If you want it done best, do it yourself! DMS developed a custom built high performance computer (HPC) for CFD calculations. But the exciting part: it cost about what you paid for that Dell workstation!
1.0 Why You Care?
DMS does not charge for our in-house computing hardware! No hidden charges when we keep it in-house. That was the cost benefit from custom building everything. The performance benefit? We can tackle several marine CFD projects without charging you for hardware.
- Ship resistance studies
- Internal flow studies
- Heat exchangers
- Rudder performance optimization
- Waterjet inlet optimization
- Propeller performance optimization
- Far more
Beyond the project benefits, this also allows DMS to pursue more ambitious research and validation studies. We do not use your money to learn on the job. Our CFD processes are tested and validated before we use them on your projects.
2.0 What We Learn From A Custom Build
A custom built HPC reflects the DMS strategy for our business model. First, develop a detailed understanding of the business needs. Build the wisdom of the reasons for business requirements before acting on them. Then invest in a solution that fits those needs. Immediate return on investment is not necessary. Long term satisfaction is more important (provided cash flow is not jeopardized).
Our decision to invest in a custom HPC hinged on two important revelations:
- HPC computers are not servers.
- The introduction of the Ryzen Threadripper CPU
2.1 HPCs Are Not Servers
Most IT administrators turn to server technology when asked to build an HPC computer. But servers and HPCs are two different categories of computers. A server is designed for high reliability. Servers should never develop hardware faults. Servers should not bet pushed to their limits. By contrast, HPC machines are intentionally pushed to the edge of processor capabilities. Imagine something similar to a hot rod. A well designed HPC spends its time with the CPU pegged at 100%, generating loads of heat. HPC computers have more in common with high grade gaming computers than servers. (Figure 1‑2)
The second element of servers is their reliability. Severs exist as critical infrastructure to business networks. Server failure means disastrous consequences: the entire business stops working. HPC failure offers relatively minor consequences. Imagine the CPU fails on an HPC, what are the consequences? One engineer stops working,and the project may lose the data generated that day. With rush shipping, it takes only 1-2 days to buy and install a replacement part. Consequences for the business: delay of three days to a single project. Relatively minor. We do not need to pay for the reliability of a server in HPC machines.
3.0 Ryzen Threadripper
Thank goodness for competition between AMD and Intel. These two giants constantly battle to manufacture the fastest CPU. When building an HPC, the CPU is the most important component. You want the most cores possible, as fast as possible. And AMD recently delivered anew product line to meet that demand: Threadripper. They pack up to 32 cores on a single chip. Just 10 years ago, it would take an entire tower of computers to achieve the same number of cores, at only a fraction of the computing power! For those of us who want small HPC capabilities, we no longer need to build a computing cluster. Threadripper allows us to compress an entire cluster into a single chip!
For DMS, this HPC hardware represents potential. The potential to research more advanced technology and prove CFD accuracy for more detailed simulations. And the potential to deliver advanced engineering at a lower cost. Thanks to the custom construction at DMS, high performance computing no longer means high cost. What can you do with affordable advanced engineering?