6 matching pages, found in 759 ms

Ask StyxWritten on your GPU machine; takes a few seconds the first time.

  1. forums.developer.nvidia.comhttps://forums.developer.nvidia.com › t › deepseek-v4-1-flash-on-4x-gb10… › 389220
    More about forums.developer.nvidia.com
    DeepSeek-V4.1-Flash on 4x GB10: 262K cold input in 27.3s, open-source runtime - DGX Spark / GB10 Projects - NVIDIA Developer Forums

    ForumNov 16, 2026. … 262,145 tokens of source code reached the first content in 27.28 seconds , with no prefix-cache hits. That is about 9,611 input tokens/s including frontend, tokenization and transport . For …

    0 points
  2. forums.developer.nvidia.comhttps://forums.developer.nvidia.com › t › nvfortran-26-9-wrong-code-cras… › 389378
    More about forums.developer.nvidia.com
    Nvfortran 26.9: wrong code / crashes with internal procedures (host association), OpenMP private array sections and MOVE_ALLOC - nvc, nvc++ and nvfortran - NVIDIA Developer Forums

    ForumNov 17, 2026. … a large Fortran CFD code (~130k lines) to nvfortran and OpenACC, we hit several wrong-code bugs that involve internal procedures or MOVE_ALLOC. Each one is reduced below to a small …

    0 points
  3. forums.developer.nvidia.comhttps://forums.developer.nvidia.com › t › nvbufsurface-memory-leak-and-t… › 389372
    More about forums.developer.nvidia.com
    NvBufSurface memory leak and timestamp drift in multi-stream RTSP DeepStream 7.0 pipeline on Jetson Orin Nano - DeepStream SDK - NVIDIA Developer Forums

    ForumNov 17, 2026. … pyservicemaker, The similar pipeline code is pyservicemaker deepstream_test3.py Related topics Topic Replies Views Activity Significant Memory leak when streaming from a clients RTSP source DeepStream SDK gstreamer , nvbugs 11 3020 August …

    0 points
  4. forums.developer.nvidia.comhttps://forums.developer.nvidia.com › t › deepseek-v4-1-flash-on-4x-dgx-… › 389371
    More about forums.developer.nvidia.com
    DeepSeek-V4.1-Flash on 4x DGX Spark GB10 (CX7 ring): full setup + cold prefill & decode numbers up to 1M context - DGX Spark / GB10 Projects - NVIDIA Developer Forums

    ForumNov 17, 2026. … reported ~9.6K @ 262K on source-code input, against ~8.1K here on CJK input. Its 1M figure is a synthetic capacity check rather than a same-corpus prefill sweep, so no comparable long …

    0 points
  5. forums.developer.nvidia.comhttps://forums.developer.nvidia.com › t › nvc-26-9-argument-dependent-lo… › 389494
    More about forums.developer.nvidia.com
    Nvc++26.9: argument-dependent lookup inside a generic lambda misses functions declared after the enclosing template - nvc, nvc++ and nvfortran - NVIDIA Developer Forums

    ForumNov 18, 2026. … comparison; it accepts the code Ubuntu 24.04.4 LTS, x86_64, glibc 2.39 Reproducer // repro.cpp #include <iostream> namespace lib { template <typename T> void call(const T& x) { [&](auto) { f(x); }(0); } } // namespace lib …

    0 points
  6. forums.developer.nvidia.comhttps://forums.developer.nvidia.com › t › nvfortran-26-9-o3-reassociates… › 389434
    More about forums.developer.nvidia.com
    Nvfortran 26.9: -O3 reassociates `(a+b)+c` despite parentheses and -Kieee; reductions vectorized with -Kieee -Mnofprelaxed, `!DIR$ NOVECTOR` ignored - nvc, nvc++ and nvfortran - NVIDIA Developer Forums

    ForumNov 18, 2026. … of the same Fortran source to give bit-identical results. Two value-changing CPU optimizations got in the way. Environment: NVIDIA HPC SDK 26.9 (nvfortran 26.9-0), RHEL 8.10, Xeon Sapphire Rapids …

    0 points

Search for “Source Code site:forums.developer.nvidia.com” on Google, or add !ddg, !w and other shortcuts to any search.