+
EDITOR’S QUESTION
/////////////////
PATRICK SMITH,
FIELD CTO EMEA,
PURE STORAGE
our AI ready infrastructure which is a joint
development with Nvidia.
It combines Nvidia's DGX –1 AI compute
servers with our FlashBlade unstructured
data storage and high speed networking and
a preconfigured software stack that really
provides rapid time to market for AI.
Our view is that the public cloud for AI
provides a great proof of concept proving
ground. But as soon as people go beyond
proof of concept with the data volumes
involved, the amount of processing
involved in AI workloads means that
they'll reach a critical mass and bring that
capability back on-prem.
What we are seeing is as part of an AI
infrastructure is people want to keep it
as busy as possible. People want to be
processing and keeping those and Nvidia
compute servers busy 100% of the day 24/7
365 days a year.
W
e're seeing that pretty much
all of the large enterprises are
deploying multi cloud today often
– whether they know it or not. So we're seeing
large enterprises certainly starting off with on-
prem cloud, where the next step beyond that
is typically the move to public cloud. simply on-prem and SaaS or whether they're
on-prem, public and SaaS.
But beyond that we find that people are not
just using one public cloud, they're looking
at more than one public cloud provider.
Often they go into that looking at arbitrage
between public clouds but actually it's about
risk mitigation rather than financial upcharge. The other thing I think that's worth saying
is as we look at the public cloud, it's not one
size fits all. We're seeing people moving
workloads into the public cloud and also
coming back out of it for workloads that
aren't optimal in the public cloud.
So today often people are using multi cloud
whether they know it or not; whether they're So if you look at our product set, one of the
products that we're talking about a lot is
www.intelligentcio.com
People also talk about hybrid cloud, on-prem
and public. We view it genuinely as multi
cloud and we think most of our customers
are moving in that direction.
And to be able to keep AI servers busy you
need really fast data processing and data
throughput to the servers. This is where
FlashBlade comes in – with its parallel low
latency performance that scales all the
way through the A.I. lifecycle be it initial
data capture and the properties associated
with the initial data capture through data
cleansing and then A.I. training. This has a
completely different workflow characteristic
to the initial data capture.
So FlashBlade's multi-dimensional
performance covers all of the sequential
right type of access together with the pure
random read access that you associate with
AI training. n
INTELLIGENTCIO
37