In case you’ve been living under a rock, the new 3rd Generation AMD Ryzen Threadrippers built on the Socket sTRX4 (TRX40 chipset) are pretty awesome to say the least. According to AMD these CPUs are:
“Built for serious visual effects artists, data scientists, or anyone who takes their time seriously, the scalable AMD TRX40 chipset offers unprecedented expansion for serious multi-GPU and NVMe arrays thanks to bleeding-edge PCIe 4.0 connectivity, exclusive to this platform in high-end desktop. Meanwhile: quad channel DDR4, available ECC support, and unlocked overclocking make TRX40 a lock for users that depend on serious computing.”
With this in mind Exxact Corporation has created 3 new Deep Learning Workstation Configurations featuring the new AMD Ryzen Threadrippers.
But first let’s examine why CPU choice is important for AI Developers.
What Role Does the CPU play in Deep Learning?
While typically the GPU is the workhorse of deep learning workloads, it’s good to understand what functions the CPU plays. A good CPU with many cores can boost preprocessing performance, keeping the system stable while other threads and programs execute.
The CPU provides vital functions for:
- Loading the mini-batch;
- Preprocessing the mini-batch;
- Executing scripts to initiate training;
- Experimental Compilers such as XLA for TensorFlow;
- For Deep Reinforcement Learning, the CPU is important for sequential tasks related to the agent, environment, and policy of the algorithm.
Of course there are many more reasons why you’d want a powerful CPU, but let’s see the workstations!
24 Core AMD Ryzen Threadripper 4 GPU Workstation
- Great for most deep learning researchers
- Budget friendly starting at $9,300
32 Core AMD Ryzen Threadripper 4 GPU Workstation
- Great for developing any deep learning application
- Up to 96GB GPU memory
- Prices start at $19,599
Elite 64 Core AMD Ryzen Threadripper 4 GPU Workstation
Our most powerful AMD Ryzen Based Deep Learning Workstation goes beyond fantastic and is powered by a AMD Ryzen Threadripper 3990X 64 Core Processor. Acceleration is powered by up to 4X NVIDIA Quadro RTX 8000 GPUs.
- Train most deep neural networks including Transformers
- Up to 192GB GPU Memory!
- Prices start at $26,500
Other key specs for the new Ryzen processors
- Quad-Channel DDR4 memory with optional ECC support
- A total of 88 PCIe® 4.0 lanes
- Multi-GPU support
- Up to 12 native USB 3.2 Gen2 Ports (Superspeed 10Gbps)
- Up to 2×4 NVMe ports
- Up to 20 SATA Ports
- SATA RAID 0, 1, 10
- NVMe RAID support
- Overclocking support
Deep Learning GPU Benchmarks
We’ve done the work so you don’t have to guess what you need! Check out any of our GPU benchmarks below to find what would work best for you:
- RTX 2080 Ti Deep Learning Benchmarks for TensorFlow
- TITAN RTX Deep Learning Benchmarks for Tensorflow
- NVIDIA Quadro RTX 6000 GPU Benchmarks for TensorFlow
- Quadro RTX 8000 Deep Learning Benchmarks for TensorFlow
- NVIDIA Quadro RTX 8000 BERT Large Fine Tuning Benchmarks in TensorFlow