Microsoft details 'planet-scale' AI infrastructure packing 100,000-plus GPUs

the register ^

[algore]

Microsoft has revealed it operates a planet-scale distributed scheduling service for AI workloads that it has modestly dubbed "Singularity".

Described in a pre-press paper [PDF] co-authored by 26 Microsoft employees, Singularity's aim is described as helping the software giant control costs by driving high utilization for deep learning workloads.

Singularity achieves that goal with what the paper describes as a "novel workload-aware scheduler that can transparently preempt and elastically scale deep learning workloads to drive high utilization without impacting their correctness or performance, across a global fleet of AI accelerators (e.g., GPUs, FPGAs)."

The paper spends more time on the scheduler than on Singularity itself, but does offer some figures to depict the system's architecture.

An analysis of Singularity's performance mentions a test run on Nvidia DGX-2 servers using a Xeon Platinum 8168 with two sockets of 20 cores each, eight V100 Model GPUs per server, 692GB of RAM, and networked over InfiniBand.

With hundreds of thousands of GPUs in the Singularity fleet, plus FPGAs and possibly other accelerators

[for-q-clinton]

Wrong. We are living with the consequences of people buying his product and making it #1.

[GaltAdonis]

Yes, you are right.
If we had had some smart-aleck time traveler guy from
42 years in the future beam back in time and say:

"No! Don't buy Windows! It's Crap! Buy this instead!"

People would have asked: "Why?"

And the time-traveling software engineer would have answered
using terminology and jargon that had not yet been invented
- and they would have bought Windows anyway because
the blue box it was in was more attractive - little
knowing that the "pretty blue box" would ultimately usher
humanity into the nightmarish scenario of "The Blue Screen of Death"
and computer viruses, malware, adware, etc.

And if the timeline had been disrupted in such a manner, there
may well have been temporal distortions that caused
people to spontaneously mutate into cute penguin-like
stuffed animals.

And then how could the human race have proceeded under
such hellishly 'cute' conditions?

Correct - It couldn't have.

So therefore, the free market should always be allowed to
make such decisions - not weirdo know-it-all time traveling geeks.

.

[martin_fierro]

[martin_fierro]