Google's followed its cloudy rivals into the servers-plus-GPUs caper, announcing that “Early in 2017, Google Cloud Platform will offer GPUs worldwide for Google Compute Engine and Google Cloud Machine Learning users.”
The Alphabet subsidiary will offer AMD's FirePro S9300 x2 to power remote workstations. If you wonder why we're writing about workstations so often at present, Autodesk's throwing an event this week. Hence workstation news from Teradici, HP Ink and AWS within a small span of days.
But we digress. Google will also offer NVIDIA's Tesla P100 and K80 GPUs for artificial intelligence, deep learning, AI and general HPC applications that need the kind of grunt only GPUs can bring.
Google says up once this service is up and running you'll be able to hang up to eight GPU dies in any non-shared-core machine and pay for them by the minute.
If you fancy this, there's a joint-the-queue form to complete. Perhaps if you do so, Google will give you some pricing information.
AWS and Azure have already announced GPU-powered instances and also offer GPU-backed cognitive services, making this another field in which cloud looks like a very fine alternative to building one's own infrastructure. That's not stopping big scientific users from ordering new GPU-assisted supercomputers. But for the rest of us, this kind of service offers the chance to grab some GPU grunt as and when needed. And perhaps once the developers we work with figure out how to make the best use of these new toys. ®