What is persistent-rnn?

pineking/persistent-rnn — explained in plain English

Analysis updated 2026-07-18 · repo last pushed 2016-06-20

C++Audience · developerComplexity · 5/5DormantSetup · hard

Mindmap

mindmap
  root((repo))
    What it does
      Speeds up RNNs on GPUs
      Caches weights in registers
      Targets small batch sizes
    Tech stack
      C++
      CUDA
      NVIDIA GPUs
    Use cases
      Real-time inference servers
      Robotics sensor processing
      Low-latency sequence models
    Audience
      CUDA developers
      ML systems engineers
    Constraints
      Basic RNNs only
      Limited GPU models

mindmap root((repo)) What it does Speeds up RNNs on GPUs Caches weights in registers Targets small batch sizes Tech stack C++ CUDA NVIDIA GPUs Use cases Real-time inference servers Robotics sensor processing Low-latency sequence models Audience CUDA developers ML systems engineers Constraints Basic RNNs only Limited GPU models

What do people build with it?

USE CASE 1

Speed up a real-time inference server that processes RNN requests one at a time

USE CASE 2

Run sequence models on a robotics system processing live sensor streams with low latency

USE CASE 3

Replace a standard cuDNN RNN call with this library for small-batch, latency-sensitive workloads

USE CASE 4

Benchmark GPU register-caching techniques against standard RNN libraries

What is it built with?

C++CUDAcuDNN-style API

How does it compare?

	pineking/persistent-rnn	achanana/mavsdk	alange/llama.cpp
Stars	—	—	0
Language	C++	C++	C++
Last pushed	2016-06-20	2024-05-20	—
Maintenance	Dormant	Dormant	—
Setup difficulty	hard	moderate	moderate
Complexity	5/5	4/5	4/5
Audience	developer	developer	developer

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · hard Time to first run · 1day+

Only works on specific older NVIDIA GPUs (TitanX, GTX 1080, GP100) and supports basic RNNs, not LSTMs.

Copy-paste prompts

Prompt 1

Explain how this library caches RNN weights in GPU register files instead of reloading them from memory each step.

Prompt 2

Help me integrate this library's cuDNN-style API into my existing CUDA RNN inference code.

Prompt 3

Walk me through the supported GPU models and batch-size limits before I try to use this in production.

Prompt 4

Show me why this library is much faster than standard RNN libraries specifically for small batch sizes.

Frequently asked questions