NVMe_Controller_V3

"Today’s processing is very data dependent. This means that the processing core is no longer the bottleneck. With the addition of multi-core there is significant trade-off and reduction available cache. In the past, hit-ratios were used to measure the performance of the system and MIPS was the final throughput number. Video processing and multi-media applications have left these metrics in the dust. At the same time, they have also caused a lot of confusion in what the metrics should be. Solid State Devices or SSD have been considered to be a solution for power reduction.

SSD, by itself, does not offer any power reduction, per se. A 7500 rpm 2.5” drive actually offers better power numbers compared to the best SSD. The advantage is the reduced usage of the cores and other devices that consume are idle more often and consume lower power.

There is a profound difference in performance and power depending on the product architecture and design. Drives consume < 5% of the total power. So, even the most power-saver will increase battery life by 10 min. The difference is in what it does with the other devices.

This is where more effort at the architecture definition phase becomes very valuable. Architecture or System models provide a means of exploring the performance and power consumption of a system for a large variety of scenarios. These are simulation models built early in the design cycle and prior to the specification is formalized. Unlike hardware prototypes, the models can be easily refined based on the simulation results. Also, ad-hoc studies such as the addition of a processor or a new application be quickly studied, sometimes even of the software is not available. A variety of analysis that are not possible with hardware prototypes such as processor stall times, or the number of IO/s or the MB/s of data accessed from the SSD.

#powerreduction, #SSD, #NVMe, #multicore #powergating"