Executable programs and extensions

Using XeStrata on Radeon: choosing a distro for RX 7000 and 9000

The AMD GPU's compute target must be included in rocBLAS for your Linux distribution.

Radeon support in XeStrata depends on the gfx code and ROCm1 rocBLAS included with the Linux distribution. RX 7900 (gfx1100), RX 7800/7700 (gfx1101), RX 9060 (gfx1200), and RX 9070 (gfx1201) appear in the tables for both Ubuntu 26.04 and Fedora 44, while RX 6700 (gfx1031) and RX 7600 (gfx1102) are listed for Fedora 44 only. Check the gfx code with xestrata --check, compare it with the distribution table, and then install the package.

Prepare XeStrata on Radeon by matching the GPU2 model with the Linux distribution. The support table below connects each product's gfx architecture code to the distribution's rocBLAS support. Choose and install a package first, then check detection with xestrata --check.

Requirements and key details
  • AMD prefill's dense matrix operations depend on the distribution's rocBLAS and hipBLASLt.
  • RX 6700 and RX 7600 are listed only for Fedora 44; RX 6600/6500/6400 are absent from both distribution tables.
  • After installing a release package, use xestrata --check and xestrata --setup to confirm device detection and a first model response.

The distribution's rocBLAS must provide the Radeon compute target

XeStrata does more than send model input to a Radeon GPU for prefill3. The dense matrix calculations used to process input rely on ROCm rocBLAS and hipBLASLt from the distribution, so the GPU's architecture target, or gfx code, must be included in those libraries. Even with the same RX product name, support can differ between Ubuntu and Fedora depending on the gfx targets included in rocBLAS.

Scroll sideways to compare all columns.

AMD gfx targets and distribution support in XeStrata xe0.1.43.1
Radeon familygfxUbuntu 26.04Fedora 44
RX 7900gfx1100YesYes
RX 7800·7700gfx1101YesYes
RX 6700gfx1031NoYes
RX 7600gfx1102NoYes
RX 9060·9070gfx1200·1201YesYes
RX 6600·6500·6400gfx1032·1034NoNo

This support table is based on the gfx targets in ROCm 7.1 rocBLAS for Ubuntu 26.04 and Fedora 44. Development checks on an actual Radeon used the RX 9060 XT, so treat other listed cards as installation compatibility and measure their speed on your own hardware.

An open handwritten notebook, dual-fan graphics card, and disc case
Check the gfx code and rocBLAS support in addition to the Radeon model name.

Install the package and start with a short request

Once the gfx code matches your distribution, choose a package from the xe0.1.43.1 release and download it with SHA256SUMS. On Ubuntu 26.04, select the recommended package for AMD RDNA2 or later; on Fedora 44, choose a listed compatible package. After verifying the file, install it with sudo apt install ./<downloaded-DEB-file> or sudo dnf install ./<downloaded-RPM-file>. Use the actual downloaded filename rather than guessing a package name that is not in the release table.

After installation, run xestrata --check again to confirm the Radeon is detected, then run xestrata --setup. Choose a model, quantization4, and context length5; the program downloads the model and accepts questions at http://127.0.0.1:8095. Keep the startup window open until you see a first response. The PC may slow briefly while GPU code is prepared, and Qwen3.8-Flash-Next downloads range from about 60 to 110 GB, so the download can take time.

A wide workstation with a laptop, two monitors, and an open desktop computer
After the GPU is detected, a short request checks the execution path.

Read your Radeon gfx code first

After installing the XeStrata release on Linux, run xestrata --check in a terminal to display the detected GPU name and AMD gfx code. For example, if it shows RX 7600 (gfx1102), compare it with the Fedora 44 row. Starting from a GPU name in Windows or another hardware utility can miss the ROCm library support that XeStrata actually sees.

To compare processing time on your PC, keep the model, quantization, and context length the same and reuse one prompt6. Record the GPU and gfx code, distribution, and package name so you can check differences after changing the setup.

A dual-fan graphics card, blank sheets, and a magnifying glass on a desk
Find the row for your gfx code in the support table for your distribution.

If the GPU is missing, check the gfx code and library separately

If Radeon does not appear in xestrata --check, first check the distribution package and GPU driver. If the device name appears but setup stops, compare its gfx code with the support table and check whether xestrata --setup printed an installation command for a missing GPU library or render group. Run the suggested command and repeat setup to check that requirement again.

If the gfx code is supported but startup still fails, confirm that the CPU7 is x86-64 with AVX2 and that Above 4G Decoding8 and Re-Size BAR are enabled in BIOS. If the model has downloaded but the browser does not respond, check the first error in the program terminal and confirm the address is http://127.0.0.1:8095. If the message reports insufficient RAM9, choose a smaller model file or shorter context in setup. A successful launch means receiving a first response in the checked environment; speed or stability on another Radeon needs separate testing.

Terminology notes

  1. ROCm — AMD’s software platform for AI and high-performance computing on GPUs. Support depends on the combination of GPU, operating system, driver, and framework versions.

    Back to the text
  2. GPU — A processor designed to handle many calculations in parallel. It performs model computations during AI inference.

    Back to the text
  3. Prefill — The stage where an LLM reads the input prompt and computes representations for its tokens. Longer prompts contain more tokens to process.

    Back to the text
  4. Quantization — A way to represent model weights or values with fewer bits, reducing storage and memory use. The coarser representation can also affect accuracy.

    Back to the text
  5. 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
  6. 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
  7. 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
  8. Token decoding — The stage that generates successive output tokens using the input and the response produced so far.

    Back to the text
  9. System RAM — System memory that temporarily holds data while programs run.

    Back to the text

Frequently asked questions

What should you check first in this article?
Radeon support in XeStrata depends on the gfx code and ROCm rocBLAS included with the Linux distribution. RX 7900 (gfx1100), RX 7800/7700 (gfx1101), RX 9060 (gfx1200), and RX 9070 (gfx1201) appear in the tables for both Ubuntu 26.04 and Fedora 44, while RX 6700 (gfx1031) and RX 7600 (gfx1102) are listed for Fedora 44 only. Check the gfx code with `xestrata --check`, compare it with the distribution table, and then install the package.