How we test

Custom PC - - LABS TEST -

There are many facets of SSD per­for­mance that can be as­sessed. Our own bat­tery of tests com­prises syn­thetic bench­marks, trace-based stor­age work­loads in PCMark 7 and PCMark 8, boot time mea­sure­ments and Iome­ter’s I/O work­load gen­er­a­tor. Prior to test­ing, we is­sue an ATA Se­cure Erase com­mand to each drive us­ing the SSD’s soft­ware pack­age if avail­able, or with the Parted Magic ( www. part­ed­ Linux build if not. This pro­ce­dure erases all data and re­sets the SSD to fac­tory per­for­mance.

We first run the syn­thetic bench­mark Crys­talDiskMark 5.0.2 to give us a quick over­view of a drive’s peak se­quen­tial and ran­dom per­for­mance, at queue depths of one and 32 - the for­mer is the most rel­e­vant to reg­u­lar users, but the lat­ter pushes most drives to their true peak per­for­mance. You can eas­ily run the bench­mark your­self to com­pare your own PC’s stor­age per­for­mance against the drives on test. We use the de­fault set­tings, but set the 32-queue-depth ran­dom test to use four CPU threads to max­imise the load on NVMe drives. Each test is au­to­mat­i­cally run five times, and we re­port the av­er­age.

For real-world test­ing, we first use PC Mark 7’s Sec­ondary Stor­age bench­mark, which loops three times and av­er­ages it­self. It uses recorded SATA traces (the ex­act traf­fic over the SATA bus at the time of record­ing) to sim­u­late per­for­mance in seven dif­fer­ent ways, in­clud­ing Win­dows pro­grams, adding pic­tures and mu­sic, video edit­ing and gam­ing. It then gen­er­ates an over­all score based on the time taken to com­plete the tests. Next up is PCMark 8, which again uses traces, but with more mod­ern pro­grams. We’ve se­lected the Pho­to­shop Heavy, Battlefield 3 and Mi­crosoft Word tests, with the re­sults this time be­ing the time taken to com­plete the trace. All PCMark re­sults in­clude idle disk ac­tiv­ity time, just as you would see in re­al­world use.

Next, we move to Iome­ter. We gen­er­ate four 64-queue-depth, four-threaded work­load pat­terns (data­base, file server, work­sta­tion and web server) de­signed to sim­u­late ex­tremely heavy sus­tained use sce­nar­ios with dif­fer­ent file sizes and write- to-read ra­tios. We run each test one af­ter the other for five min­utes, each us­ing fully ran­dom data – eas­ily enough to stress a mod­ern SSD con­troller. The num­ber re­ported is the av­er­age IOPS (in­put/out­put oper­a­tions per sec­ond) of all four tests.

The last test times how long it takes to boot a clean Win­dows 10 64-bit in­stal­la­tion us­ing the freely avail­able BootRacer, which mea­sures boot times down to a thou­sandth of a sec­ond. We in­stall the chipset, graph­ics, USB and au­dio driv­ers and re­boot the sys­tem five times to al­low Win­dows to get its caching in or­der. We then take an av­er­age of five cold boot times, which in­volve restart­ing the sys­tem and a full reload­ing of the op­er­at­ing sys­tem with all nec­es­sary driv­ers and ser­vices. Next, we take an av­er­age of five fast star­tups, which is the de­fault be­hav­iour fol­low­ing a shut­down (dif­fer­ent to re­sum­ing from standby). Here, Win­dows has saved the ker­nel and loaded driv­ers into the hi­ber­na­tion file (hi­ber­fil.sys), which it then loads back into sys­tem mem­ory when you boot up – for this part, only the load­ing of the desk­top en­vi­ron­ment is timed.

All tests are per­formed on an MSI X99A God­like Gam­ing moth­er­board us­ing an In­tel Core i7-5960X and 16GB of G.Skill Rip­jawsV DDR4 RAM. All CPU power-saving fea­tures are dis­abled.

Our SSD test suite in­cludes Iome­ter, Crys­talDiskMark and BootRacer

Newspapers in English

Newspapers from UK

© PressReader. All rights reserved.