AceCAST 5: what's new
AceCAST 5 is available now. Here is what changed: WRF 4.8, four physics updates and four new model outputs.
HPCwire profiled TempoQuest in April 2023. This is our own account of the subject: why weather models are moving to GPUs, what has been measured since, and what TempoQuest offers today.

Forecast teams keep asking for finer grids and more frequent runs, and the CPU clusters that weather models were written for are expensive to grow. GPUs suit the highly parallel arithmetic of a weather model, but code that has been built up over decades does not run on them without serious rework. In April 2023, HPCwire profiled TempoQuest's part in that shift in an article titled Meet TempoQuest, a Startup Bringing Weather Code Into the Accelerated Era. Here is our own account of the same subject, and of what has changed since. For the original reporting, read HPCwire.
HPCwire interviewed TempoQuest's COO. The profile set out the problem TempoQuest works on. WRF, the Weather Research and Forecasting model maintained by NCAR, is used around the world for regional forecasting, and it was written for CPUs. Commercial users in fields such as renewable energy, agriculture, aviation and logistics want finer and more local forecasts than public models provide. As he told HPCwire: "What they need is very, very high-resolution, precise and hyper-local forecasts."
The profile's performance figures reflect early 2023. Measurements published since, with their sources, are below.
A weather model is less a single program than a collection. A dynamical core moves air around the grid. Separate physics schemes handle clouds, rain, radiation, turbulence and the land surface, and a forecaster picks a combination of them in a configuration file called a namelist. The schemes were written and tuned by different scientists over many years.
Two things follow from that. First, the whole chain a forecaster selects has to run well on the GPU, not only the heaviest piece. Second, the answers have to stay scientifically equivalent to those of the CPU model, or the forecaster has no reason to trust them.
For a buyer, the practical test is simple. Does your existing WRF setup run, with the same inputs and outputs, and how long does it take? AceCAST is built for that test. It keeps the WRF workflow, with WRF-compatible inputs, namelists and outputs, and it runs on Linux systems with NVIDIA GPUs, on premises or in the cloud.
Later the same month, on April 28, 2023, Microsoft and TempoQuest published a benchmark on the NVIDIA Technical Blog. It was run on Microsoft Azure with two earlier AceCAST releases. The GPU side was a virtual machine with 8 NVIDIA A100 GPUs. The CPU side was virtual machines with 120 AMD EPYC Milan-X cores each.
See the benchmark and cost example for the separate releases and configurations.
Separately, Veer Renewables published a validation of AceCAST. It reported AceCAST performing at roughly five times the speed of CPU-based WRF simulations, with only minimal variances between WRF and AceCAST.
All of these figures describe earlier releases on particular hardware and test cases. Speed depends on the domain, the physics options and the machine, and the cost example uses list rates recorded in November 2026. Our article on the Azure benchmark goes through the numbers.
TempoQuest is based in Las Vegas, Nevada. It has four products.
AceCAST 5 is available now. Regional MPAS on GPU is coming soon.

AceCAST 5 is available now. Here is what changed: WRF 4.8, four physics updates and four new model outputs.

Aerava runs a weather model around the site you pin and checks the result against your own storm, wind and turbine cut-out limits. Here is what you can run today, what it costs and what it says about accuracy.

One published 2023 benchmark and two Azure list prices, worked line by line: what a GPU VM costs against one CPU VM and against sixteen, and what the sums leave out.
Your forecast
Tell us the decision it has to serve. One short form reaches the team, and we review every request.