MLPerf Storage v1.0 Benchmark Measures Performance of Storage Systems
MLCommons has announced results for its MLPerf Storage v1.0 benchmark suite, which measures storage system performance for machine learning (ML) workloads in an architecture-neutral, reproducible manner. In other words, the company says, the benchmark measures how fast storage systems can supply data when a model is being trained.
The MLPerf Storage benchmark includes three models – 3D-UNet, Resnet50, and CosmoFlow – and emulates storage demands across various system configurations, including different accelerators and workloads.
According to the announcement, the results show that “as accelerator technology has advanced and datasets continue to increase in size, ML system providers must ensure that their storage solutions keep up with the compute needs.”
As MLCommons states, “storage system architects now have few design tradeoffs available to them: the systems must be high-throughput and low-latency, to keep a large-scale AI training system running at peak load.”
Read more at MLCommons.
Related content
-
IO500
The IO500 is a less well known but very useful HPC benchmark with innovative applications.
-
Fundamentals of I/O benchmarking
Admins often want to know how to measure the performance of a specific solution. Care is needed, however: Where there are benchmarks, there are pitfalls. -
Defining measures
IOPS is mentioned quite often when benchmarking and testing storage systems, but what does it really mean? We discuss and explain what an IOPS is and how to measure it. -
What is an IOPS Really?
IOPS is mentioned quite often when benchmarking and testing storage systems, but what does it really mean? We discuss and explain what an IOPS is, and how to measure it.
-
Realistic Ceph Performance Tuning Measures
Administrators regularly criticize Ceph performance tuning, because some elements can be tweaked effectively, while others are difficult or impossible to fine tune. We look at how to measure performance, compare actual requirements with what is technically possible, and determine realistic objectives.