What should you check first in this article?

XeStrata runs on x86-64 Linux PCs with Ubuntu 26.04 or Fedora 44 and an Intel Arc, RDNA2-or-later Radeon, or Pascal-or-later NVIDIA GPU. Install the package for your distribution, then run `xestrata --setup`; it checks the GPU and RAM, asks about the model and context, and opens a local chat page. Qwen3.8-Flash-Next needs about 60–110 GB for model files plus about 6 GB for the MTP file, so check available RAM and NVMe storage first.

Executable programs and extensions

What is XeStrata? Running Qwen3.8 on Intel, AMD, or NVIDIA

XeStrata lets you run Qwen3.8-Flash-Next locally across GPUs from different vendors.

XeStrata runs on x86-64 Linux PCs with Ubuntu 26.04 or Fedora 44 and an Intel Arc, RDNA2-or-later Radeon, or Pascal-or-later NVIDIA GPU¹. Install the package for your distribution, then run xestrata --setup; it checks the GPU and RAM², asks about the model and context, and opens a local chat page. Qwen3.8-Flash-Next needs about 60–110 GB for model files plus about 6 GB for the MTP³ file, so check available RAM and NVMe⁴ storage first.

XeStrata is a Linux program for running Qwen3.8-Flash-Next locally. Choose a supported Intel Arc, AMD Radeon, or NVIDIA GPU, install the runtime⁵, and use its browser chat interface to send prompts⁶ and receive answers on your PC.

It is an option when you want to try a server-sized model on a personal PC. Before installing, match the Linux distribution and check both the selected model file's RAM requirement and storage size, not just the GPU name. That helps avoid stopping the first run because of a large download or insufficient memory.

Contents

Requirements and key details

Official packages are DEB for Ubuntu 26.04 and RPM for Fedora 44; other distributions require a source installation.
Matching the GPU list is not enough. Depending on the Qwen3.8-Flash-Next file, plan for about 32–62 GB of RAM and 60–110 GB of model storage.
After installation, run xestrata --setup to download a model; it opens http://127.0.0.1:8095 in your default browser.

XeStrata runs a model using your GPU and RAM

XeStrata is a local inference⁷ tool that runs language models such as Qwen3.8 on a Linux PC and returns answers to prompts. The GPU handles part of the model computation, while RAM supplies expert weights that are not selected. Check both VRAM⁸ and the RAM needed for the chosen model files and context length⁹.

Start by choosing a GPU and model file. XeStrata recommends at least 12 GB of VRAM, though it may run more slowly with less. Qwen3.8-Flash-Next also needs RAM for its expert weights. IQ2_XS is about 68 GB on disk and calls for about 48 GB of RAM; IQ3_XXS is about 76 GB and about 60 GB of RAM.

A blank monitor, desktop tower, and RAM modules on a workbench — XeStrata computes on the GPU and reads the remaining expert weights from RAM.

Match the Linux distribution and GPU path before installing

Prebuilt packages are provided for Ubuntu 26.04 and Fedora 44. Windows is unsupported, and WSL 2 may work. Other Linux distributions need a source installation. The PC needs an x86-64 CPU¹⁰ with AVX2. For a discrete GPU, the project calls for Above 4G Decoding and Re-Size BAR enabled and CSM disabled in BIOS.

Intel support covers Arc A and B series, AMD covers RDNA2 and later, and NVIDIA covers Pascal and later. AMD also requires a supported GPU architecture and distribution rocBLAS library. RX 6700 and RX 7600 are listed only for Fedora 44; RX 6600 is not listed for either distribution. After installation, compare the GPU and gfx code from xestrata --check with this table.

XeStrata AMD gfx and distribution matrix

A dual-fan graphics card, RAM modules, an NVMe SSD, and an open blank notebook — Before installing, check the memory and storage space needed by the GPU and model.

Install the package and get a first response

Download a package matching your GPU and distribution from XeStrata xe0.1.43.1, along with SHA256SUMS. Ubuntu 26.04 and Fedora 44 commands are separate below; choose the CUDA¹¹ variant for the GPU generation. Intel and AMD can run without CUDA GPUs, although the distribution package names include a CUDA variant. NVIDIA Pascal and Volta need the Ubuntu CUDA 12.4 package; Fedora has no binary package for them, so use the source-install path.

Verify downloaded package checksums

sha256sum -c SHA256SUMS --ignore-missing

Ubuntu 26.04: install the CUDA 13.1 package

sudo apt install ./xestrata-contrib-cuda13.1_0.1.43-2_amd64.deb

Fedora 44: install the CUDA 13.4 package

sudo dnf install ./xestrata-contrib-cuda13.4-0.1.43-2.x86_64.rpm

Ubuntu 26.04: CUDA 12.4 package for NVIDIA Pascal/Volta

sudo apt install ./xestrata-contrib-cuda12.4_0.1.43-2_amd64.deb

After installation, run xestrata --setup. It checks the GPU, RAM, and CPU, then asks for a model, quantized file, and context length. Accept the recommendation or choose a file and shorter context that fit your available memory. The Qwen3.8-Flash-Next IQ2_XS model file is about 68 GB and the MTP file is separate, so leave room on the NVMe SSD¹². Afterward, xestrata --check shows the detected GPU and AMD gfx code.

After the download and initial GPU code preparation, the default browser opens http://127.0.0.1:8095. Send a short question; receiving an answer confirms the first run. The initial launch reads model data into RAM and prepares GPU code, so the PC may respond slowly for one to three minutes. Keep the terminal open until setup finishes.

Choose a model and start the first run

xestrata --setup

xe0.1.43.1 release and packages

An open PC with two graphics cards beside a box, notebook, and screwdriver — Once the model is ready, use a short question in the local web interface to check the first response.

If the first run stops, check the device and setup log

If xestrata --setup stops because it cannot find a GPU library or the render group, run the installation command printed in the terminal and then run xestrata --setup again. For an AMD card, check that the gfx code shown by xestrata --check appears in the rocBLAS table for your distribution. If checksum verification fails, confirm that you selected the right package and downloaded it completely, then compare it with the release's SHA256SUMS.

If the model downloads but the browser does not respond, check the terminal error and logs in ~/.local/state/xestrata, and confirm the address is http://127.0.0.1:8095. If the message reports insufficient RAM, reopen setup and choose a smaller quantized file or shorter context. If the first answer arrives with your usual applications open, the PC has usable headroom. If RAM remains insufficient, use a PC with more RAM or choose a smaller model.

Terminology notes

GPU — A processor designed to handle many calculations in parallel. It performs model computations during AI inference.
Back to the text
System RAM — System memory that temporarily holds data while programs run.
Back to the text
MTP — A training objective or model component for predicting multiple future tokens at once. In speculative decoding, those predictions can be proposed as candidate tokens.
Back to the text
NVMe — A storage protocol used by SSDs communicating over PCI Express.
Back to the text
Runtime — The software environment that provides facilities needed while a program runs. In local AI, it can also refer to a model execution engine.
Back to the text
Prompt — The input text sent to a model. For generative models it describes the desired result and strongly shapes the output.
Back to the text
Inference — The process of using a trained model to compute an output for an input. Here, local inference means running the model on the user’s device.
Back to the text
VRAM — Memory used by a graphics card’s GPU to store model weights and intermediate values.
Back to the text
Context window — The token span of input and generated content a model can handle in one request. The supported limit and memory use depend on the model and runtime settings.
Back to the text
CPU — The central processor that runs the operating system and general-purpose instructions; in AI workloads it also handles tasks such as preprocessing and data movement.
Back to the text
CUDA — A software platform for general-purpose computing on NVIDIA GPUs.
Back to the text
SSD — A data storage device that uses flash memory and retains data without power.
Back to the text

Frequently asked questions

What should you check first in this article?: XeStrata runs on x86-64 Linux PCs with Ubuntu 26.04 or Fedora 44 and an Intel Arc, RDNA2-or-later Radeon, or Pascal-or-later NVIDIA GPU. Install the package for your distribution, then run `xestrata --setup`; it checks the GPU and RAM, asks about the model and context, and opens a local chat page. Qwen3.8-Flash-Next needs about 60–110 GB for model files plus about 6 GB for the MTP file, so check available RAM and NVMe storage first.
How much memory does it need?: Start by choosing a GPU and model file. XeStrata recommends at least 12 GB of VRAM, though it may run more slowly with less. Qwen3.8-Flash-Next also needs RAM for its expert weights. IQ2_XS is about 68 GB on disk and calls for about 48 GB of RAM; IQ3_XXS is about 76 GB and about 60 GB of RAM.
When is it better not to choose this?: Start by choosing a GPU and model file. XeStrata recommends at least 12 GB of VRAM, though it may run more slowly with less. Qwen3.8-Flash-Next also needs RAM for its expert weights. IQ2_XS is about 68 GB on disk and calls for about 48 GB of RAM; IQ3_XXS is about 76 GB and about 60 GB of RAM.