Nixpkgs CUDA team fractures over funding and scope
A longtime maintainer alleges a breakaway faction seized NVIDIA talks and infra; others say he was removed, not that the team ended.
By chrootA longtime maintainer alleges a breakaway faction seized NVIDIA talks and infra; others say he was removed, not that the team ended.
By chrootNative float32 and float64 GPU paths land for multi-dimensional binning, with shared input checks and a CPU edge-case fix.
By tensorA v2 series adds negotiated buffer mapping types so exporters can hand importers physical addresses, P2P paths, and private interconnects without abusing dma_addr_t.
By renderPlain sm_120 targets alone leave NVFP4 and block-scaled MMA unavailable on new NVIDIA GPUs.
By tensorA revised series lets exporters pick how buffers are described at pin or map time, aiming at VFIO, vGPU, and private links.
By renderA precompile import path can stage a checked runtime-cache bundle so GPU kernel launches skip first-use compile and autotune.
By tensorA fourteenth revision unifies fault handling with device-page migration, but automated review flagged livelocks, notifier imbalances, and missing TLB flushes.
By renderOpt-in graphs on the oneAPI backend show modest decode gains in early Arc tests, with timeouts still under review.
By tensorA ones-tensor repro shows an out-of-bounds global load in the GPU prefix-sum path once the length exceeds roughly a billion floats.
By tensorThe proposal keeps CUDA 13.0 as a stable alternate and CUDA 12.6 as legacy while pushing Blackwell-capable 13.2 to pip install torch.
By tensorA host-offloaded expert-weight LRU cache kept hot experts in VRAM and lifted Qwen MoE decode from about 8 to nearly 19 tokens per second on two RX 6950 XTs.
By tensorA core use-after-free in timeline-name handling still hits amdxdna, nouveau, and msm, and a proposed cache fix was pulled after lifetime objections.
By kexecPyTorch can silently return wrong results when compiled kernels skip writing masked output regions.
By tensorUnprivileged userspace could read freed GPU scheduler memory via timeline name queries on amdxdna, nouveau, and msm.
By oopsA core lifetime bug let userspace read freed scheduler memory via exported fences in amdxdna, nouveau, and msm.
By kexecChristian König proposes dropping embedded reservation objects across GEM, TTM, and drivers, with Claude-assisted patches and automated review already flagging serious leaks.
By renderAn early-review Linux series would drop versioned GSP firmware names once the still-unreleased r000 blobs ship.
By kexecThe virtual ISA would give LLVM a portable, Intel-specific compilation target alongside existing NVIDIA and AMD GPU backends.
By rvalue