Buying a new workstation for Ansys LS-DYNA is not the same as buying one for general FEA or CFD. The LS-DYNA solver puts unusual demands on a machine, and the parts that matter most depend on the type of analysis you run. This means that a high price tag or large core count on a spec sheet won’t necessarily guarantee you faster results – so you need to carefully consider where it is worth spending your hard-earned money. This guide covers the three things that LEAP’s Explicit Dynamics team advises you to look at first:
- what your analysis type needs,
- how cores and memory fit together, and
- a short checklist to take to your supplier to narrow down your options
1. Start with your analysis type: explicit needs speed, implicit needs memory
LS-DYNA users broadly fall into two camps, and each should spend the budget differently.
Explicit analysis (crash, drop, impact, blast) takes millions of very small time steps. Each step is quick, but the CPU constantly needs fresh data from memory.
- CPU: look for a high base clock speed and a large cache. The base clock is the speed the chip holds when every core is busy, which is how LS-DYNA runs. The cache is fast memory on the chip itself, so a bigger cache means fewer trips out to RAM.
- Memory: bandwidth matters more than size. Bandwidth is how fast data moves between RAM and the CPU, and explicit runs are usually limited by it. Aim for no more than about four cores sharing each memory channel.
- Storage: run jobs on a fast local NVMe drive. Check the read and write speeds, as these drives vary widely.
Implicit analysis (static loading, slow events, springback) solves one very large system of equations, and that changes the priorities.
- Memory: above all, buy enough to solve “in-core”, meaning the whole model fits in RAM. If it doesn’t, LS-DYNA moves the work to disk and slows sharply.
- CPU: the implicit solver scales well, so more cores reduce run time.
- Storage: fast NVMe is the fallback.
A few rules apply to both: use ECC memory, fill every memory channel, and plan for long runs with good cooling and, where needed, a battery backup (UPS). The graphics card is used for pre- and post-processing but not by the solver.
2. How cores and memory channels fit together
The most overlooked part of an LS-DYNA build is the link between the CPU and its memory. Every CPU has a memory controller with a fixed number of channels, the lanes data travels along between RAM and the cores. The CPU and motherboard decide how many there are, and you can’t add more later.
Take a 32-core CPU with eight memory channels. Each channel is shared by four cores, a comfortable ratio for LS-DYNA. Put the same 32 cores on a four-channel platform and eight cores share each lane, so the cores spend more time waiting for data.
When sizing a machine, we work through four checks in order:
- Licence: buy at least as many cores as your LS-DYNA licence covers. LS-DYNA splits the model into one piece per core, and unused licence cores are wasted money.
- Elements per core: for MPP runs, aim for no less then 10,000–20,000 elements per core. A one-million-element model would technically scale up to 100 cores.
- Cores per channel: ideally four or fewer. This is the check that most often separates a well-balanced machine from an expensive disappointment.
- Memory stick: identical memory stick in every channel. For explicit work, eight 16 GB sticks (128 GB) is usually plenty. For implicit, choose larger identical sticks, such as 8 × 32 GB or 8 × 64 GB, which keep the same speed with far more capacity.
3. The buying checklist
Before you sign off on a quote, run through these.
Do:
Match the licence. Get at least as many physical cores as you are licensed for. Fill every memory channel with identical stick. On an eight-channel CPU, 8 × 16 GB is about twice as fast as 4 × 32 GB, even though both give 128 GB. Buy RAM at the CPU’s rated speed; faster sticks won’t run any faster. Choose ECC memory. It corrects memory errors, and a single error can ruin a run lasting several days. Run jobs with a fast internal NVMe drive. Result files can reach hundreds of gigabytes. Plan for long runs. Use good cooling, and add a UPS if power interruptions are a risk.
Don’t:
Count threads as cores. A 32-core CPU often lists 64 threads; only the physical cores count. Choose CPUs that mix performance and efficiency cores. LS-DYNA runs at the pace of the slowest core. Add cores without adding memory channels. The extra cores mostly wait for data. Overspend on the graphics card. For most LS-DYNA work a modest card is enough. Run jobs from a network drive or spinning hard disk. They are far too slow for large result files; keep those for archiving.
Three questions are worth asking any supplier: how many memory channels does the system have, are they all filled, and is the memory ECC?
Final thoughts
The right workstation for Ansys LS-DYNA usage is a balanced one, not simply the most expensive. If you’re unsure how many cores your models need, we advise testing a representative benchmark model at increasing core counts to develop a reliable guide to scalable performance. At LEAP, now part of PDSVISION, our Explicit Dynamics team has extensive experience in this area and we’re happy to help you with benchmarking models that will help you arrive at the right hardware choice. You can reach out to us here.