OBS cuts the cord on Kepler GPU NVENC support — version 31.0.0 Beta 1 no longer works with GTX 600 and GTX 700 GPU hardware encoders

News
By
published

If you want to encode on these very old GPUs, you must use an older OBS version.

Nvidia GTX 690
(Image credit: Nvidia)

OBS Studio is dropping support for Nvidia's first generation of GPUs, which featured hardware GPU encoding support. The latest version of the streaming and recording software, 31.0.0 Beta 1, no longer works with GTX 600 and 700 series Kepler GPUs.

The developers behind OBS did not give a reason for dropping Kepler GPU support, but it was inevitable that this would happen at some point. Three years ago, Nvidia dropped driver support for the Kepler-powered GTX 700 series lineup, making software compatibility with the ancient GPU architecture rather pointless for modern-day applications.

The Kepler GPU architecture was Nvidia's first-ever GPU design to integrate an NVENC encoder engine, and it was equipped with a hardware accelerator. Kepler featured Nvidia's first-generation NVENC engine, capable of only encoding in H.264. It wasn't known for being a super high-quality encoder and was often outperformed by software-based encoders that operated on the CPU. It was also very limited in features, only supporting Chroma 4:2:0 and lacking a lossless encoding option. Regardless, it laid the groundwork for future NVENC encoders to compete with software encoders in terms of quality and features.

If you are one of the few people still operating a GTX 600 or 700 series card for encoding purposes, prior versions of OBS Studio still support the NVENC encoder on Kepler-based GPUs.

OBS Studio 31.0.0 Beta 1 adds many new features and changes to the streaming/recording application, including new Nvidia blur, background filters, and first-party YouTube chat features. Nvidia-specific changes include a refactored NVENC implementation with various improvements. SDK 12.2 features such as split encoding, B-Frames references, and Target Quality VBR mode from old SDKs are now supported. This NVENC refactoring could be why Kepler support was stripped, but we can't be sure.

See more GPUs News
Aaron Klotz
Aaron Klotz
Contributing Writer

Aaron Klotz is a contributing writer for Tom’s Hardware, covering news related to computer hardware such as CPUs, and graphics cards.

12 Comments Comment from the forums
  • DS426
    "The Kepler GPU architecture was Nvidia's first-ever GPU design, and..."

    What?? No, the GeForce 256 was nVidia's first GPU. Need to reword as yes, NVENC was introduced with Kepler.
    Reply
  • josmat
    "If you want to encode on these very old GPUs..."

    These young people today, they have an extremely distorted notion of time.
    Reply
  • kyzarvs
    DS426 said:
    "The Kepler GPU architecture was Nvidia's first-ever GPU design, and..."

    What?? No, the GeForce 256 was nVidia's first GPU. Need to reword as yes, NVENC was introduced with Kepler.
    Exactly - my first GPU was a GeForce 2 GTS! A very long time before Kepler.
    Reply
  • Eximo
    josmat said:
    "If you want to encode on these very old GPUs..."

    These young people today, they have an extremely distorted notion of time.

    In terms of GPUs capable of NVENC, these are the oldest. CUDA was only 5 years before that. The beginning of the GPGPU era was in the early 2000s.

    Prior to that you wouldn't even have the notion of running non-display tasks on a GPU.
    Reply
  • AkroZ
    Eximo said:
    In terms of GPUs capable of NVENC, these are the oldest. CUDA was only 5 years before that. The beginning of the GPGPU era was in the early 2000s.

    Prior to that you wouldn't even have the notion of running non-display tasks on a GPU.
    GPGPU existed long before CUDA. Since Vertex Shaders was introduced with the Geforce 3 (2001), the programming pipeline was sometimes used for other tasks as you can read the resulting data instead of displaying it, but this was hacky. Some IT papers were showing good results in using the GPU instead of the CPU for parrallel tasks.
    One day someone published a driver to program nvidia graphic cards, nvidia hired him and some years laters cuda was born.
    Reply
  • Eximo
    AkroZ said:
    GPGPU existed long before CUDA. Since Vertex Shaders was introduced with the Geforce 3 (2001)
    Second sentence mentions it.
    Reply
  • crunchylayer4
    DS426 said:
    "The Kepler GPU architecture was Nvidia's first-ever GPU design, and..."

    What?? No, the GeForce 256 was nVidia's first GPU. Need to reword as yes, NVENC was introduced with Kepler.
    NV1 was their first GPU. SGS Thomson manufactured the chips for them. The Diamond Edge 3D board also functioned as an audio interface and game controller input. It came bundled with Sega Virtua Fighter. Back in the day, I bought one at Fry's Electronics.
    Reply
  • Giroro
    There's probably some really good technical reason behind the scenes for why they had to drop support, but I'm not seeing what that would be.
    My initial reaction was "At least it's not Maxwell, they shouldn't drop support for Maxwell", even though Maxwell is only 2 years newer, and I have no idea why that's where I draw the line. I thought maybe because they still sell the GT 730... which is Fermi/Keppler based and lacks NVENC regardless of what die was thrown in there.

    Then I remembered that the Nintendo Switch uses a Maxwell chip from 2015. That's weird to think about.
    Reply
  • das_stig
    Time for anything but a 900 series to be retired, replacements cheap on eBay that just wipe the floor with them performance and power wise.
    Reply
  • evdjj3j
    josmat said:
    "If you want to encode on these very old GPUs..."

    These young people today, they have an extremely distorted notion of time.
    I thought the same thing when the author described the architecture as ancient.
    Reply