How your HPC colocation decisions boost your HPC capacity

Posted: 8 Apr 2020

Chances are, if your workloads include heavy compute capacity, you’re leaving money on the table. It doesn’t have to be so.

In the UK, Ireland and mainland Europe, a vast majority of data center-based compute takes place in what’s commonly known as the “DFLAP” marketplace.

Dublin, Frankfurt, London, Amsterdam and Paris (DFLAP) are Europe’s centerpiece data center locations today. Some 72% of all Dutch data centers, for instance, are in Amsterdam — whose premium-priced real estate houses a vast majority of all compute for enterprises in Benelux countries. The same goes for London and Dublin and their respective lion’s share of compute for all the United Kingdom and Ireland.

Yet when you consider facts, there is really no good reason for DFLAP data centers to be running power demanding, heavy compute workloads. DFLAP cities have some of the highest commercial real estate prices in Europe and their electricity rates and access to power are, what one would expect for high-density urban environments. These might be perfectly fine for Paris, Frankfurt, London or Amsterdam, but they’re simply no match in comparison to other regions’ electricity rates and power capacity (and we are not even adding sustainability considerations to the equation).

Consider, for instance, the “RCOSH” cities: Reykjavik, Copenhagen, Oslo, Stockholm and Helsinki. Nordic countries represent the strongest package of: (sustainable) electric prices, electric grid reliability, well-trained workforce and (free) Nordic air cooling. Some Nordic colocation solutions represent up to 50% more cost-efficient operations than running one’s cluster in mainland Europe.

Splitting off HPC workloads

In many enterprise compute environments, HPC applications have been slow to migrate to the cloud. In fact, in many organizations, HPC workloads continue to be run in-house. This could be the case for an automotive manufacturer, who still runs their crash simulations on in-house clusters or a business whose sales division runs their most compute-intensive forecasts and analytics on the in-house data center.

Moving HPC to an off-site colocation data center represents a real opportunity for savings. Therefore, it’s important to think outside the traditional DFLAP data center sites.

Colocating one’s HPC to a data center sited where power and cooling are the most cost-efficient and sustainable is good business, plain and simple. Considering that cost savings can then be channeled back into more compute cycles, making the right HPC colocation decision becomes one of compute capacity. For any given budget outlay, would you rather have access to more CPU/GPU cycles or less?

Few organizations would consciously choose the latter. And yet, implicitly they’re doing it every day they run their high-density workloads in-house or within the traditional DFLAP data center ecosystem.

In short, organizations need to think of their HPC colocation problem as one of moving photons, not electrons. They should think, that is, not of the power bills their local data center might generate but rather of the photons of data traffic they could be sending to a RCOSH data center — whose power and cooling costs are going to be much more cost-competitive and efficient.

Do more with less with your HPC budget. Download our case study to discover how HPC colocation to a Nordic data center can change your organization’s entire compute infrastructure — and bring in substantial savings as well as maximize your ROI.

Fill the form below to download the Case study. Right arrow