4 ms·

Hm, isn't NodeJS not suitable for heavyweight CPU tasks ?

aastronaut 3y ago
> [...], written in TypeScript and accelerated with C++ bindings to BLAS and LAPACK C++ addons for Node.js are not part of the JavaScript runtime.
- revskill 3y ago
  But it still blocks event loop ?
  - byteofbits 3y ago
    Native bindings can use libuv to offboard work to other threads and then re trigger the JS execution when it completes. This is how a lot of the native node libraries work under the hood to allow parallel IO operations! It’s also an incredibly powerful performance optimisation tool in more complex, scale out NodeJS deployments.
    - revskill 3y ago
      So in theory, if we ported python libraries into JS, the performance could be better than Python ?
      - byteofbits 3y ago
        At work we do some pretty significant parallelised computation (using Node and Rust) and that has equivalent performance to a relevant python based library. There are nuances to how data enters and exits the VM runtime which means I couldn’t say with confidence if performance would be better or worse in the general case - but either direction is at least technically viable.
        NickChasanis 3y ago
        i was thinking about that, seems like a good deal overall with limited overhead.
      - pjmlp 3y ago
        Python bindings for native libraries would still be the same, they are the same native libraries, while pure Python code would speed up thanks to JavaScript runtimes having JITs.
pjmlp 3y ago
Just like Python, but at least has a JIT on the box.
- huac 3y ago
  Arguably better async primitives too. Python's main advantage at this point feels like a stronger ecosystem of packages calling into native code (eg numpy et al mostly call optimized C routines). If Typescript adds more of those libraries (like this one), then I could see it gaining a lot of adoption.
  - sroussey 3y ago
    I think Bun (which has typescript embedded but written in zig) would be a great place to experiment. I could see it adding some tensor types that drop into zig code that uses webgpu bindings. And add some operator overloading so you could just add t1 + t2 etc.

> [...], written in TypeScript and accelerated with C++ bindings to BLAS and LAPACK C++ addons for Node.js are not part of the JavaScript runtime.

revskill 3y ago
But it still blocks event loop ?
- byteofbits 3y ago
  Native bindings can use libuv to offboard work to other threads and then re trigger the JS execution when it completes. This is how a lot of the native node libraries work under the hood to allow parallel IO operations! It’s also an incredibly powerful performance optimisation tool in more complex, scale out NodeJS deployments.
  - revskill 3y ago
    So in theory, if we ported python libraries into JS, the performance could be better than Python ?
    - byteofbits 3y ago
      At work we do some pretty significant parallelised computation (using Node and Rust) and that has equivalent performance to a relevant python based library. There are nuances to how data enters and exits the VM runtime which means I couldn’t say with confidence if performance would be better or worse in the general case - but either direction is at least technically viable.
      - NickChasanis 3y ago
        i was thinking about that, seems like a good deal overall with limited overhead.
    - pjmlp 3y ago
      Python bindings for native libraries would still be the same, they are the same native libraries, while pure Python code would speed up thanks to JavaScript runtimes having JITs.

pjmlp 3y ago

Just like Python, but at least has a JIT on the box.

huac 3y ago
Arguably better async primitives too. Python's main advantage at this point feels like a stronger ecosystem of packages calling into native code (eg numpy et al mostly call optimized C routines). If Typescript adds more of those libraries (like this one), then I could see it gaining a lot of adoption.
- sroussey 3y ago
  I think Bun (which has typescript embedded but written in zig) would be a great place to experiment. I could see it adding some tensor types that drop into zig code that uses webgpu bindings. And add some operator overloading so you could just add t1 + t2 etc.