is that a RAID array has to run at the speed of the slowest device in the array. If an SSD takes 10 ms for one write in 10,000, but you've got four of them in RAID 0, then now at least one of the four will take 10 ms for one write in 2,500. More severe corner cases could actually make RAID 0 slower for real-world use than a single drive, even if the averages look just peachy.
What the SSD DC S7300 was able to offer is random write performance that never dipped below about 29K 4 KB writes per second. That sort of consistency means you can use them in RAID 0 without the hiccups in performance. And it means consistent performance for an end user, rather than occasionally having to wait 100 ms on a single write, as even some of the good SSDs will make you do now and then.
The SSD DC S7300 has 256 GB of physical NAND flash, but sets aside about 1/4 of it as spare area unavailable to the end user. Most consumer SSDs set aside about 7% (advertised as 64, 128, 256, or 512 GB drives) or 13% (advertised as 60, 120, 240, or 480 GB drives) of capacity. So Anandtech said, let's take the top end consumer SSDs and set aside 1/4 of the capacity as spare area and see what happens.
The Corsair Neutron (LAMD controller) went from a minimum of about 11k IOPS to 29k. The OCZ Vector (Indilinx Barefoot 3 controller) went from a minimum around 1k to 29k. The Samsung 840 Pro (newest Samsung controller) went from a minimum around 400 to 25k. All three were able to roughly match the consistency of the SSD DC S7300 by setting aside more spare area. And it's probably not a coincidence that the LAMD SSD did by far the best at the default amount of spare capacity, considering that it sets aside 13%, while the other two only force you to set aside 7%.
I'm still of the view that among good SSDs, performance differences don't matter much for consumer use. But if you want the highest end SSD performance you can get, more spare area is the way to go.