Performance Comparisons – Sequential and Random
Starting again with sequential writes, we find the 800P's going slightly faster than the 860 PRO (a SATA SSD). This is a definite improvement over the Optane Memory parts, but still falls way short of a 960 EVO or a 900P.
Sequential reads were also improved with 800P, but still falls middle fo the road compared to competing parts. Do note that both 800P capacities come in just a hair above the 32GB Optane Memory part, so it looks like we will not be seeing any additional throughput out of this particular controller implementation.
NAND SSDs are surprisingly competitive at low QD random writes. This is for a few reasons. To explain better, let's review what happens when a typical NAND-flash SSD writes or reads:
- Writes: Host sends data to SSD. SSD receives data and acknowledges the IO. SSD then passes that data onto the flash for writing. All necessary metadata / FTL table updates take place.
- Reads: Host requests data from SSD. SSD controller looks up data location in FTL, addresses and reads data from the appropriate flash dies, and finally replies to the host with the data, completing the IO.
Something interesting comes out of this data. Note how the 32GB Optane Memory part hits saturation at QD2, while both 800P's reach 2x the amount at QD4. This suggests we might have a 4-channel controller at play.
Alright, now things get serious. Everything with 3D XPoint installed completely dominates the chart, especially at lower queue depths. Everything with NAND takes far longer to service incoming requests, resulting in significantly lower IOPS figures at the more common client queue depths.