Compute

Along with our usual gaming results, we’ll also take a cursory look at compute performance for the GTX 1650 Super. This is one area where NVIDIA has developed an unusual advantage, at least for the moment, as AMD’s OpenCL driver for the Radeon RX 5000 series cards is currently broken and unfit for production use. I don’t expect anyone to be using a GTX 1650 Super for any serious compute work – generally if you need GPU compute, you’re after higher-end GPUs to really push performance – but if you do find yourself buying a $160 card for compute purposes, among modern cards NVIDIA is currently the only game in town for both OpenCL and CUDA.

Compute: LuxMark 3.1 - Hotel

Compute: CompuBench 2.0 - Level Set Segmentation 256

Compute: CompuBench 2.0 - N-Body Simulation 1024K

Compute: CompuBench 2.0 - Optical Flow

Compute: V-Ray Next Benchmark (CUDA)

Compute: Folding @ Home Single Precision

The Division 2, Grant Theft Auto V, & Forza Horizon 4 Synthetics
Comments Locked

67 Comments

View All Comments

  • Ryan Smith - Monday, December 23, 2019 - link

    Yeah, it's a ratio test, and both scores fluctuate depending on things like memory bandwidth and fill rates. In this case lower bandwidth cards tend to do better, since they aren't as likely to be bottlenecked elsewhere (whereas the 2080 Ti has bandwidth to spare for days).

    It's imperfect, to say the least. But people have been asking for the data, so here it is.
  • sheh - Monday, December 23, 2019 - link

    That's strange.

    I thought, maybe, faster cards don't bother compressing since they don't need it and it uses more power. But other than that, I thought it's just a question of the supported algorithms.
  • harobikes333 - Sunday, December 22, 2019 - link

    Considering these current GPUs seem pretty darn similar. My pick between NVIDIA & AMD would the AMD card simply for the fact that NVIDIA needs competition in the future.
  • sheh - Monday, December 23, 2019 - link

    That's strange.

    I thought, maybe, faster cards don't bother compressing since they don't need it and it uses more power. But other than that, I thought it's just a question of the supported algorithms.
  • sheh - Monday, December 23, 2019 - link

    (Ignore the above. AnandTech's commenting system bugs...)
  • sharathc - Wednesday, December 25, 2019 - link

    See the pic from a distance, it looks like owl 🦉
  • jmunjr - Monday, January 6, 2020 - link

    One good thing about the SUPER variant of the 1650 is it adds the Turing version of NVENC which will boost performance for live streaming and Plex transcoding. The base 1650 used the Volta NVENC for some reason.

Log in

Don't have an account? Sign up now