How do you evaluate the environmental impact of cloud HPC vs on-premise HPC?

One important but sometimes underestimated factor is the influence cloud HPC has on the environment. Businesses need to carry?out careful due diligence to make sure that cloud providers correspond with their sustainability goals and principles, even though they may brag about their usage of renewable energy and carbon offset programs. Reducing the possible environmental impact of cloud-based HPC operations can be accomplished in part by being aware of cloud providers' environmental policies and practices.

Cloud HPC offers significant advantages in scalability and operational costs compared to capital expenditure. It enables quicker implementation and supports numerous engineering applications for concept validation. However, in our initial implementation of HPC for automotive engineering applications, we found that running applications 24/7 on the cloud was extremely costly. The expense was high enough that we could purchase and maintain a server for the same cost as running these applications in the cloud. Therefore, a hybrid approach is the most effective solution.

A public (Shared) cloud can offer scalability that a private (on-prem) cloud may not be able to offer. However, an on-prem cloud gives the CIO/IT administrators an opportunity to purchase more exact configurations needed for their particular type of HPC, i.e. more GPUs per server, more memory per core, local storage, etc.

Benefits: Cloud HPC leverages data centers optimized for energy efficiency, often powered by renewable energy. Providers handle e-waste responsibly and offer scalable resources, reducing the need for over-provisioning. Drawbacks: Potential higher energy use due to data transfer and less transparency in energy sources used. Dependence on provider’s sustainability practices can vary.

Perhaps the most significant way to reduce a data centers environmental impact is to know where your power comes from. When using some cloud providers, a user may have no idea if the public cloud data center is using renewable power or not. Also, an on-prem data center can be configured for exactly the use that the enterprise or organization requires for their workloads. And, there may be situations where a on-prem data center can acquire the latest technology first.

On-premise HPC environment comes with these benefits:1.Customization: On-premise HPC environments allow for easy customization of scripts to meet specific needs. 2. Servers can be partially allocated to different applications, maximizing resource utilization. 3. Keeping data within an on-premises environment ensures a higher level of security. Drawbacks: 1. Maintenance Overhead: Significant overhead costs are associated with maintaining a datacenter. 2. Capital Investment: A substantial capital investment is required for hardware, and with rapid advancements in technology, keeping up with market changes is challenging. 3. Skilled Personnel: Maintaining the environment requires experienced administrators and specialized knowledge.

One other point to consider is the type of HPC workload being evaluated and where your data set is being generated. If the bulk of your data is being generated from an on-premise source then you need to factor in the costs associated with data transfer. If you are training you are going to be ingesting and transferring a lot of data over time.

To best my knowledge On-premise HPC systems provide full control over hardware and software configurations. For HPC servers running 24/7, on-premise systems can be more cost-effective in the long run. This can yield a high return on investment (ROI) through reduced ongoing operational costs compared to cloud services. Drawbacks of on-premise Setting up an on-premise HPC system requires substantial upfront investment in data center infrastructure, server purchases, and maintenance, In house team needs to maintain and upgrade. Periodic upgrade of hardware needs to be done

Benefits: Greater control over energy sources, potentially allowing the use of local renewable energy. Customizable for specific efficiency needs and can reduce data transfer energy costs. Drawbacks: Typically lower utilization rates lead to higher energy waste. Requires significant investment in cooling and infrastructure, which may not be optimized for energy efficiency. Generates more e-waste managed by the organization.

Conduct a Life Cycle Assessment (LCA) to evaluate the environmental impact from cradle to grave. For cloud HPC, assess the provider's data center operations, energy sources, and hardware lifecycle. For on-premise HPC, evaluate manufacturing, operational energy use, cooling requirements, and disposal of equipment. Compare total carbon footprint, energy consumption, and waste generated.

In my experience, benchmarking is a crucial step in cloud implementation. While many vendors provide their own statistics on the benefits of moving to the cloud, it's essential to determine what works best for our organization and support that decision with solid data. Here are key factors to consider: Performance Metrics: Compute Performance: Network Latency and Throughput: Data Transfer Speeds: Upload/Download Speeds: Bulk Data Migration: Support Response Times: Implementation and Integration: Deployment Speed: integration with Existing Systems: By benchmarking these aspects with detailed metrics, you can make an informed decision about the best cloud environment for your organization.

Use benchmarking methods to measure and compare the environmental impact. Metrics include Power Usage Effectiveness (PUE) for energy efficiency, carbon intensity of the energy used, and resource utilization rates. Tools like SPECpower or Green500 can provide standardized benchmarks. Compare these metrics for both cloud and on-premise setups to determine which offers better environmental performance based on your specific use case.

Qarnot has developed a distributed and sustainable alternative by deploying its servers directly where IT waste heat can be valorized: near major heat consumer sites. The Qarnot model has been designed with the aim of optimizing data center energy consumption, considering it not only as a consumer but rather as an energy producer.. Using patented direct water cooling technology, their digital boilers (QBx) repurpose the heat from servers to warm water up to 65°C. This hot water is then used to heat buildings, swimming pools, and industrial sites. The system operates in a closed loop, conserving all the water and maximizing heat recovery, capturing up to 95% of the heat emitted by the servers.