dob
November 13, 2020, 10:15pm
1
Questions are frequently asked about what GPUs one has to have to transcode video effectively on the Livepeer network. The short answer is that any NVIDIA GPU with NVENC/NVDEC chips listed on their support matrix should theoretically work. But to make it more concrete, there’s a new crowd source list at this URL:
# GPU Support in Livepeer
Livepeer enables node operators to transcode video on GPUs while concurrently mining cryptocurrencies and performing other CUDA operations such as machine learning. As there is a very wide range of GPU hardware out there, this document aims to crowdsource a list of specific GPU models that have been tested and verified to work on Livepeer. If you've tested an additional model, please submit an update to this document.
**Overview**
* Livepeer supports transcoding on NVIDIA GPUs with NVENC/NVDEC. Any GPU [listed here](https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new) with those chips should theoretically work. Note that different models may be subject to different session limits that restrict the amount of video that can be transcoded.
* See [this document which lists tested driver versions](https://github.com/livepeer/go-livepeer/blob/master/doc/gpu.md) on the NVIDIA cards.
| GPU Model | Tested Transcoding | Tested Concurrent Ethash Mining | Notes |
|------------|--------------------|--------------------------|---------|
| NVIDIA GeForce GTX 1060 | :white_check_mark: | :white_check_mark: | |
| NVIDIA GeForce GTX 1070 | :white_check_mark: | | |
| NVIDIA GeForce GTX 1070 Ti | :white_check_mark: | :white_check_mark: | |
| NVIDIA GeForce GTX 1080 | :white_check_mark: | | |
| NVIDIA GeForce GTX 1080 Ti | :white_check_mark: | | |
| NVIDIA Tesla T4 | :white_check_mark: | | |
| NVIDIA GeForce GTX 1660 Ti | :white_check_mark: | | |
| NVIDIA GeForce GTX 1660 SUPER | :white_check_mark: | | |
If you’ve run Livepeer with a GPU not listed, and you’ve verified that it works, please submit an update to this list, which you can do by clicking the “edit” button on Github.