Internals and Simple Tests

Internals:

Opening these up we see that the V2 (right) can achieve its 1TB capacity with half the number of packages as the V1 model (left). The V2 employs 4x 8-die packages (each die provides 32GB or 256Gbit of storage). Note: the Samsung T3 was able to fit 2TB into four packages but had to stack them 16-high.

ATTO:

Right out of the box I partitioned and ran ATTO on each:

850 EVO V2 (48-Layer VNAND):

850 EVO V1 (32-Layer VNAND):

Sequential and Random (during pre-condition run):

I found that the best way to compare sequential performance of the two models actually came about during pre-conditioning for the Latency Percentile testing (on the next page). I pre-conditioned these by applying the same workloads to both variants (plus a third 1TB 850 PRO 32-Layer for additional comparison). Simply monitoring the throughputs to the two EVOs revealed virtually identical results at all stages of the process. Below is one example:

850 EVO V2 (48-Layer VNAND):

850 EVO V1 (32-Layer VNAND):

It was very clear that any sequential data testing was not going to show any differences here, but the same applied during the random access phase of conditioning. Here’s what the two drives looked like while sustaining 4k random writes to the first 8GB:

850 EVO V2 (48-Layer VNAND):

850 EVO V1 (32-Layer VNAND):

One note about ‘average response time’ in Windows Task Manager – I’ve found it to be artificially inflated during high QD activity. In this case I was writing at QD=32 and the actual average response time was ~0.35ms. The read and write speed figures can be taken at face value though, and in this case we saw no more than a 1MB/s difference (0.29%) between both units, with the 48-Layer V-NAND coming in a hair faster than the older flash.

« PreviousNext »