Well, it's been a fortnight since my original post to this thread, and I've been keeping my eyes peeled for additional Predator reviews, particularly ones with ATTO and AS SSD graphs such as those provided by Nathan Kirsch on pages 3 and 6 in his gracious review here:
http://www.legitreviews.com/480gb-hyper ... iew_160496. During that time, 4 more major tech sites have published online reviews which include an ATTO benchmark graph that we can look at -- again, for indications of how consistent the early test samples of the 480GB Predator are in showing a dramatic write transfer speed drop involving files between 64KB and 2048KB in size. One site, whose ATTO graph I didn't include in my original post (since it was a virtual duplicate of the benchmark response Legit Reviews' report included) has performed the ATTO test using 2 480GB Predator drives in RAID0; I'm including both here since this post is something more of a broad -- albeit very informal -- survey of the Predator's performance under ATTO than was my initial post. (The fifth site's review lacked the standard ATTO bar graph, and none showed AS SSD results in graphic form -- hence, my focus on ATTO alone.)
Here's the ATTO graph from the first published review of the Predator:
http://www.thessdreview.com/our-reviews ... w-480gb/2/ (at the bottom of the page); this review followed LR's:
http://www.bjorn3d.com/2015/03/480gb-hy ... 0p2480g/5/ (also at the bottom).
Next, came this one -- and when I saw it, my jaw dropped, and my heart leapt:
http://proclockers.com/reviews/storage/ ... page=0%2C3 (the first benchmark in the review, I might add)! Now
this was what I had hoped to see to begin with!
Such high consistency in data transfer rates across the range of file sizes is a thing of technical beauty!
Overall, it looks as smooth as Samsung's two M.2 PCIe drives'.
After that gem came a disappointing repeat of the previous ATTO graphs, but with a twist -- all ATTO benchmarks shown till now have been with the queue depth set at 4; this review shows the results with depths of 2, 4 and 10:
http://www.custompcreview.com/reviews/k ... w/23727/4/. Irregularities abound, but they vary with the change in queue depth.
This is the graph of a pair of Predators in RAID0 configuration:
http://www.thessdreview.com/our-reviews ... as-fast/2/. The extreme drops in speed seen in all but one ATTO run have been moderated significantly, but as the review's author noted, "... these results do not show perfect scaling... "
Last, I'm going to toss in a graph of the ATTO performance of Intel's new 750 NVMe PCIe 3.0 x4 SSD, in part, to illustrate the unique shape of the overall curve another high-performance PCIe SSD's ATTO benchmark takes:
http://www.thessdreview.com/our-reviews ... industry/3, but also to invite comparison of this unit's performance to the 2 Predators in RAID0.
I might add here that the Plextor M6e in both its variants has a quirky dip in its curve -- but in the
read area, instead:
http://www.legitreviews.com/plextor-m6e ... w_144812/3,
http://www.legitreviews.com/480gb-hyper ... w_160496/3 (second graph from top).
What a graphical single-benchmark comparison of the performance of these 7 SSDs means, I won't attempt to say, other than to comment that various types of NAND, coupled with different controller chips, combined with particular versions of their associated firmware, as implemented by individual manufacturers -- along with inevitable variance in the mass-production process, no matter how stringent the QA and exacting the QC -- leads to performance differences between various models of a type of product. (Duh!)
But to get back to the real reason I assembled these posts -- the inconsistencies shown in two important indicators of measurable performance of Kingston's 480GB HyperX Predator M.2 PCIe 2.0 x4 SSD have me puzzled. And now, with one test instance's graphic output showing a drive with practically none of the unevenness demonstrated by 5 other similar units under the same test, with the same settings, I could
really use input from my fellow LR members as to
what's going on! If what we're seeing were due to the need for further tweaking of the Predator's firmware, then why did the one sample -- at least, in that one instance -- show "normal" behaviour? Ar-r-rgh!
I just wanna know that if I buy one o' those things, it'll behave properly, and I won't be shredding my wig! ~X(