Could an NVMe™ solid state drive (SSD) make coffee breaks less spontaneous? I can answer that with a workload example from the new Micron® 9300 series of high-performance SSDs, our fastest and largest-capacity flash drives for cloud and enterprise, which launched today.
In today’s Machine Learning (ML) process there are many times in the day when data scientists must iterate through the Extract, Transform, and Load (ETL) and Learning/Training processes. These two distinct processes are sequential in nature; you wait for one process to complete before starting the next. When running ML on legacy rotating media, this leaves many opportunities for coffee (or tea) breaks throughout the day.
So, how can the Micron 9300 SSD with NVMe help hard-working data scientists in getting to data insights and learning quicker?
Micron 9300 SSD –3.5 GB/second of Read and Write Bandwidth
The Micron® 9300 SSDs deliver up to 15.36TB of usable capacity, enabling more data to be cached from data lakes in NVMe storage near the CPU/GPU complex. And with 3.5 GB/second of bandwidth for both sequential reads and writes, the 9300 SSDs can parallelize the ingest (ETL load) and export (Learning) steps of the ML process. Because NVMe transfer speeds are 7X that of SATA SSDs*, deploying the 9300 SSD creates the opportunity to parallelize the sequential processes of ETL loading and Learning.
The answer then is yes. The 9300 SSD performance benefits when coupled with newer ML software / development stacks can speed time to data insights. But those coffee break opportunities become harder to find.
9300 Performance Pays
Engineered to accelerate massive data sets and analytics, the 9300 NVMe SSD running on PCIe brings more of your data closer to the CPUs and GPUs with streamlined NVMe protocols. With up to 850,000 random read and 310,000 random write IOPS performance, the new series of drives can perform nearly nine times faster than leading enterprise SATA SSDs, while being 35% more power efficient**. This acceleration can provide improved results for a variety of performance-critical workloads: training and caching in AI and machine learning; high-frequency trading; caching and database acceleration; OLTP; and performance-focused block & object stores.
The balanced 9300 read/write performance delivers faster time to data insights enabling AI/ML/DL businesses to achieve higher ROI on their infrastructure. And, like data scientists are discovering as ML algorithms become more sophisticated, iterating with accuracy and efficiency helps deepen relationships to add value.
On a future blog, we’ll look at combining 3D triple layer cell (TLC) NAND in the 9300 SSD with 3D quad level cell (QLC) NAND in the Micron 5210 SSD to deliver a tiered-storage solution that can accelerate Red Hat® CEPH to Hadoop® YARN workloads.
For More Information
Getting too much caffeine as a data scientist? See about parallelizing your ETL and Learning cycles with the tech briefs and more at micron.com/9300.
*Comparing bandwidth from datasheet specs; the SATA interface SSDs top out at 540MB/s while the 9300 tops out at 3.5GB/s, which is more than 7x.
**Comparing Micron 9300 PRO vs Micron 5200 SATA ECO (7.68TB) datasheet spec for random read performance (850K vs 95K), sequential read performance and power (3500MB/s @ 12W vs 540MB/s @ 2.5W).