Hyperion Research has released its 2021 Annual HPC ISC Market Update, and this year’s release included a 100+ slide study of the landscape presented by six analysts. We will not cover all 100+ slides. Bottom line: Despite the pandemic, the vast 2020 on-premise HPC market (servers, storage, etc.) grew slightly (~ 1.1%) to $ 27.75 billion, while cloud-based HPC surged (~ 10%) to $ 4.3. billion. Hyperion announces a slight decline next year before resuming growth.
“The year has actually been a positive one and it has been a positive one because the Fugaku machine, which was in the billion dollar range, actually changed the market from [being] down 7-8% to increase only slightly, ”said Earl Joseph, CEO of Hyperion. Besides Fugaku, which came online during the pandemic and has remained at the top of the Top500 list ever since, there was a high-end server force driven by sales of exascale systems and other large systems. The category of workgroups has fallen the most, notes Joseph, hit hard by the migration of HPC to the cloud in a trend that is likely to continue.
Notably, the surge in sales of very large supercomputers, of which Fugaku is an example, prompted Hyperion to reorganize its supercomputer segments. It will now follow three sub-segments in the future: Leadership / Exascale class systems; Supercomputers-Large ($ 3M and over); Entry-level supercomputers ($ 500,000 to $ 3 million). Note that the lower limit of the global supercomputer division ($ 500,000) has not changed.
“The [are a couple] of the reasons why we do this. Exascale / leadership class machines are on a very different trajectory than the rest of the market. First of all, just the size of these systems with their price tags; we are now talking over $ 300 million, a lot at $ 500 million, [some] which actually reached a billion dollars. You can also see over the next five years [that their] the growth path is very different at 23 percent. It’s a substantial change, so look to the future for us to divide the market this way, ”Joseph said.
In at least one sense, the new labels make sense; they more accurately reflect the power of the subcategory systems, i.e. an entry-level supercomputer is just that, and can help clarify some of the previously opaque content of the category. Hyperion supercomputer.
Capturing the full presentation of Hyperion is beyond the scope of this article; Also, many numbers (naturally) haven’t changed since Hyperion’s last market update presented in May at the HPC User Forum, which HPC wire covered (see Hyperion: HPC Server Market Gains 1 Percent in 2020, Storage Ready for ‘Tipping Point, and Hyperion Provides Snapshot of Quantum Computing Market).
To make it easier to find what you’re looking for in the latest update, Hyperion has broken its lengthy ISC presentation into separate video segments by category and published links to each segment. These links are included at the end of the article.
Hyperion is currently forecasting 6.8% growth in on-premises servers of all types through 2024 after a slight decline in 2021. Interestingly, in the server category, Hyperion claims that dedicated HPDA / AI server revenue is increasing 4 times faster than all on-site servers.
While server sales represent the bulk of the HPC market (42% according to Hyperion), the storage category has long occupied the second location (17% in 2020). Hyperion senior analyst Mark Nossokoff has pleaded for storage to become even more important. Driven by the familiar deluge of data in computing, not just in HPC, also by the growth of data-intensive AI technologies, and finally by efforts to co-locate memory closer to compute, storage is about to carve out an even bigger slice of the HPC calculate the pie, he argued.
“Storage continues to be the biggest growth statement of the various major areas of the HPC market, forecasting a growth rate of 8.5% over five years, peaking at $ 8 billion in revenue in 2024” , Nossokoff said. “This represents about 20% of overall user spend on on-premises HPC infrastructure. For every dollar spent on on-premises servers, approximately 40 cents is spent on storage. HPC cloud storage is growing faster, more than double the growth rate of on-premises storage over the forecast period, accounting for about one-third of user spending on HPC resources in 2019. Cloud storage is expected to grow. reach nearly $ 3 billion. 2024. “
As proof of the growing importance of storage in HPC, Nossokoff cited specifications for several upcoming systems (see slides below).
“Several of the Department of Energy’s new and upcoming leadership machines have shared more details regarding their respective storage architectural elements. NERSC has invested heavily in Perlmutter to optimize Luster for heterogeneous, large block sequential, and small block random concurrent access for what should be one of the fastest AI-powered HPC machines when fully operational ” , Nossokoff said.
“Lawrence Livermore gave an overview of the El Capitan storage subsystem. Dubbed Rabbit, the subsystem will simultaneously support access to the shared file system and a low-latency direct path from the compute nodes. And while not exclusively storage related, the Stream Resource Manager is designed to have local knowledge of resources and optimally allocate them at a more granular level than traditional resource managers. Finally, Oak Ridge (National Lab) recently shared the storage specifications for Frontier which is expected to be the first exascale computer in the United States. The specs are impressive in every way, ”said Nossokoff.
Nossokoff noted Samsung’s decision to move persistent memory support from NVMe to CXL and said, “This reflects a potential accelerated shift in CXL adoption within HPC for the CPU, GPU, memory hierarchy. and storage, ”adding“ emerging consumption and utility models, as evidenced by HPE’s support for HPC workloads on Green Lake and the shift of VAST from a hardware infrastructure provider to a software.
File systems also continue to evolve. In addition to the previously mentioned Luster innovation, Seagate announced the release of CORTX, an object archive file system, in conjunction with its on-premises storage infrastructure and storage utility strategy. in the cloud. Another interesting development in the file system has been HPE’s support for the IBM Spectrum scale, reflecting the growing adoption of HPC-enabled AI infrastructures in the enterprise.
Last but not least, in the world of HPC, he said, several large HPC sites have shared their strategies for increasing the use of the cloud for HPC storage. “Almost four exabytes of storage are expected to be deployed under the UK Met-Microsoft deal to meet the premier’s next-generation requirements for weather research and data storage and retrieval research within its archives. active data. To complete this non-exhaustive cloud storage activity, Riken announces a partnership with Oracle for its cloud storage needs on Fugaku.
Two bonus slides
Video links to the HPC Hyperion Research Marketplace update during ISC21: