NVIDIA HAS TAKEN the covers of HGX-2, a cloud server platform designed to mix high performance and artificial intelligence (AI) computing into a unified architecture.
Featuring double precision floating point cores in the form of FP64 and FP32, HGX-2 has been designed to crunch calculations for scientific and simulation workloads while using FG16 and Int8 data to train AIs and power their uses - commonly referred to as inference.
Putting that into good old plain English; HGX-2 is a powerful data centre grade platform for handling high-end workloads like 3D modelling and simulation in the cloud and compute-intensive AI work.
It essentially forms the building blocks on which customer companies can build their own advanced systems for high-performance computing and AI tasks. Nvidia boasted that HGX-2 can achieve "record AI training speeds of 15,500 images per second on the ResNet-50 training benchmark" and can replace "up to 300 CPU-only servers"; that's some serious tech willy-waving.
"The world of computing has changed," said Jensen Huang, Nvidia's leather jacket-sporting founder and chief executive, at the GPU Technology Conference in Taiwan. "CPU scaling has slowed at a time when computing demand is skyrocketing. Nvidia's HGX-2 with Tensor Core GPUs gives the industry a powerful, versatile computing platform that fuses HPC and AI to solve the world's grand challenges."
To back up the bluster, Nvidia's HGX-2 makes use of Team Green's NVSwitch interconnect fabric which links together 16 Nvidia Tesla V100 Tensor Core GPUs to work together as one single giant GPU, which according to Nvidia, delivers two petaflops of AI performance - imaging how good it could make games look.
You may already have seen HGX-2 in action without knowing it as it forms the base of Nvidia's DGX-2 AI powering system designed to power deep learning workloads.
HGX-2 may just be an evolution of its predecessor, HGX-1, but that platform found use in the data centres of big tech firms like Amazon Web Services, Microsoft and Facebook. So an updated platform could be ready for use with other companies looking to follow the successes of those thee big players.
It's often easy to forget that Nvidia does a lot more than just GPUs, but that's easily done especially when it reportedly has a fresh family of pixel pushing graphics cards just around the corner. µ
But we probably won't see it until next year
Why stick a finger in a dyke when you can ram the entire boy in the hole, eh?
Reminds us that we're supposed to be able to trust them
'Exclusive' model starts shipping on 29 June