Processor
CPU performance isn’t especially important for TensorFlow, but there are certain features that are only available in higher end CPUs. PCIe lanes is the most notable.
Since TensorFlow users will want to use multiple GPUs whenever possible, they need to have enough PCIe lanes to support all of them.
For this reason, you may want to consider a dual-CPU workstation. Also note that these platforms often feature technologies designed to boost AI performance.
Not all user workflows are the same though, so be sure and contact us for an optimized configuration.
Memory
The amount of memory you will need depends on the size of your datasets and how many programs you will have open at any given time alongside TensorFlow.
For extremely large datasets, you should take advantage of the high capacity and multi-channel memory support available in multi-CPU platforms.
Graphics Card
TensorFlow users should build their systems around GPU power. You should opt for the most powerful GPU possible, and more than one of them if possible. Nvidia GPUs are the standard for TensorFlow as they offer the greatest compatibility, stability and performance.
Higher end Quadro cards may be the best option due to their additional VRAM, and reduced heat output.
Storage
In the past, computers were held back by slow mechanical hard drives.
Unless you are storing files which are not accessed too often, in which case mechanical hard drives might be a better choice, Solid State drives should be used for everything else.
Having everything stored on SSDs means you’ll be able to copy, move, open and save files quickly and PC and program start times will be reduced.
Recommended Systems
Apollo
from R 29 995 ex VAT
Our Apollo workstation sets the standard for high-performance workstation.
Vulcan
from R 34 995 ex VAT
Our Vulcan features cutting-edge tech with next-level performance.
Hydra
from R 79 995 ex VAT
Our Hydras feature AMD Threadripper CPUs with support for up to 2TB RAM.