Press release

RiftStack Opens Registration for the Emmy Hackathon 2026, a Seven-Day Remote Contest to Write a PTX Code Generator for an Open-Source ML Compiler

September 29, 2026

RiftStack, the team behind the CloudRift GPU cloud platform, today opened registration for the Emmy Hackathon 2026, a seven-day remote event built around Emmy, its open-source ML compiler. The hackathon runs from Wednesday, October 21 to Tuesday, October 27, 2026. It is open worldwide to anyone aged 18 or older, working alone or in a team of up to three, and awards $1,500 for first place, $750 for second and $500 for third.

Every participant gets the same task: build the part of a GPU compiler that generates PTX, the assembly-level language for NVIDIA GPUs. Entries are written in Python, and no GPU is needed, because RiftStack runs and tests every submission on its own hardware. The problem set, the tooling and the submission instructions are released at kickoff on October 21, so everyone starts from the same point. Submissions close at 23:59 Anywhere on Earth on October 27.

Emmy lowers PyTorch graphs to CUDA through six inspectable intermediate representations and picks kernel schedules by search rather than fixed heuristics. On Gemma 4 12B, Emmy-generated kernels reached up to 1.6x the throughput of NVIDIA's cuBLASLt library on the RTX 5090. Every optimization a compiler finds still has to be turned into code the GPU can execute, and that last step is the one participants build. Emmy is published under the Apache-2.0 license and available on GitHub.

The hackathon follows last year's event, where standout submissions led to interviews with the engineering team. Entries that stand out this year may again be shortlisted for an interview with RiftStack.

Participants can prepare with NVIDIA's PTX ISA reference and CUDA C++ Programming Guide, and by reading the Emmy source. Registration, the reading list and the Discord invite are at https://riftstack.ai/hackathon.

About RiftStack

RiftStack builds the operating system for sovereign AI deployments: a control plane that turns datacenters into AI cloud offerings, with built-in LLM inference, GPU virtualization, and the Emmy ML compiler. RiftStack is SOC 2 Type II certified and an NVIDIA Inception member.

Media contact: Heiko Polinski, heiko@cloudrift.ai.

All press releases