Machine for DeepLabCut ver2.3 (February 2024 version)

Case No.PC-10086A customer who viewed the site asked us about a workstation for DeepLabCut ver2.3. We are looking for an optimal configuration with a budget of around 100 million yen.

Regarding the level of processing power, it is necessary to have a configuration that can run DeepLabCut at a speed equal to or faster than the workstation equipped with 6000 x NVIDIA RTX A1 that the customer is currently using, and we also request that DeepLabCut ver2.3 be installed in advance. It has received.

Other specific specifications are as follows.

CPU Ability to run DeepLabCut without problems
memory 128GB
GPU Something that can perform DeepLabCut analysis at the same speed as RTX A6000 (Geforce RTX4090, etc.)
storage 1TB M.2 SSD x2

Based on the information received from the customer, we proposed the following configuration.

CPU Core i9-14900K (3.20GHz 8 cores + 2.40GHz 16 cores)
memory 128GB
Storage 1 1TB SSD M.2 NVMe Gen4
Storage 2 1TB SSD M.2 NVMe Gen4
video NVIDIA Geforce RTX4090 24GB
network on board (2.5GBase-T x1) Wi-Fi x1
Housing + power supply Middle tower type housing + 1500W
OS Microsoft Windows 11 Professional 64bit
Others DeepLabCut installation

This configuration is equipped with the latest 2024th generation Core i2 as of February 14. I chose the Core i9-24K with a total of 9 cores to suit my budget.

In addition, when actually executing DeepLabCut, most of the load is placed on the GPU, and it is thought that there are very few situations where the CPU is placed under a high load.
Therefore, if CPU performance is not important, you can change to a lower model such as Core i7.

GeForce RTX4090and RTX A6000, which one is best?

In accordance with the customer's wishes, we selected an NVIDIA GeForce RTX4090 24GB GPU.
Comparing NVIDIA GeForce RTX4090 24GB and NVIDIA RTX A6000 48GB, there are differences in chip generation and VRAM capacity. The chip generation used in the RTX4090 is newer, but the VRAM capacity is half that of the A6000.
DeepLabCut's GPU processing is based on TensorFlow, and its behavior while using GPU also follows the characteristics of TensorFlow. When executed with TensorFlow's default settings, DeepLabCut allocates as much video memory as possible and executes the process. Therefore, although there may be an effect due to the difference in VRAM capacity, the number of CUDA cores on the RTX4090 is over 18,000, which is nearly twice that of the A6000. In actual use, it is thought that the difference in the number of CUDA cores will have a greater effect than the difference in VRAM capacity, so we can expect an improvement in analysis speed compared to the RTX A2.

In addition, the GPU requirements posted in the DeepLabCut official repository recommend 8GB or more of VRAM. Considering that value, it can be said that the 4090GB of RTX24 has sufficient VRAM.


The configuration of this case study is based on the conditions given by the customer.
We will flexibly propose machines according to your conditions, so please feel free to contact us even if you are considering different conditions than what is listed.

■ Keywords

・What is DeepLabCut?

DeepLabCut is an open source deep learning tool for analyzing animal behavior. It can identify specific body parts of animals from videos and track them without markers, providing highly accurate movement analysis.

Reference: DeepLabCut — The Mathis Lab of Adaptive Intelligence *Jumps to an external site