Skip to content

Releases: TheLitis/xVRAM

Release list

xVRAM Developer Preview 1 — explicit CUDA workloads beyond VRAM

xVRAM Developer Preview 1 — explicit CUDA workloads beyond VRAM Pre-release
Pre-release

Choose a tag to compare

View all tags
@github-actions github-actions released this 11 Sep 19:25
preview-2026.09.11.1
c95513c

xVRAM Developer Preview 1

A Windows x64 distribution of the existing explicit CUDA/cuBLAS integration,
with a public-API C++ example and a bounded, evidence-producing demo workflow.

Included

  • Release SDK and existing CLI benchmarks, app-local pinned cuBLAS/nvCOMP libraries,
    public headers, CMake targets, licenses and notices.
  • xvram-preview-gemm: small standalone FP32 example, full CPU FP64 reference,
    explicit error/cleanup handling, using only installed public C API headers.
  • preview.py verify, smoke, oversubscribe, validate: integrity checks,
    real isolated runs and the established Phase 6a report/trace validators.
  • New result directory for every run, provenance and file hashes, stdout/stderr,
    raw JSON/trace and offline HTML. Missing/failed GPU runs never count as passes.
  • Documented Astra boundary at 98e41acc, evidence limitations, conceptual gallery
    asset and Product Hunt submission kit in docs/preview.

Download the Windows ZIP and SHA256SUMS.txt, then read its README.md.
Python 3.10+ with the pinned validator dependency, a compatible NVIDIA GPU/driver,
Visual C++ v14 x64 runtime and sufficient available RAM are required. No OpenAI API
key or automatic telemetry upload is involved.

Validation boundary

Publication requires the existing Windows/Linux CI matrix and separate preview
unit, build, installed-example and packaging checks on the same source revision.
CI has no GPU: this is not fresh RTX 3070 hardware certification. Historical
maintainer results are linked and labeled. Record a new run of this exact package
before presenting new measurements. Unit-test fixtures are not hardware evidence.

Limitations

The production residency architecture, ABI and frozen schemas are unchanged.
This profile is explicit, synchronous FP32 integration with compression disabled:
not unchanged-executable interception, arbitrary kernels, general PyTorch/LLM
support, training or a universal speedup. Phase 6b proof gates remain unproven.
The SDK stays 0.1.0-dev and is not for production workloads.

xVRAM is Apache-2.0. Bundled components retain their own terms. Publishing this
release is not a Product Hunt submission or an assertion of organizer approval.

Source: c95513c58c0b8375ba58b67864068882462ee70a

CI: https://github.com/TheLitis/xVRAM/actions/runs/34637159922

Package checks: https://github.com/TheLitis/xVRAM/actions/runs/34638080265

Loading