32-Layer vs. 48-Layer V-NAND
Since Samsung’s August 2015 announcement of their upcoming 48-layer V-NAND, we’ve seen it trickle into recent products like the SSD T3, where it enabled 2TB of capacity in a very small form factor. What we have not yet seen was that same flash introduced in a more common product that we could directly compare against the old. Today we are going to satisfy our (and your) curiosity by comparing a 1TB 850 EVO V1 (32-layer – V2) to a 1TB 850 EVO V2 (48-layer – V3).
While Samsung has produced three versions of their V-NAND (the first was 24-layer V1 and only available in one of an enterprise SSDs), there have only been two versions of the 850 EVO. Despite this, Samsung internally labels this new 850 EVO as a 'V3' product as they go by the flash revision in this particular case.
Samsung’s plan is to enable higher capacities with this new flash (think 4TB 850 EVO and PRO), they also intend to silently push that same flash down into the smaller capacities of those same lines. Samsung’s VP of Marketing assured me that they would not allow performance to drop due to higher per-die capacity, and we can confirm that in part with their decision to drop the 120GB 850 EVO during the switch to 48-layer in favor of a planar 750 EVO which can keep performance up. Smaller capacity SSDs work better with higher numbers of small capacity dies, and since 48-layer VNAND in TLC form comes in at 32GB per die, that would have meant only four 48-layer dies in a 120GB SSD.
Other companies have tried silently switching flash memory types on the same product line in the past, and it usually does not go well. Any drops in performance metrics for a product with the same model and spec sheet is never welcome in tech enthusiast circles, but such issues are rarely discovered since companies will typically only sample their products at their initial launch. On the flip side, Samsung appears extremely confident in their mid-line flash substitution as they have voluntarily offered to sample us a 1TB 48-layer 850 EVO for direct comparison to our older 1TB 32-layer 850 EVO. The older EVO we had here had not yet been through our test suite, so we will be comparing these two variations directly against each other starting from the same fresh out of the box and completely unwritten state. Every test will be run on both SSDs in the same exact sequence, and while we are only performing an abbreviated round of testing for these products, the important point is that I will be pulling out our Latency Percentile test for detailed performance evaluation at a few queue depths. Latency Percentile testing has proven itself far more consistent and less prone to data scatter than any other available benchmark, so we’ll be trusting it to give us the true detailed scoop on any performance differences between these two types of flash.
Read on for our comparison of the new and the old!
(I just referred to a 3D Flash part as 'old'. Time flies.)