Hip: C++ Heterogeneous-Compute Interface for Portability

2025-05-317:373612github.com

HIP: C++ Heterogeneous-Compute Interface for Portability - ROCm/hip

You can’t perform that action at this time.


Read the original article

Comments

  • By isusmelj 2025-06-035:58

    Is there anything like this also supporting other GPUs? Thinking of Apple Silicon or embedded ones in phones etc.

  • By fursund 2025-06-0212:59

  • By Ygg2 2025-06-0212:064 reply

    What is the relation between this and Zluda[1]?

    [1]https://github.com/vosen/ZLUDA

    • By TheCycoONE 2025-06-0217:371 reply

      Since no one answered you, HIP is a compile time target, you use hipify to transform your cuda code, then you compile it with hipcc, and hopefully everything works; or you target HIP directly.

      ZLUDA was a translation layer, ala wine for cuda applications.

      • By Ygg2 2025-06-0221:43

        Thanks a thousand times over.

    • By krapht 2025-06-0212:291 reply

      HIP is something you can put in production and Zluda is a proof-of-concept?

      • By Ygg2 2025-06-0212:421 reply

        Ok, but it's that their only difference? Zluda was supported by AMD for some time. Is it just that Hip is currently more complete?

        • By 3836293648 2025-06-0218:441 reply

          You using HIP was supported by AMD. Zluda was getting support from AMD. Not the same kind of support

          • By Ygg2 2025-06-0223:33

            Yeah, I get it now. HIP converts CUDA code to HIP code, then converts that code to ROCm (AMD) or PTX(nVidia). Zluda converts CUDA code to PTX, but that CUDA license was changed during 2021 to prevent these kinds of efforts.

            This is why AMD dropped all involvements with Zluda (they were contributing code to it).

    • By spmurrayzzz 2025-06-0215:45

      There's no relation per se (apart from AMD once being involved, before asking the developer to take the project offline).

      It is worth noting however that they accomplish their goals very differently. ZLUDA is now DoA-ware because of nvidia's use constraint in their license regarding translation layers. HIP doesn't have this issue.

    • By almostgotcaught 2025-06-0212:352 reply

      Lol that's like asking what's the difference between the empire state building and the shed in my backyard: one is the runtime that the company that owns/makes the GPUs has been building for 10 years and the other is a (good) solo project.

      • By Ygg2 2025-06-0212:41

        Well, god forbid people ask questions.

HackerNews