Has high performance computing (HPC) reached a cloud tipping point?

clock • 2 min read

High performance workloads require a specialist environment that provide enterprises with low latency and high bandwidth. The capital cost of this is huge - often over $100m - and the requirement for regular refreshes means that costs can continue to accrue. However, the specialist nature of these workloads means that enterprises face a dilemma: should they continue investing in on-premise HPC infrastructure or turn to cloud to augment, or even replace those facilities?

Computing research found that while a high proportion of IT decision makers (29 per cent) have 100 per cent on-premise HPC infrastructure, there is an almost identical number of IT decision makers (28 per cent) that are cloud-only. There are others that are supplementing with an occasional cloud ‘burst' (25 per cent). The reliance on cloud is likely to ramp up in the next three to five years, with the vast majority (71 per cent) of those who already use for HPC workloads stating they plan to use cloud at least a little more. While half of those who run on-premise HPC plan on using more cloud or fully migrating to the cloud for HPC workloads.

These IT decision makers may be taking these decisions because those who have shifted to the cloud have had a broadly positive experience, citing a reduction of operating costs, increased business agility, avoidance of large up-front capital costs, better performance and even improvements in security. At a time when businesses are searching for new ways of working and additional revenue streams, it's interesting that a significant number agree that their cloud HPC is empowering ‘the creation of new business models'.

But what cloud?

There are different pre-requisites to a HPC cloud provider compared to cloud computing in general and that's perhaps why there are some surprises: of the hyperscalers Microsoft leads Amazon, while Google's lead over Oracle and IBM is wafer thin. This suggests that selecting a cloud HPC supplier requires more of an open mind, as the usual candidates may not be best suited to an enterprise's needs. However, a large majority of IT decision makers did opt for a cloud provider because they already use other cloud services from that provider. The question is whether that provider is most compatible with on-premise HPC or has the technical superiority of solution.

There seems to be two main camps: those who have large sunk costs in on-premise HPC that will be conscious of getting the most of their money by retaining on-premise HPC until its junk metal - this may be with the help of cloud computing, and those that are either already cloud-only, or aspire have a larger cloud presence.

Cloud is likely to become the dominant HPC approach - and enterprises are seemingly taking incremental steps to up their use of cloud HPC workloads.

To find out more, read Computing's research paper: The need for speed: achieving success in high-performance computing. Sponsored by Oracle.

You may also like
Oracle promises AI everywhere at Cloudworld 2024

Business Software

Integrated across the Fusion Suite

clock 19 March 2024 • 6 min read
US House passes bill paving the way for TikTok ban

Social Networking

Rumble reveals intention to purchase TikTok, but Oracle remains absent from the fray

clock 14 March 2024 • 3 min read
Oracle's controversial stewardship of Java: The good and the bad

Open Source

Oracle is doing a good job in keeping Java relevant. But that's pretty much where it ends, says Azul CTO

clock 15 February 2024 • 5 min read

Sign up to our newsletter

The best news, stories, features and photos from the day in one perfectly formed email.

More on Big Data and Analytics

Belfast to spearhead UK's digital revolution with £37m Digital Twin Centre

Belfast to spearhead UK's digital revolution with £37m Digital Twin Centre

Aim is to foster innovation across engineering sectors

clock 02 May 2024 • 2 min read
Even CERN has to queue for GPUs. Here's how they optimise what they have

Even CERN has to queue for GPUs. Here's how they optimise what they have

'There's a tendency to say that all ML workloads need a GPU, but for inference you probably don't need them'

John Leonard
clock 17 April 2024 • 4 min read
Partner Content: Why good data is the foundation of AI success

Partner Content: Why good data is the foundation of AI success

Does your organisation have the right quantity and quality of data to make its AI ambitions a reality?

clock 04 April 2024 • 2 min read