leonarski_fandClaude Opus 5
4e6600a96c
CMAKE_CUDA_ARCHITECTURES had no sm_70 entry, and PTX only ever JIT-compiles forwards, so a V100 had no runnable code in the fatbin at all - every kernel launch failed with "no kernel image is available for execution on the device". Append 70 only for a CUDA 12 toolkit: CUDA 13 removed offline compilation for Volta, so an unconditional entry would break the RHEL 9, Ubuntu and Windows builds. 12.8/12.9 still emit it but warn on every .cu, hence -Wno-deprecated-gpu-targets. The append goes after ENABLE_LANGUAGE(CUDA), where the nvcc version is known, matching the existing sm_121 handling. Verified: all 15 CUDA sources compile for sm_70 (including ffbidx, which already guards on __CUDA_ARCH__ >= 700/800), and cuobjdump shows an sm_70 cubin in the built rugnux binary. Consequence worth documenting: a V100 can only run the artefacts built with CUDA 12 - the RHEL 8 packages and the portable Linux .tgz. Document that alongside the minimum NVIDIA driver of every released artefact (525.60.13 for CUDA 12, 580.65.06 for CUDA 13), which applies because the CUDA runtime is linked statically and cuFFT is bundled, so the driver is the only NVIDIA component the target host must supply. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
9.1 KiB
Release contents
This page describes what a Jungfraujoch release ships and what each artefact needs on the target machine — which CPU instruction set the binaries were compiled for, which CUDA toolkit they were built against, and which runtime libraries are bundled rather than expected from the host.
The artefacts in the table below are built and published by the continuous-integration pipeline
(.gitea/workflows/build_and_test.yml) when a tag is pushed. For how to install and configure the
result see Deployment; for the package-repository URLs see
Linux package repositories.
Artefacts
| Artefact | Distributed via | Contains |
|---|---|---|
.rpm / .deb packages |
package repositories | The full server stack: jfjoch (broker, frontend, FPGA and detector tools), jfjoch-writer, jfjoch-viewer (incl. the XDS plugin), jfjoch-driver-dkms |
jfjoch_viewer-<version>-linux-cuda<major>.tgz, ...-linux-cpu.tgz |
Gitea release page | Portable Linux viewer package: jfjoch_viewer, rugnux, jfjoch_extract_hkl, jfjoch_recompress and the license notices |
jfjoch-viewer-<version>-win64-cuda<major>.exe, ...-win64-cpu.exe |
Gitea release page | Windows installer with the same four programs, plus the Qt runtime |
jfjoch-writer .rpm / .deb |
Gitea release page | The writer alone, for a file-writing machine without the rest of the stack |
libjfjoch_xds_plugin.so.<version> |
Gitea release page | XDS HDF5 read plugin (built on RHEL 8); see Integration with MX software |
jfjoch-client |
PyPI and the Gitea PyPI index | Generated Python OpenAPI client |
| Documentation | Read the Docs and the gitea-pages branch |
This documentation set |
The FPGA firmware (.mcs) images are attached to the release as well. The firmware is stable and is
carried from version to version, and is rebuilt with Vivado (see FPGA smartNIC) when it
needs to change — so a card keeps its image across a software upgrade unless the release notes say
otherwise.
CPU instruction set
The architecture flags live in the CI configuration rather than in CMakeLists.txt, so a site
building from source picks its own (x86-64-v4 on an AVX-512 cluster, -march=native, or the plain
baseline the compiler defaults to). The released binaries are compiled to a fixed floor:
| Release | Flags | Minimum CPU |
|---|---|---|
Linux (all packages, and the portable .tgz) |
-march=x86-64-v3 -flto=auto |
AVX2 + FMA + BMI2 — Intel Haswell (2013) / AMD Zen (2017) and newer |
| Windows installer | /arch:AVX |
AVX — Intel Sandy Bridge (2011) / AMD Bulldozer and newer |
The Windows floor is lower because MSVC has no spelling for the x86-64-v2 level; /arch:AVX is
the nearest one and implies SSE4.1/4.2, which is what actually matters — without it Eigen has no
vectorised round and falls back to a libm call per element. Link-time optimisation is applied on
Linux only.
A binary will fault with an illegal instruction on a CPU below its floor. If you must run on older hardware, build from source without the flags.
Operating-system floor
The .rpm / .deb packages are built per distribution (RHEL/Rocky 8 and 9, Ubuntu 22.04 and 24.04)
and are tied to it. The portable viewer .tgz is built on RHEL 8, the oldest supported
distribution, so its glibc floor is low enough to run on any newer Linux — that is what it is for,
and why it replaces the per-distro packaging of the viewer on the release page. The Windows
installer is built and verified on Windows 11.
CUDA and non-CUDA builds
Every binary artefact is released in two variants, cuda<major> and cpu. The CUDA variant adds
the GPU fast-feedback indexer (ffbidx), the GPU FFT indexer and GPU image processing; the CPU-only
variant runs the same pipeline on the CPU with the FFTW indexer, at much lower throughput.
The CUDA toolkit used is the one on the corresponding build machine: CUDA 12 for the RHEL 8 packages, CUDA 13 for RHEL 9, Ubuntu and Windows. The major version is part of the artefact and repository name, so a download is self-identifying. Building from source needs CUDA 12.8 or newer.
A CUDA build does not require a CUDA machine. Of the CUDA components only cuFFT is linked dynamically — the CUDA runtime and the fast-feedback indexer are linked statically — and cuFFT itself has no link-time dependency on the NVIDIA driver library. Jungfraujoch asks how many CUDA devices are present at start-up and treats "none" (including "no driver installed") as zero GPUs, falling back to the CPU path. So a CUDA build starts and runs correctly on a machine with no NVIDIA GPU at all, provided the cuFFT runtime can be loaded:
- Portable
.tgzand Windows installer — cuFFT is part of the distribution, shipped next to the executable (on Linux found through an$ORIGINrpath). Nothing else is needed: no CUDA toolkit, and on a GPU machine only the NVIDIA driver. .rpm/.deb— cuFFT comes from the distribution's own CUDA packages, so that one dependency is managed centrally with the rest of CUDA. Install the cuFFT package alongside, or use thenocudarepositories on a machine where CUDA is not wanted.
The cuFFT runtime is large (the Windows DLL is ~256 MB), so the CUDA artefacts are correspondingly bigger than the CPU ones — the other reason for shipping both.
On a machine with an NVIDIA GPU, take the CUDA variant: only that one uses the GPU.
GPU generations and the NVIDIA driver
A CUDA variant carries compiled device code for a fixed set of GPU generations, and which generations those are follows from the CUDA toolkit it was built with. The CUDA runtime is linked statically, so the only NVIDIA component the target machine has to supply is the driver — there is no CUDA-toolkit version requirement on the host.
| Artefact | CUDA toolkit | GPU generations | Minimum driver |
|---|---|---|---|
RHEL 8 packages, portable Linux .tgz |
12.9 | Volta (V100) through Blackwell: sm_70, 75, 80, 86, 89, 90, 100, 120, 121 |
525.60.13 |
| RHEL 9 and Ubuntu packages, Windows installer | 13.x | Turing (T4) through Blackwell: the same list without sm_70 |
580.65.06 (Linux), R580 (Windows) |
any cpu / nocuda variant |
— | — | none |
A V100 needs the CUDA 12 build. CUDA 13 dropped offline compilation for Volta, and the PTX a
fatbin also carries only ever JIT-compiles forwards, so a CUDA 13 artefact contains nothing a V100
can execute: every kernel launch fails with no kernel image is available for execution on the
device. On a V100 host take the RHEL 8 packages or the portable Linux .tgz. Nothing older than
Volta is supported.
Newer GPUs never need a newer build — the highest generation in the list ships PTX as well as SASS, which the driver JIT-compiles for a GPU that came out after the release.
The minimum driver above is the floor for the whole CUDA major version, which is what applies here
because the CUDA runtime is statically linked
(CUDA minor version compatibility).
Newer drivers are always fine; they are backward compatible. A driver from the same release as the
build toolkit (575.57.08 for the CUDA 12.9 build, 610.43.02 for a CUDA 13.3 one) additionally rules
out the single caveat of minor version compatibility — a call into a driver API newer than the
installed driver, which fails with cudaErrorCallRequiresNewerDriver.
Windows installer
The Windows artefact covers jfjoch_viewer and the portable analysis CLIs only; the rest of
Jungfraujoch (broker, receiver, FPGA host, detector control) is Linux-only.
The toolchain bounds of the released installer are:
- Visual Studio 2026 with the C++ (MSVC) toolset. MSVC is not optional — CUDA on Windows builds through it — and it is what the release is compiled with.
- CUDA Toolkit 13.3 for the
cuda13variant. - Qt 6.11 for MSVC (
msvc2022_64), including Qt Charts. - Ninja as the generator; zlib and Eigen 3.4 supplied from a build prefix.
The installer is generated with NSIS and bundles the Qt runtime (via windeployqt) and, on the
CUDA variant, the cuFFT DLL — so the end user installs neither Qt nor a CUDA toolkit. The two
variants share an install directory and Start Menu group and replace each other (CUDA is a strict
superset); they are told apart by the installer filename and the Add/Remove Programs entry:
| Build | Installer file | Add/Remove Programs |
|---|---|---|
| CUDA (default) | jfjoch-viewer-<version>-win64-cuda<major>.exe |
Jungfraujoch (CUDA) |
| CPU-only | jfjoch-viewer-<version>-win64-cpu.exe |
Jungfraujoch (CPU) |
To build the viewer yourself on Windows, see jfjoch_viewer ▸ Building from source on Windows.
Licenses
Every package variant carries the project license, the third-party manifest and the verbatim
license texts of the bundled dependencies under share/doc/jfjoch. See
Third-party software notices.