TensorFlow

Processor

This application relies heavily on the GPU to perform calculations, therefore a processor with a low core count and low clockspeed is acceptable.

Memory

Benchmarks indicate that TensorFlow performs best when the amount of system memory is double (or more) the graphics card memory. So if you have one RTX 2080Ti card (which is 11GB), then we recommended 24GB RAM. If you have two RTX 2080Ti cards (22GB), then we recommended 48GB RAM.

Storage

This application relies more on reading data than writing data as it analyzes and processes training datasets. Extreme read rates are not required however, therefore we recommend a 512GB SSD for the operating system and program files. Furthermore, because deep learning requires large datasets to be processed, storing this data locally is important. Therefore a 4TB, 6TB or 8TB hard drive (or multiple hard drives) is recommended depending on your storage needs.

Graphics

The RTX 2080Ti has become the defacto graphics card for deep learning and TensorFlow offsets all the computing of data to the GPU. For maximum performance, recommend two NVIDIA GeForce RTX 2080Ti graphics cards.

Read More: Adding more than 2 Graphics Cards

Chassis

The HP Z8 G4 Workstation is recommended if you would like dual processors and/or multiple high performance graphics cards.

The HP Z6 G4 Workstation is recommended if you would like to start with a single high-end processor and single high-performance graphics card with the potential of upgrading in the future. With the Z6 G4, you can add a second matching processor and double the memory using the optional second processor riser. The power supply will also support two mid-range graphics cards.

The HP Z4 G4 Workstation is recommended for majority of workflows. Since Intel now produces single Core i9 and Xeon processors with up to 18-cores and high clock speed, there is no need for dual processors unless you will have heavy CPU rendering or intensive tasks demanding more than 18-cores. Also, the HP Z4 G4 has an optional 1000W power supply that will support the most powerful graphics cards inluding the GeForce RTX 3080 and Quadro RTX A5000.

Specialty Hardware

For massive computes, we carry NVIDIA Tesla K80 and P100 accelerators installed in worlds leading Cubix Xpander Rackmount series expansion chassis. The top selling, Xpander Rackmount 2, Gen 3 expands one PCIe 3.0 x16 slot in the HP Z Worskation into two dual-slot PCIe 3.0 x16 slots for passively cooled cards. Add 2 4 8 or 16 Tesla compute processors to for extreme compute power to take on any advanced product. Shave hours, days, or weeks off your estimated completion time to maximize your productivity.

For rigorous rendering and advanced deep learning solutions, we carry the Cubix Xpander Desktop Elite series expansion chassis ideal for NVIDIA GeForce GTX and RTX GPUs. The most popular in the series, Xpander Desktop Elite, Gen 3 - 1500w, expands one PCIe 3.0 x16 slot in the HP Z Worskation into four dual-slot PCIe 3.0 x16 slots for actively cooled cards. The 1200w model also available for a mixture of GPUs and specialty hardware cards, and the 8 and 16 GPU model for a grid computing solution.