Repository navigation
Releases: etcofficials/GPUFusion
Release list
GPU Fusion 1.1.0
Per-application GPU selection that says what it can and cannot do on your PC — and checks it.
What's new
- Choose a GPU per application by name: Automatic, the integrated GPU (Intel), the discrete GPU (NVIDIA) or Both. Each choice shows whether Windows can enforce it on this PC (verified, expected, cannot be enforced, unknown) with the evidence. A Windows value is written only where the probe verification shows it selects that GPU, and it is read back after saving. The raw Windows values remain under Advanced.
- Verify on this PC (
--verify-app <exe>on the command line): GPU Fusion starts the program, follows every process it starts, and reports in four stages whether the requested GPU actually did the rendering — counting only 3D/compute engine time of the program's own processes, never total GPU load, video or copy engines, or other programs. If a launcher starts the real renderer, GPU Fusion names it and offers to configure that executable. - Dual-GPU check for a program: which kinds of multi-GPU use are possible on this PC, from the hardware and the graphics APIs the program links. No setting can make an existing single-GPU program use two GPUs, and GPU Fusion says so.
- The GPUs page shows how Windows classifies each adapter (boot device, hybrid flags) and which one drives the primary display.
What was found on the reference PC (GeForce GT 730 + Intel HD Graphics 2000)
Windows ranks the GT 730 first for every preference — it is the boot device and drives the only display, neither GPU has a hybrid classification, and the Intel driver is WDDM 1.2. NVIDIA selection works (verified with real rendering); Intel selection cannot be enforced by Windows (with "Power saving" set, rendering stays on the GT 730 — measured, not assumed). Programs that let you choose their GPU themselves can still use the Intel GPU, and GPU Fusion verifies that too. Details: docs/GPU_SELECTION.md.
Fixes
- GPU choices match saved verification results by GPU model, not LUID (Windows assigns new LUIDs at every boot).
Download
- GPUFusion-Windows-x64.zip — portable · GPUFusion-Setup.exe — per-user installer · SHA256SUMS.txt —
sha256sum -c SHA256SUMS.txt
MIT License. Qt 6 is used under the LGPLv3 (notices and license texts are in the package).
GPU Fusion 1.0.1
Licensing and packaging release — the application itself is unchanged from 1.0.0.
Changes
- GPU Fusion is now released under the MIT License; the ZIP and the installer include
LICENSE.txt. SHA256SUMS.txtnow uses LF line endings, sosha256sum -c SHA256SUMS.txtworks on every platform.
Download
- GPUFusion-Windows-x64.zip — portable: extract and run
GPUFusion\GPUFusion.exe(data stays next to it). - GPUFusion-Setup.exe — per-user installer, no administrator rights (Start-menu entry, optional desktop shortcut, uninstaller).
- SHA256SUMS.txt — verify with
sha256sum -c SHA256SUMS.txtorGet-FileHash -Algorithm SHA256 <file>.
Requires Windows 10/11 64-bit and a DirectX 11 graphics adapter. For what GPU Fusion does, see the 1.0.0 release and the README.
Qt 6 is used under the LGPLv3 (notices and license texts are in the package).
GPU Fusion 1.0.0
First release of GPU Fusion — a native Windows (C++20, Qt 6) app that shows what your graphics adapters really do, manages the Windows per-app GPU preference safely, and measures whether two GPUs can work together.
Download
- GPUFusion-Windows-x64.zip — portable: extract and run
GPUFusion\GPUFusion.exe(data stays next to it). - GPUFusion-Setup.exe — per-user installer, no administrator rights (Start-menu entry, optional desktop shortcut, uninstaller).
- SHA256SUMS.txt — verify with
Get-FileHash -Algorithm SHA256 <file>.
Requires Windows 10/11 64-bit and a DirectX 11 graphics adapter.
Highlights
- Live monitoring of every physical GPU, including WDDM 1.x GPUs that Task Manager cannot show; unsupported metrics are labelled, never shown as 0.
- Per-application GPU usage; GPU preference changes are journalled, undoable and verified with a probe.
- Diagnostics with a privacy-safe exportable report.
- Benchmarks with warm-up, repeated runs, bit-exact output checks and noise-aware comparisons.
- Workload Lab (measured on a GeForce GT 730 + Intel HD Graphics 2000): a compute workload split by measured speed runs 1.14–1.19× faster than the GT 730 alone; decoding one video stream per GPU is 2.2× slower than decoding both on the Intel GPU — reported as measured.
See CHANGELOG.md, docs/MULTI_GPU_RESEARCH.md and docs/ACCEPTANCE.md.
Qt 6 is used under the LGPLv3 (notices and license texts are in the package). GPU Fusion itself is released under the MIT License.