Storage system suppliers showcase modern options to maintain tempo with quicker accelerators.
MLCommons® introduced outcomes for its industry-standard MLPerf® Storage v1.0 benchmark suite, which is designed to measure the efficiency of storage programs for machine studying (ML) workloads in an architecture-neutral, consultant, and reproducible method. The outcomes present that as accelerator know-how has superior and datasets proceed to extend in dimension, ML system suppliers should be sure that their storage options sustain with the compute wants. It is a time of speedy change in ML programs, the place progress in a single know-how space drives new calls for in different areas. Excessive-performance AI coaching now requires storage programs which might be each large-scale and high-speed, lest entry to saved knowledge turns into the bottleneck in your complete system. With the v1.0 launch of MLPerf Storage benchmark outcomes, it’s clear that storage system suppliers are innovating to satisfy that problem.
Model 1.0 storage benchmark breaks new floor
The MLPerf Storage benchmark is the primary and solely open, clear benchmark to measure storage efficiency in a various set of ML coaching situations. It emulates the storage calls for throughout a number of situations and system configurations masking a variety of accelerators, fashions, and workloads. By simulating the accelerators’ “suppose time” the benchmark can generate correct storage patterns with out the necessity to run the precise coaching, making it extra accessible to all. The benchmark focuses the take a look at on a given storage system’s skill to maintain tempo, because it requires the simulated accelerators to take care of a required degree of utilization.
Three fashions are included within the benchmark to make sure numerous patterns of AI coaching are examined: 3D-UNet, Resnet50, and CosmoFlow. These workloads supply quite a lot of pattern sizes, starting from lots of of megabytes to lots of of kilobytes, in addition to wide-ranging simulated “suppose occasions” from a couple of milliseconds to some hundred milliseconds.
The benchmark emulates NVIDIA A100 and H100 fashions as representatives of the presently accessible accelerator applied sciences. The H100 accelerator reduces the per-batch computation time for the 3D-UNet workload by 76% in comparison with the sooner V100 accelerator within the v0.5 spherical, turning what was sometimes a bandwidth-sensitive workload into way more of a latency-sensitive workload.
As well as, MLPerf Storage v1.0 contains assist for distributed coaching. Distributed coaching is a crucial situation for the benchmark as a result of it represents a standard real-world observe for quicker coaching of fashions with massive datasets, and it presents particular challenges for a storage system not solely in delivering larger throughput but in addition in serving a number of coaching nodes concurrently.
V1.0 benchmark outcomes present efficiency enchancment in storage know-how for ML programs
The broad scope of workloads submitted to the benchmark mirror the big selection and variety of various storage programs and architectures. That is testomony to how necessary ML workloads are to all forms of storage options, and demonstrates the energetic innovation taking place on this house.
“The MLPerf Storage v1.0 outcomes exhibit a renewal in storage know-how design,” stated Oana Balmau, MLPerf Storage working group co-chair. “In the intervening time, there doesn’t seem like a consensus ‘better of breed’ technical structure for storage in ML programs: the submissions we acquired for the v1.0 benchmark took a variety of distinctive and inventive approaches to offering high-speed, high-scale storage.”
The ends in the distributed coaching situation present the fragile steadiness wanted between the variety of hosts, the variety of simulated accelerators per host, and the storage system with a purpose to serve all accelerators on the required utilization. Including extra nodes and accelerators to serve ever-larger coaching datasets will increase the throughput calls for. Distributed coaching provides one other twist, as a result of traditionally totally different applied sciences – with totally different throughputs and latencies – have been used for transferring knowledge inside a node and between nodes. The utmost variety of accelerators a single node can assist might not be restricted by the node’s personal {hardware} however as a substitute by the power to maneuver sufficient knowledge rapidly to that node in a distributed surroundings (as much as 2.7 GiB/s per emulated accelerator). Storage system architects now have few design tradeoffs accessible to them: the programs should be high-throughput and low-latency, to maintain a large-scale AI coaching system operating at peak load.
“As we anticipated, the brand new, quicker accelerator {hardware} considerably raised the bar for storage, making it clear that storage entry efficiency has turn out to be a gating issue for total coaching pace,” stated Curtis Anderson, MLPerf Storage working group co-chair. “To forestall costly accelerators from sitting idle, system architects are transferring to the quickest storage they’ll procure – and storage suppliers are innovating in response.”
MLPerf Storage v1.0
The MLPerf Storage benchmark was created via a collaborative engineering course of throughout greater than a dozen main storage resolution suppliers and tutorial analysis teams. The open-source and peer-reviewed benchmark suite provides a degree enjoying subject for competitors that drives innovation, efficiency, and vitality effectivity for your complete {industry}. It additionally gives important technical info for patrons who’re procuring and tuning AI coaching programs.
The v1.0 benchmark outcomes, from a broad set of know-how suppliers, exhibit the {industry}’s recognition of the significance of high-performance storage options. MLPerf Storage v1.0 contains over 100 efficiency outcomes from 13 submitting organizations: DDN, Hammerspace, Hewlett Packard Enterprise, Huawei, IEIT SYSTEMS, Juicedata, Lightbits Labs, MangoBoost, Nutanix, Simplyblock, Volumez, WEKA, and YanRong Tech.
“We’re excited to see so many storage suppliers, each massive and small, take part within the first-of-its-kind v1.0 Storage benchmark,” stated David Kanter, Head of MLPerf at MLCommons. “It exhibits each that the {industry} is recognizing the necessity to hold innovating in storage applied sciences to maintain tempo with the remainder of the AI know-how stack, and likewise that the power to measure the efficiency of these applied sciences is important to the profitable deployment of ML coaching programs. As a trusted supplier of open, honest, and clear benchmarks, MLCommons ensures that know-how suppliers know the efficiency goal they should meet, and customers can procure and tune ML programs to maximise their utilization – and in the end their return on funding.”
View the Outcomes
To view the outcomes for MLPerf Storage v1.0, please go to the Storage benchmark outcomes.
Join the free insideAI Information e-newsletter.
Be part of us on Twitter: https://twitter.com/InsideBigData1
Be part of us on LinkedIn: https://www.linkedin.com/firm/insideainews/
Be part of us on Fb: https://www.fb.com/insideAINEWSNOW
