Performance Comparisons – Client QD Weighted
These results attempt to simplify things by focusing on what really matters – the Queue Depths that folks actually see when using these products. A dimension is eliminated from the previous charts by applying a weighted average to those results. The weights were derived from trace recordings of moderate to heavy workloads, which still ended up running closer to QD=1-2 even on a slower SATA SSD. The intent here is to distill the results into something for those wanting 'just the facts' to grab and go when making their purchasing decisions. Don't be alarmed by the low figures. Remember, these are low queue depths – the place where these SSDs actually operate when in use by those not just running benchmarks all day!
Again, all Optane parts turn in outstanding random read performance at the queue depths that matter, leaving anything with NAND in the dust. 800P also turns in some very good random write figures as well. Note the falloff there as we progress to lower die counts of the 32Gb and 16GB Optane Memory parts.
For sequential transfers, there's just no beating a solid PCIe 3.0 x4 SSD like the Samsung 960 and the SSD 900P.
This % read sweep chart demonstrates the opposing performance characteristics of NAND and 3D XPoint media. Flash can do well with random writes but rapidly falls off as the workload shifts towards random reads. Optane *loves* random reads, and does reasonably well with random writes provided there are enough dies to spread the load across. It is worth noting that the 960 EVO (with the help of its SLC cache) was able to beat both 800P's in a pure random write workload, as seen on the left edge of the above chart.