Why do Crystaldiskmark and Atto Benchmarks report different values?

[corrado33]

Testing a few flash drives that I own and I realized that crystal disk mark and atto disk benchmark report different values, even for 4k file transfers. A very quick google search says that crystal disk mark uses incomprehensible data where atto uses compressible. 

 

Does anyone know, in more detail, why these benchmarks differ? 

[dobo2001]

20 minutes ago, corrado33 said:

Testing a few flash drives that I own and I realized that crystal disk mark and atto disk benchmark report different values, even for 4k file transfers. A very quick google search says that crystal disk mark uses incomprehensible data where atto uses compressible. 

 

Does anyone know, in more detail, why these benchmarks differ? 

CrystalDiskMark tests using uncompressed data. This is why Sandforce controller SSD's usually struggle in this benchmark as Sandforce doesn't do well with uncompressed data. ATTO uses compressed data, which is what you see 90% of the time in your documents, music, and operating system. This kind is easier to transfer leading to sometimes higher results. Manufacturer speeds are usually based off of ATTO benchmarks.

 

That's pretty much the only difference. 

[Mira Yurizaki]

I've heard that some SSD controllers (I think SandForce was one of them) compress data before saving it in order to minimize writes. If you have compressible data, it could inflate the actual performance of the SSD.

[corrado33]

Thank both of you. I mostly saw differences in write speeds. Crystal disk mark had very slow speeds for some flash drives, while atto had fast speeds for the same drives.

 

Since I'm working on a "budget flash drive showdown" to post here on the forums, (Teaser: ADATA drives preform very well...) I think I may do a "real world" test as well and have a folder filled with a crap ton of text documents (compressible) and a folder filled with movie/mp3 files (incompressible) and time how long it takes to transfer over each folder using the command line. Possibly in linux. Or perhaps I could just create junk files with random data and with 0s to represent incompressible vs compressible data respectively.