Scientific simulations. AI training. Robotics and drones. Put GPU resources and guided workflows to work on your own research, with results you can inspect.
A static showcase from EMT Aerospace Technologies S.r.l. GPU workloads run separately on EMT Compute using paid resources.
EMT brings together aerospace R&D, artificial intelligence, robotics and autonomous systems. EMT Compute connects that engineering focus with accessible, on-demand computing workflows.
Use a supported scientific preset or bring your own code. Compare parameter sets and collect results, provenance and program-declared calculation steps.
Guided choices for data, model, training method, resources and costs, with checkpoints. Deploy supported one-click models on a suitably sized single machine with HTTPS access.
Develop control strategies in simulation: locomotion, manipulation, drone flight and driving. Organize experiments around task settings and measurable outcomes.
Physical deployment requires its own integration and validation.
Explore robotics workflows ↗Workflow overviews describe supported product paths, not results from a robotics or model-training benchmark. Availability, licensing and workload compatibility apply.
Bring your Python program, command and input files. Record equations, inputs, outputs and explanations with ctx.explain(...) or EMT_CALCULATION_JSON. Use independent parameter sweeps across nodes, or your own compatible distributed implementation.
Inspect recipes, recorded loss and evaluation metrics, learning rate, gradient norm and runtime details. Export the complete event history. Observed dataset hashes help check remote-data continuity when resuming from a recorded checkpoint.
Begin with a guided experiment or prepare your own workload. Review resources, duration and the EUR rental quote before launch. Connect your tools through REST, Postman, n8n and MCP.
Release 0.9.1. Metrics are shown when recorded. Assigned or visible GPUs are distinct from actual device usage; calculations supplied by custom code remain attributable to its author. Older training checkpoints may lack a dataset-hash baseline.
Sixteen custom-code runs performed a float64 matrix multiplication on CUDA across two H100 nodes. Every saved element was checked against the expected result. Select a run to inspect its recorded sum.
A[i,j] = i × 256 + j
B = s × I
C = A @ B = s × A
N = 256² = 65,536
sum(C) = s × N × (N − 1) / 2Each run checks 65,536 elements. The analytical scalar sum provides a separate comparison with the CUDA result. This demonstrates the stated calculation, not general solver accuracy or GPU throughput.
Loading recorded data…
The selector reads saved results. It does not launch GPU work or perform a live benchmark.
The original test includes a deliberate 45-second pause to exercise worker allocation across both nodes. That time is not a matrix-performance measurement. This public extract excludes private account, job and infrastructure identifiers.
Run independent experiments across multiple workers, or use compatible distributed code when the method requires communication between nodes. Choose from current availability and review a fresh quote before launch.
A cluster does not automatically distribute arbitrary code. Interconnect performance must match the workload; the recorded test does not qualify InfiniBand/RDMA bandwidth.
Inspect inputs, runtime, hashes, outputs, numerical checks and the calculation steps declared by the program. Custom experiments can record those steps with ctx.explain(...) or EMT_CALCULATION_JSON.
Reports document execution; they do not reconstruct missing intermediate steps or prove an arbitrary program correct.
Start with 24 requests for resources, quotes, existing rentals, training jobs and experiment reports. Public discovery works without a key.
Get the API collection ↓Connect to 21 tools for GPU resources, custom experiments and training inspection. Use an appropriately scoped API key and a client that supports authorization headers.
MCP setup and documentation ↗Import the downloadable workflow for quotes, budget checks, job observation and report retrieval. Paid launch is disabled by default.
Get the n8n workflow ↗Tell us what you want to calculate, the data you have, your success criteria and timing. We can scope a paid pilot with agreed technical objectives and budget, then evaluate a continuing integration.
info@emtaerospacetechnologies.com