distributed-compute

You orchestrate distributed compute by launching a Conductor service and connecting Carriage workers to distribute WorkAssignments across Metal GPU and NEON CPU devices. This architecture coordinates load balancing and task scheduling across all connected hardware.

The Conductor and Carriage architecture is a distributed compute model where the Conductor service manages task scheduling and load balancing, while Carriage workers execute WorkAssignments. It handles data sharding and fault tolerance across multi-device networks.

Yes, you can use NEON pre-filtering before Metal GPU dispatch. The three-gear engine coordinates GPU dispatch and NEON pre-filtering to optimize workload distribution and processing efficiency across the distributed compute network.

Fault tolerance in this distributed compute network operates through JSON-based network messaging with continuous heartbeat monitoring. If a device fails, the Conductor detects the lost heartbeat and triggers failover to redistribute WorkAssignments to available Carriage workers.

Yes, distributed compute supports JSON-based network messaging for multi-device coordination. It uses JSON messages with heartbeats to maintain communication between the Conductor and Carriage workers, ensuring reliable failover and continuous task scheduling.

You need data sharding for distributed computation when running tasks across multi-device networks to balance the processing load. Sharding splits your data so the Conductor can efficiently distribute WorkAssignments across Metal GPU and NEON CPU workers.