RAID Testing (VROC and RST)
For this review, I tested a set of four 118GB 800P's and compared them to a set of four 960 PRO 512GB parts (we didn't have four 960 EVO 250's). Arrays of 2-4 drives were tested on X299 VROC, while a pair were tested on Z270 (but we have enough knowledge of RST's performance to make educated guesses at 3-drive performance).
For sequential maximum throughput, we get an expected linear progression with the VROC configuration, while RST tapers off to a max throughput at ~3.5GB/s, limited by DMI 3.0.
Random performance ramps up similarly for all groups tested, with only the single 118GB running out of steam and leveling off early. Overall we can see that even though the 800P in a RAID configuration starts at a lower IOPS value, it still far exceeds that of the 960 PRO.
This chart is a segment of the data from the previous one. Focusing on QD1 and plotting additional SSDs across the axis (similar to the top chart on this page, but for IOPS instead of sequential throughput). Note that random performance does not scale higher when drives are added to the array because we are measuring at QD1 here. Can't scale higher when only a single request is coming in at a time.
Latency figures are about what we would expect. Note that even with a single 118GB 800P, we come nowhere near that oddball 6.75us rating we saw in the specs earlier. The realistic latency figures for the 800P fall in the high 11's. X299 VROC arrays adds ~5us per IO, while Z270 RST arrays add ~2.5us per IO.