Download

Get PXA

The current release is v3.1 for Linux, as a tarball or a container. The one-file downloads for Linux, Windows, macOS and Android come with 4.0 and are in testing now.

available

Linux x86_64

Engine + PXA Control tarball (CUDA 12.8 runtime inside, sm_60/61/70), Ubuntu 24.04 and a 22.04 variant

v3.1 · tar xzf, then ./pxa

Get it

available

Linux container

ghcr.io/poisonxa16/pxa:v3.1, same binaries, Compose file

v3.1 · docker run --gpus ... ghcr.io/poisonxa16/pxa:v3.1

Get it

4.0

Linux one-file installer

pxa-<version>-linux-x86_64.run: verifies its own sha256, unpacks once, adds a menu entry, opens Control signed in

4.0-pre build (file name still says v3.2pre) · built and tested; ships with 4.0

in testing

Windows x64

PXA Control .exe (tray app) + engine zip

4.0-pre build (file name still says v3.2pre) · Control exe built and smoke-tested; engine + PXQN DLL build in progress; unsigned until release

in testing

macOS (Intel)

PXA Control as one file; drives a rig on your network (no CUDA on a Mac)

4.0-pre build (file name still says v3.2pre) · built and smoke-tested on macOS 13.7 Intel

in testing

Android

PXA Control as an installable PWA and a small APK wrapper for your rig on the LAN

4.0-pre build (file name still says v3.2pre) · emulator-tested on Android 14; needs a WebView from Chrome 86 or newer

Tarball

Two commands

tar xzf pxa-v3.1-linux-x86_64-cuda12.8-sm60_61_70.tar.gz
cd pxa-v3.1-linux-x86_64-cuda12.8-sm60_61_70
./pxa --doctor     # checks your cards, driver and CPU; starts nothing
./pxa              # opens PXA Control

Or the one-line installer, which checks your CPU, glibc and cards, downloads the right tarball and verifies its checksum:

curl -fsSL https://raw.githubusercontent.com/poisonxa16/pxa/main/install.sh | bash

Container

Or Docker

docker run -d --name pxa --gpus '"device=0,1"' --shm-size=1g \
  -p 8080:8080 -v /path/to/models:/models:ro \
  -v pxa-cache:/work/.cache/pxa -e PXA_CACHE_DIR=/work/.cache/pxa \
  ghcr.io/poisonxa16/pxa:v3.1

Hosts where --gpus fails with an ldcache error (Unraid among them): use --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=0,1.

Requirements

What you need

Cards	Tesla P100 (sm_60) or V100 (sm_70), tuned and measured; GTX 1080 Ti and other sm_61 cards work. RTX 20-series and newer are not in this build.
Driver	NVIDIA 570 or newer; the package brings its own CUDA 12.8 runtime.
System	Linux x86-64, glibc 2.38 (Ubuntu 24.04+) or the 22.04 variant; any x86-64 CPU (no AVX2 needed).
Disk	About 5 GB for the package, plus your models.
RAM	Plan 64 GB of system RAM for the 32 GB Flash-Next file on one card.

Models

Start with a PXA quant

One 16 GB card and a 27B model: the one-card mix or PXQN3bal. Two 16 GB cards: PXQN4, or PXQN5 for the best quality. Standard GGUF files run too.

All on huggingface.co/poisonxa; who may download each file is set on its model card.