NVTabular Support Matrix
We offer the following containers:
Merlin Inference: Allows you to deploy NVTabular workflows and HugeCTR or TensorFlow models to the Triton Inference server for production.
Merlin Training: Allows you to do preprocessing and feature engineering with NVTabular so that you can train a deep learning recommendation model with HugeCTR.
Merlin TensorFlow Training: Allows you to do preprocessing and feature engineering with NVTabular so that you can train a deep learning recommendation model with TensorFlow.
Merlin PyTorch Training: Allows you to do preprocessing and feature engineering with NVTabular so that you can train a deep learning recommendation model with PyTorch.
The following tables provide the software and model versions that NVTabular version 0.6 supports per container.
Table 1: Support matrix for the Merlin Inference (merlin-inference) container
DGX |
|
DGX System |
|
Operating System |
Ubuntu version 20.04 |
NVIDIA Certified Systems |
|
NVIDIA Driver |
The 21.06 release is based on NVIDIA CUDA version 11.4, which requires NVIDIA Driver version 465.19.01 or later. However, if you’re running on Data Center GPUs (formerly Tesla) such as T4, you can use any of the following NVIDIA Driver versions:
NOTE: The CUDA Driver Compatibility Package doesn’t support all drivers. |
GPU Model |
|
Base Container Image |
|
Container Operating System |
Ubuntu version 20.04 |
Base Container |
Triton version 21.07 |
CUDA |
11.4 |
RMM |
21.06 |
cuDF |
21.06 |
cuDNN |
N/A |
HugeCTR |
3.1 |
NVTabular |
0.6 |
Table 2: Support matrix for the Merlin Training (merlin-training) container
DGX |
|
DGX System |
|
Operating System |
Ubuntu version 20.04 |
NVIDIA Certified Systems |
|
NVIDIA Driver |
The 21.06 release is based on NVIDIA CUDA version 11.4, which requires NVIDIA Driver version 465.19.01 or later. However, if you’re running on Data Center GPUs (formerly Tesla) such as T4, you can use any of the following NVIDIA Driver versions:
NOTE: The CUDA Driver Compatibility Package doesn’t support all drivers. |
GPU Model |
|
Base Container Image |
|
Container Operating System |
Ubuntu version 20.04 |
Base Container |
N/A |
CUDA |
11.4 |
RMM |
21.06 |
cuDF |
21.06 |
cuDNN |
8.2.2 |
HugeCTR |
3.1 |
NVTabular |
0.6 |
Table 3: Support matrix for the Merlin TensorFlow Training (merlin-tensorflow-training) container
DGX |
|
DGX System |
|
Operating System |
Ubuntu version 20.04 |
NVIDIA Certified Systems |
|
NVIDIA Driver |
The 21.06 release is based on NVIDIA CUDA version 11.4, which requires NVIDIA Driver version 465.19.01 or later. However, if you’re running on Data Center GPUs (formerly Tesla) such as T4, you can use any of the following NVIDIA Driver versions:
NOTE: The CUDA Driver Compatibility Package doesn’t support all drivers. |
GPU Model |
|
Base Container Image |
|
Container Operating System |
Ubuntu version 20.04 |
Base Container |
nvcr.io/nvidia/tensorflow:21.07-tf2-py3 *Customized with TensorFlow version 2.4.2 |
CUDA |
11.4 |
RMM |
21.06 |
cuDF |
21.06 |
cuDNN |
N/A |
HugeCTR |
3.1 |
NVTabular |
0.6 |
Table 4: Support matrix for the Merlin PyTorch Training (merlin-pytorch-training) container
DGX |
|
DGX System |
|
Operating System |
Ubuntu version 20.04 |
NVIDIA Certified Systems |
|
NVIDIA Driver |
The 21.06 release is based on NVIDIA CUDA version 11.4, which requires NVIDIA Driver version 465.19.01 or later. However, if you’re running on Data Center GPUs (formerly Tesla) such as T4, you can use any of the following NVIDIA Driver versions:
NOTE: The CUDA Driver Compatibility Package doesn’t support all drivers. |
GPU Model |
|
Base Container Image |
|
Container Operating System |
Ubuntu version 20.04 |
Base Container |
|
CUDA |
11.4 |
RMM |
21.06 |
cuDF |
21.06 |
cuDNN |
N/A |
HugeCTR |
N/A |
NVTabular |
0.6 |