What computer hardware do you recommend for running Bladed?
We are unable to offer specific hardware recommendations because we do not have benchmarking figures for Bladed simulations from a range of CPU architectures or machine configurations.
However, there are a few high level pointers:
The simulation code (dtbladed.exe) is single threaded, therefore we recommend choosing a CPU with fast single threaded performance (see https://www.cpubenchmark.net/singleThread.html).
We recommend dedicating one CPU core for:
- each running simulation
- each instance of the user interface
- the Batch manager and runner (there is one runner per machine accepting batch jobs)
- one core at all times for the operating system.
Regarding parallel instructions, Bladed does not support AVX. Dtbladed 4.5 and later use SSE2 but not AVX. Dtbladed 4.3 does not support any parallel instruction set.
A Bladed simulation typically consumes around 250 MB RAM. When running a large number of parallel simulations on a single process server, throughput is affected by the performance of the memory subsystem and disk write times. We therefore recommend choosing a machine equipped with low latency RAM and use of SSD disks for result output.
DNV now offers Bladed users the option to run simulations in the Amazon cloud (AWS). This allows you to securely upload and run a large number of simulations in parallel using dtbladed 4.3 or later, and download the results to your network on completion. The benefits of this arrangement are that you pay just for the cloud CPU time you use and do not need to maintain your own HPC infrastructure.