Power Consumption and Thermals
Let's take a look at a couple of ATTO runs. First the SSD 750:
Now for the SM951 NVMe:
The SM951 sees a distinct advantage at smaller transfer sizes. Small file writes are far superior to the SSD 750, and read performance 'ramps up' far sooner than the Intel unit. Take note of how close those results are at the top end, especially the ultimate throughputs seen (the bottom few entries, which are large enough sizes to be more sequential than random). With those up for reference, I now direct you to the power consumption of both solutions as they turned in those above figures:
Wow! With very similar ultimate read and write performance, the SM951 is consuming roughly half the power of the SSD 750. It's almost a neat halving on just about everything from idle to random to larger transfers that behave more sequential in nature (the results at the right edge of the plot).
An additional note on idle consumption. The SM951 is rated at a much lower idle power figure (only 2mw), but that is only for the lower power state used on mobile platforms. This also negatively impacts response time as it takes the drive a bit more time to recover from this much lower power state. We chose to test as this SSD as it would be used in a default desktop configuration.
While it is far more power efficient, it doesn't have any heatsink area at all, and during sustained writes or heavy random IO, it will rise in temperature given the chance:
If you do manage to find one of these and put it into your desktop system, I'd recommend having at least *some* airflow passing across the M.2 slot. We saw no thermal throttling but with a fan at the slowest speed pointed towards the SM951, we recorded that 156F figure above. So long as we had minimal airflow across the M.2 slot, none the M.2 SSDs tested hit their thermal throttle limits, regardless of applied workload.