Set the tempo. Not the machine.
Tutti is an execution layer for GPU workloads. Tell it the GPU memory, budget and deadline. It picks the capacity, starting with AWS Spot, and keeps the run moving when that capacity is reclaimed.
- GPU memory
- 24 GB min
- Software
- CUDA 12
- Play until
- 18:00
- Spend under
- $42
Spot is cheap. Babysitting it isn’t.
Spot GPUs cost far less than On-Demand, but AWS can reclaim them with two minutes’ warning. Most teams decide the savings aren’t worth a job dying at 3 a.m.
With Tutti you don’t pick instance types. You describe the workload and set the limits; Tutti decides where it runs, splits it into short chunks, and requeues only the chunk that was interrupted.
Built for ML and infra teams with batch GPU jobs that split into short, independent pieces: batch inference, video generation, transcoding.
- 24 GB min
- CUDA 12
- by 18:00
- under $42
- queued · 640 chunks
- running on Spot capacity A
- capacity left the stage · chunk 212 requeued
- new seat found · Spot capacity B
- completed · 0 chunks lost
Completion, not machine choice.
Capacity left the stage. Your run didn’t.
When AWS reclaims Spot capacity, only the unfinished chunk is requeued. A new seat opens. The run carries on.
A batch workload, split into short chunks.
capacity: not yet selectedBuilding carefully. Starting with AWS.
Tutti is in active development. We’re validating execution and recovery for chunked workloads on AWS Spot, and looking for design partners who run batch GPU work.
What we’re measuring
- How often Spot capacity is reclaimed, and how much warning arrives
- Work lost per interruption, and retry overhead
- Jobs completed by their deadline
- Cost per completed job, against an On-Demand control run
Your work deserves an encore.
Tell us what you run. We’re looking for design partners with batch GPU work, and the founder reads every request. Give Tutti the conditions; we’ll keep the GPUs playing.
