Hardware FAQs
Answers to common questions about Rescale hardware: which browser to use, how On Demand priority options differ, and how to pick a coretype and core count for a simulation. For deeper detail on any topic, follow the links into Coretypes, GPU and accelerators, or Regions.
System requirements
Which browsers are supported
The Rescale platform delivers the application through any HTML5 browser. Rescale officially supports the latest versions of:
For the best performance and feature coverage, use Google Chrome.
What display resolution is required
The Rescale platform supports most resolutions. The minimum supported resolution is 1920 x 1080.
On Demand pricing and priority
What is On Demand Priority
On Demand Priority (ODP) is the default pricing and priority level for general use and high availability. An ODP job runs to completion without interruption from the cloud service provider, so it is the right choice for long-running or time-sensitive jobs.
What is On Demand Economy
On Demand Economy (ODE) is a lower-priced priority level. ODE jobs enter the run queue and may experience a moderate delay before starting. Once running, the underlying hardware can be reclaimed by the cloud service provider at any time, with or without warning.
Rescale attempts to restart a reclaimed ODE job based on a configurable policy. You can choose, for example, to always restart on a non-preemptible instance, or to attempt up to five restarts with the last restart falling back to On Demand Priority.
When should I pick ODE versus ODP
Use ODP when:
- The job is time sensitive or has a deadline.
- The job is long running and a restart is expensive.
- You cannot tolerate reclamation mid-run.
Use ODE when:
- The workload is flexible and you care more about price than latency.
- The job is short or checkpoints frequently, so a restart is cheap.
- You are running a parameter sweep where individual replicas are disposable.
You can combine the two inside a Coretype Set: start with ODE for cost, then fall back to ODP if ODE is reclaimed or unavailable. See Coretypes for details on Sets.
Why don't I see ODE on every coretype
Not every coretype has an ODE variant. The priorities that appear in the Price column on the Hardware step depend on what is configured for the selected coretype. Coretype Collections, for example, are ODP-only today, so the ODE column is hidden.
Choosing hardware for a simulation
How do I decide on the number of cores
Core count is driven by your model size, model fidelity, and cost target:
- Model size. Cells, grid points, nodes, elements, degrees of freedom.
- Model fidelity. Time accurate vs steady, single vs double precision, solver choice.
- Cost focus. Minimize dollar cost per solution.
- Cost-performance focus. Balance time to solution and compute cost.
For CFD, a common starting point is 75,000 to 100,000 cells per core. Start with a smaller core count to validate the model, then scale up once you know the solver converges cleanly.
Which coretypes does Rescale recommend for CFD and FEA
For CFD or FEA workloads, the typical starting points are Emerald or Amber. For FEA with higher memory requirements, start with Melanite, Zinc, or Topaz.
You can find the full per-coretype spec (cores per node, memory, storage) on the Hardware Settings section of a job, or in Coretype Explorer. For example, Emerald has 36 cores per node with 144 GB memory and 1.3 TB storage per node; Melanite has 48 cores per node with more memory, storage, and cores than Emerald.
Where do I compare coretypes side by side
Use Coretype Explorer. You can sort by memory per core, network bandwidth, GPU model, or price, and put multiple coretypes next to each other before you commit a job to one.
What about GPU workloads
See GPU and accelerators. GPU coretypes pair an NVIDIA GPU (A100, H100, V100, or P100 in the current portfolio) with a CPU host. Use them for AI training, GPU solvers, and NVIDIA Modulus workflows.