SSD Write Performance

[justin4pack]

I am using 2 ssd's atm. 1 Patriot Torqx 2 (64G) and 1 Kingston HyperX (120G).

I am getting really bad write speeds for both. Read speeds seem to be right around what quoted but write is less then half what's quoted. Any obvious reason this would be the case? When disabling write caching it made it worse. The torqx is coming at ~80 seq and the Brand new Kingston is at ~170 Seq.

The torqx is Rated 230 Write and the Kingston at 510 write.. wtf

 

[Hellhammer]

What software are you using for benchmarking?

 

[greenhawk]

I am using 2 ssd's atm. 1 Patriot Torqx 2 (64G) and 1 Kingston HyperX (120G).


write is less then half what's quoted. Any obvious reason this would be the case?

The torqx is Rated 230 Write and the Kingston at 510 write.. wtf

a guess would be that the specs are "upto" which are for the largest unit in the range. The smaller units have less flash chips and so have less channels for writing with. Meaning the smallest in a range will have the lowest write speed, sometimes noticiably far from the "upto" spec.

I could not find a actual suitable review that comparied each version against each other, but for the torqx2, I did find this post from about 2 years ago

http://www.overclock.net/t/1190745/patriot-torqx-2-64gb/10#post_16072881

shows write speed for the 64GB version about 80MB/s.




Another issue with some SSD's using "upto" numbers is that the numbers could come from the drive compressing data on the fly. Real world data might be uncompressable or the test program is using uncompressable data, meaning you get raw performance and not the best case numbers.

 

[justin4pack]

It was with crystal mark. @ greenhawk, I guess that fine print always gets ya lol

 

[Hellhammer]

It was with crystal mark. @ greenhawk, I guess that fine print always gets ya lol

CrystalDiskMark uses incompressible data by default, which is the Achilles' Heel of SandForce as it relies on real-time data compression. Try benching with ATTO and you should see figures similar to the marketing ones.

 

[Emulex]

the write speeds need to be normalized - ie write continuosly random data for 5 minutes then run a 15 minute benchmark.

Many of the drives consumer-wise are "bursty" or as-it happens at that second, since who on earth would continuosly write to a consumer drive?