256GB Samsung SM951 and Native Command Queuing.

[crABtoad]

Ran into something strange today.

I'm running a 256GB SM951 on an ASUS Impact VII, 4790K.

Ran CrystalDiskMark 3.0.4 and received sequential reads at ~1600MB/s. This is in line with basically every review site that has a CDM benchmark for the drive.

Legit Reviews has a CDM 4 (beta) benchmark that shows the drive doing sequential reads at ~2400MB/s. Apparently this version of CDM supports Native Command Queuing, which I thought only applied to mechanical drives until today. I installed the beta version of the drive and my personal results jumped up to this level as well.

What do you all make of this? Is NCQ really that much faster, or is this a peculiarity with running beta benchmark software?

I haven't ran any other software benchmarks to corroborate as of yet, figured I'd drop a line in here first.

Cheers.

 

[toast0]

On spinning drives, command queuing helps because it lets the drive reduce the amount of seeking by doing work out of order. For an SSD, command queuing would probably help because it eliminates delay between filling one read and getting the next; there could potentially be benefit from combining adjacent reads or writes, but I don't know if the controllers do any of that.

 

[crABtoad]

Yeah, after looking through some stuff it seems likely that SSDs this fast are being bottlenecked host side, and need a local queue to make sure they have enough stuff to do. Still, 50% performance increase from flipping a switch seems like a big thing.

Is this something that needs to be enabled on a per-application basis? The fact that CDM 3.0.4 doesn't support it, and CDM 4 beta *does*, seems to indicate that this isn't much of an amazing feature if no-one is going to write applications to take advantage of it.

On a side note: Damn fast drive...like "woah" fast. Going to be an exciting next couple of years for secondary storage.