AMD announces Radeon Instinct, the beginning of AI

We are in for a wild freakin' ride in 2017, with AMD kicking things off today with the announcement of something radically new: Radeon Instinct. What is Radeon Instinct? There's no easy answer to that, but it is the next big thing in cloud computing - something being touted as the machine intelligence era.

Back in the 1960s, the big thing at the time were the massive main frames that would take up entire buildings, requiring radical amounts of cooling and physical space - while not providing much power (compared to the insane amounts of data crunching power we have now).

In the 80s and 90s it shifted to client-server operations, and then in the last 15 years we've seen a massive shift towards cloud computing.

Cloud computing isn't disappearing any time soon, but it will be morphing into its own beast - the machine intelligence era, and the market for this is virtually unlimited. AMD is at the forefront of this, alongside competitors NVIDIA, Intel, Qualcomm and more - but AMD has somewhat of an ace up its sleeve with Radeon Instinct.

Some would think that machine intelligence isn't needed just yet, but there are so many things that are controlled by computers and technology in our day-to-day lives - and it's only extending into more things as time goes by. Autonomous cars are the next big thing, smart homes are becoming more of the norm, nano robots are on the way, and so much more.

This is how compute infrastructure is done today.

While this is how it'll be done in the near future, with open accelerators in the middle.

AMD is well aware of the problems, as we need a better way of sifting through all of the data created per day.

Or as AMD puts it, a human brain in your hand.

This is where the Radeon Instinct Initiative is born, ready for new markets that use a common infrastructure that will better use the investments and scale quickly into multiple industries.

Right now, supercomputers can't learn easily - with the usual 'is this a dog' being asked with a picture, and if it doesn't know the answer - it has to work it out. Eventually, this will take seconds - versus hours, and this is where the fun begins.

Radeon Instinct Accelerators - The Future of Machine Learning

AMD is announcing 3 new Radeon Instinct Accelerators today, with some Vega GPU love thrown in. All of the Radeon Instinct accelerators are built exclusively by AMD and are passively cooled.

The first is the Radeon Instinct MI6, a passively-cooled inference accelerator with 5.7 TFLOPs of comptue performance, 224GB/sec memory bandwidth, and uses less than 150W.

Next up, we have the MI8 which is a small form factor accelerator with 8.2 TFLOPs of performance, 512GB/sec memory bandwidth, and under 175W TDP.

Finally, we have a Vega-powered Radeon Instinct MI25 Vega with NCU accelerator, which is a passively-cooled training accelerator. The MI25 has 2x packed math, the spiffy new High Bandwidth Cache and Controller (which we can't talk too much about now, but you don't have long to wait), and under 300W TDP.

How Will Radeon Instinct Accelerators Be Used?

We can't have all of that power without something to throw it at, right? We've got to run into the arms of Skynet, which is where MIOpen comes into play, super high performance - a deep learning library for Radeon Instinct.

Just how much faster is MIOpen? AMD has compared it to GEMM-based Convolutions, and against both the Maxwell- and Pascal-based Titan X graphics cards.

Hardware virtualized deep learning will be capable of much more than we're used to now, with datacenters set to explode with compute power.

Our First Real Taste of Zen

AMD teased the new Radeon Instinct accelerators with their next-gen Zen architecture, under the Naples platform during the AMD Tech Summit.

Naples has been optimized for GPU and accelerator throughput computing.

New Servers Powered by Radeon Instinct

The raw GPU performance of the Radeon Instinct accelerators is already being put to use in 3 new servers, starting with a 100 TFLOPs beast in the Inventec K888 G3 with Radeon Instinct.

Inside, there's 24 x DDR4 DIMMs, support for either 12 x 3.5-inch/24 x 2.5-inch HDDs/SSDs, 4 x PCIe 3.0 x16 ports, and 2 x PCIe 3.0 x8 mezz, as well as 4 of the new Radeon Instinct MI25 accelerators. The system uses a single 1600W PSU, with a redundant PSU just in case of power outages.

Performance wise, we're looking at up to 100 TFLOPs, 120TB of storage, and up to 768GB of DDR4 RAM.

Next up, we have the Falconwitch with Radeon Instinct, offering 400 TFLOPs of performance - or around 9x the bandwidth of the best GPU-based server on the market.

The Falconwitch PS1816 takes 16 x Radeon Instinct MI25 accelerators across its 288 x PCIe 3.0 lanes, and has 6600W of PSU to power the entire server. Falconwitch designed the PS1816 for machine learning, NVMe over Fabric, and more.

But you haven't seen anything yet - the Inventec Rack with Radeon Instinct rocks a freakin' mind blowing 3 petaflops of GPU performance (1000 TFLOPs = 1 PFLOP of GPU performance).

Inside, we have an enormous 120 x Radeon Instinct MI25 accelerators - yes, 120 of them. Absolutely insane.

Wrapping Up

If you thought this was everything, you're wrong - there's so much more that AMD has up its sleeve for the immediate future on December 13 with its New Horizon event, and into 2017 - starting at CES 2017 in January.

The launch of Radeon Instinct was a big one, as it'll make the entire world look to AMD for super-fast server and supercomputer hardware, something that companies have needed to tap the likes of Intel and NVIDIA... until now. AMD has positioned itself beautifully with Radeon Instinct, and as I said - this is just the beginning.

I remember the feeling that I had when I saw AMD announce the 3 petaflops of GPU performance with the final Inventec Rack system with 120 x Radeon Instinct MI25 accelerators. I had goosebumps. We're not even seeing the end game result here folks, this really is a seed that was planted around 5 years ago now, and AMD are just picking off the first fruits of their labor.

AMD is going to have a tectonic shift in 2017 in all departments, the company is firing on all cylinders - my technology instincts are telling me, that or it's my spider senses.