Interesting write behaviour from HyperX 3K 120GB

[nisio]

It gets the average read of 555MB/s as promised, yet the write will settle at ~350MB/s (510MB/s promised average). I'm using the latest BIOS, known good cable, SATA 3 (intel), and the same version of ATTO as advertised; needless to say I'm a bit stumped. Any ideas are appreciated

Here's a cap for disambiguation: http://imgur.com/7trrH

 

[=Wendy=]

Welcome to the forum.
The advertised specs will have been measured with the SSD connected as a spare drive, and empty. You have benchmarked yours as a system drive.

The advertised specs will have more than likely been measured on a Windows 7 PC, with Intel RST SATA drivers installed, with the SATA mode set to AHCI mode, and with "C" power saving states disabled in the system BIOS.

Check that your SATA is in AHCI mode, if not you can change it to AHCI mode, but before doing so you will need to apply the following "fix it".
http://support.microsoft.com/kb/922976

 

[nisio]

The advertised specs will have more than likely been measured on a Windows 7 PC, with Intel RST SATA drivers installed, with the SATA mode set to AHCI mode, and with "C" power saving states disabled in the system BIOS.
http://support.microsoft.com/kb/922976

Thanks

I did my due diligence in regards to the priors i.e. made sure AHCI was enabled along with a couple other whatnots.

I haven't looked into power saving states in the BIOS, or the percentage of performance degradation after filling the drive with X amount of data. Should I be expecting a ~21% drop in write performance even with TRIM et c. enabled?

 

[groberts101]

Should I be expecting a ~21% drop in write performance even with TRIM et c. enabled?

For a smaller/midsized Sandforce based drive?.. at least that much and sometimes even more if you fill it up beyond about 70% and test it with benchmarks. Random/incompressible benchmarks such as Crystal Diskmark(write 20 gigs at default 5 x 1000MB test size) and AS SSD(writes about 3 gigs each time) are the worst since they touch more physical flash without compression availability for that type of data.

They just work differently than other drives and have built in algorithms that purposely throttle them to preserve lifespan.

Only 2 ways to get out of those type of temp throttles too.

1. Leave the drive powered on(no S3 or deep sleeps) for extended periods of time(sometimes days of non-use) to let the built in "time counter vs writes" algorithm counteract the throttling effect.

2. Secure erase to instantly wipe the counters clean and then reinstall or reimage the OS back again.

Another way to help avoid throttling as quickly is to unallocate some additonal slack space for the controller to keep a larger fresh block reserve and/or allow a few overnight logoffs(make sure the system keeps power to the hard drive or GC won't function) to allow the drives counter to equalize things while GC and partial block consolidation does its magic. Good luck with it.

 

[nisio]

Interesting stuff... thanks for that.

Then again, should I really be seeing that kind of difference at ~30% usage? If that's the case I'll shut up and eat my brussel sprouts