Performance Comparisons – Sequential and Random

Just as we did on the last page, I'll start us off with Sequentials:

comp-seq-r.png

comp-seq-w.png

Since the last review, I've tested *a lot* of SSDs on the new suite, but for these comparisons, I'm keeping the selection limited to a reasonable 8 devices. I've included mostly SATA parts, but also tossed in an Intel 600p to represent budget NVMe. Recall that all data points here are the result of cumulative sampling across multiple percentages of allocated space. If you see what looks like an outlier or a weird dip, rest assured that apparent aberration was repeatable and consistent.

Speaking to the above two charts, the majority of modern SATA products perform reasonably well and within a tight performance grouping. The Intel 600p is able to use its higher interface bandwidth to pull away from the SATA parts, but only at higher queue depths (which are not representative of actual client use queue depths).

Now for random:

comp-rnd-r.png

comp-rnd-w.png

While consumer usage rarely exceeds QD=8, we extend these charts out to QD=32 for SATA reviews (256 for PCIe / NVMe) to evaluate manufacturers claimed performance maximums. Our exclusive burst test is the only way to properly evaluate the random write performance of caching SSDs. Outliers in this test are the 600p's impressive random write performance (even at low QD), but that same NVMe SSD actually comes in behind most of the pack in random reads. The Western Digital Blue essentially did the inverse of the 600p, turning in relatively low random write performance while coming on very strong in random reads – even at low QD. Another interesting data point here was that the 500GB 750 EVO appears to have just slightly edged out the 850 EVO across nearly the entire QD sweep. We will know exactly where it fell at lower QD in our weighted numerical charts on the next page.

« PreviousNext »