HPC comes with particular considerations that distinguish it from regular computing in terms of storage and care. Energy consumption is high, both for powering and cooling the machines, and this has an impact on cost, whether you are storing on-premises or colocating in a data centre.
High performance computing is changing the world. From the cyber heart that allows surgeons to simulate surgery, to the engineers streamlining bikes and 18-wheel lorries, it’s all made possible with HPC. Once solely available to universities and major corporations, these powerful systems are entering mainstream use thanks to their costs falling considerably in recent history. But what impacts the cost of HPC?
The cost of hosting HPC
The main cost of housing HPC is the energy requirement, which comes from two main power demands.
1. Power consumption from the hardware
When we talk about HPC, we’re talking about clusters of GPUs and CPUs, networked together to provide super computing power. The more processing units you have, the more power you need. So, standard computing may have a requirement of 3 – 7 kW per rack, HPC could push that up to 40kW per rack.
Energy prices vary widely, of course, but let’s estimate an average UK electricity price of 15.75p per kWh (as per figures on government website uSwitch). You can see how much difference the increased energy requirement of HPC makes to your bill – from 47.25p for a 3kWh rack to £6.30 for the 40kWh used to run an HPC rack.
How to maximise cost efficiency
If you’re getting the most from your HPC then that energy consumption is necessary. However, you can control your costs by making sure you don’t use and pay for power you don’t need. For the sake of an easy visual, let’s talk about kettles:
A single kettle uses about 3 kWh. Now imagine you have 10 kettles constantly boiling. That’s the energy requirement of your 30 kW HPC rack. In order to get a reasonable return on investment you’d need to make 60 cups of tea from those kettles every hour (i.e. utilise their full capacity). Do you have a need for that much tea? If not, then that power is going to waste. The same applies to your HPC application. If you are running more GPUs and CPUs than you need, you are paying for power you’re not using.
2. Power consumption from cooling
The other significant cost comes from cooling. All that power generates a lot of heat, which has to be controlled to reduce the risks of damage to electronics or – worse still – fire.
Cooling can account for up to 50% of energy costs for an HPC system, since HPC’s increased power consumptions and high density exponentially increases the heat it produces.
How to maximise cost efficiency
While air cooling works well for standard racks, liquid cooling is the more efficient cooling option for HPC.
It may sound nonsensical to put liquids in close proximity to electronics, but in reality, liquid cooling works in a similar way to radiators or fridges, cycling cold liquids through leak-proof, enclosed systems to bring cooling to exactly where it’s needed.
Liquid cooling can be 3 – 4 times more efficient than air cooling and can be done with water or other liquids using a variety of different systems and techniques. This increased efficiency reduces the money you have to spend on powering the cooling system. We use rear-door cooling, which is one of the most energy-efficient cooling option.
So, is it more expensive?
Yes and no. Housing your HPC system on-premises will cost you a lot more than your standard computer system because you will need to install additional cooling and power infrastructure, as well as running costs. However, data centres typically benefit from lower power costs and access to more efficient cooling methods, which increases their power usage effectiveness (PUE). You also get the benefit of other economies of scale, like better connectivity costs. The other thing to consider is density. If you have four 7 kW standard density racks that could be condensed into one 30 kW high-density rack, you could actually end up saving money through cost efficiencies.
In short, HPC-ready data centres are built for exactly this job and so can perform it more efficiently and more effectively than your standard server room. That being said, be aware that not all data centres are designed to provide the power and cooling requirements of HPC. Do your research before making any decisions about where to house your HPC application.
What are your priorities and what are they worth?
Most likely you are considering HPC because of what it can offer you, and the value of all that processing power to your business will outweigh the cost.
Of course, that value depends on building the right solution and running it properly. Otherwise you risk losing productivity, missing opportunities to innovate, running up debt and generally failing to capitalise on what should be a hugely exciting tool.
That’s where partnering with an HPC-ready data centre can really help since it enables you to gain the benefits of their experience, the economies of scale for both power and cooling and comes with the added advantage of built-in resilience. Data centres such as ours are designed to provide you with 99.999% uptime, whatever’s going on in the world, from pandemics to power cuts and everything in between. These are benefits worth paying for if you’re serious about harnessing the potential of HPC for your business.