WORKLOADS
A simple test: if it can wait for a queue and run for hours, it is a network job. If it must answer now, it stays on your own systems.
DESIGN TARGETS — E31 NETWORK IS IN ACTIVE DEVELOPMENT AND HAS NOT SHIPPED
RUNS ON THE NETWORK
GPU-heavy jobs that can run asynchronously.
Submitted, queued, run on compute rented for the duration, returned as output.
TRAIN & TUNE
Training, fine-tuning, retraining
- Model training
- Fine-tuning
- Retraining on new data
- Evaluation jobs
Hold the GPUs a run needs for the hours it takes. A new model comes back; the compute goes back to the pool.
PROCESS AT VOLUME
Video, image, and dataset processing
- Video processing
- Image processing
- Large dataset processing
- Data transformation
A backlog of footage or a large dataset, worked through as one batch instead of one file at a time.
SIMULATE & SYNTHESIZE
Simulation, rendering, synthetic data
- Simulation
- Rendering
- Synthetic data generation
Thousands of runs, frames, or generated samples that no single workstation should carry.
INFER & ANALYZE
Batch inference and analytics
- Batch inference
- Large-model inference
- Analytics
- Other GPU-heavy asynchronous jobs
Everything collected since last time, run through a model or an analysis in one pass.
THE SHAPE OF A NETWORK JOB
Four properties every network job shares.
- DURATION
- Minutes to many hours. A job has a beginning and an end.
- TIMING
- Scheduled or asynchronous. It can wait for a queue; it does not need an answer now.
- OUTPUT
- An artifact: a model, processed files, results, metrics. Not a live stream of decisions.
- FAILURE
- If the job is late or has to restart, nothing physical is affected.
REAL-TIME
STAYS ON YOUR OWN SYSTEMS · NOT E31
- Control loops and motion
- Navigation and safety systems
- Millisecond inference
- Live perception at the sensor
- Dedicated local or edge hardware
BATCH / ASYNCHRONOUS
SUBMITTED TO E31 NETWORK · BY THE HOUR
- Training, fine-tuning, retraining
- Batch and large-model inference
- Video, image, and dataset processing
- Simulation, rendering, synthetic data
- Evaluation, transformation, analytics
DOES NOT RUN ON THE NETWORK
Anything real-time.
E31 Network is not robot infrastructure, not a control plane, and not a runtime for autonomous systems. Real-time work stays on dedicated local infrastructure, edge devices, robots, or customer-owned systems.
Control loops, motion, navigation
On the robot or its controller, as the vendor shipped and certified them.
Safety systems
E-stops, interlocks, speed and separation monitoring: dedicated local hardware, never a remote job.
Millisecond inference
Anything that must answer inside a control cycle runs on the edge device beside the sensor.
Live perception
Obstacle detection and the sensor processing a machine needs to act now.
Always-on services
Systems that must be reachable every second belong on infrastructure you own or dedicate.
PLC and machine control
Programmable logic keeps final authority over the actuation it already owns.