What could you do with easy access to monster GPU?

King Kong The Eighth Wonder Of The World © UNIVERSAL ALL RIGHTS RESERVED

This article was also featured on CodeX.

Imagining a future where we escape the chains of the PCIe connector… and unleash remote GPU

Most of us imagine that we’re facing forward into the future, with the past behind. Some cultures picture things in reverse — if the past is visible but the future is not, aren’t we really facing the other direction?

Regardless of how we view our experience moving through time, our vision of the future depends on our ability to clearly see the present:

‍

It’s presently clear… the PCIe connection is throttling accelerated computing

Our most impactful innovations often look like step-changes to most of the world, but the myth of the lone inventor is just that  —  a myth. In truth, there are always at least a handful, and often hundreds, of people who are close enough to any technology space that The Big Breakthrough can be seen, by them, for what it is  —  a push through a meaningful, but incremental, final barrier that makes the innovation viable.

In the case of Remote GPU (rGPU for short), our team at Juice Labs is inviting customers, partners, and individual leaders in the space to join us in “seeing the present before anyone else does.”

Here is the present that we see:

1. GPU hardware, the primary resource behind accelerated computing, is big, heavy, hot, and power-hungry — and therefore non-portable and cumbersome.
2. GPU hardware needs to run “right next to” (in terms of the physical pipeline carrying data) the applications it is accelerating.
3. That physical pipeline, the PCIe connection, is oppressive — it dictates that the precious resource of acceleration can only be used where (cumbersome, non-portable) GPU hardware is physically running.

Software eats hardware — and unleashes rGPU 

More positively, another part of the present that we see is that technology trends are aligning to make replacing that oppressive hardware connection — abstracting it with software that can run over a network, from anywhere — possible.

The vision of remote access to GPU from anywhere is the vision we call rGPU. We believe the time is now to align around this vision.

Imagine:

1. Your smartphone running native apps with graphics acceleration provided by a powerful GPU located “at the edge” — in your home via wifi, or on a nearby cell tower via 5G
2. Playing any PC game you own, natively, on a cheap laptop or desktop, using GPU from the cloud, or from another machine in your home
3. Running ML experimentation, development, or training from your local machine, and just “turning up the GPU” whenever you want — without having to switch environments
4. Small, lightweight IoT devices that can run beefy AI inference models with the full power of edge or cloud GPU
5. Wide pools of shared GPU capacity within or between data centers, with dynamic resource-balancing among all application hosts (physical, virtual, or container)

These are just a few examples  —  the possibilities are as numerous as the potential uses for accelerated computing.

 What could you do with easy access to monster GPU?

If your organization struggles with the expense or inconvenience of today’s GPU status quo, reach out to us at info@juicelabs.co or @juicegpu  —  we’d love to understand your challenges, and talk about how we can help.

Steve Golik is co-founder of Juice Labs, a startup with a vision to make computing power flow as easily as electricity.

‍