0% found this document useful (0 votes)

50 views

HPC Trends For 2015

The document discusses trends in high-performance computing (HPC) for 2015. It finds that two-thirds of respondents could use a 10x performance increase in the next five years, with over a third needing a 1000x increase. The top barrier to greater scalability is the scalability of software. It predicts that in 2015, server vendors will reshuffle market shares, specialized systems will overtake standard clusters, and ARM and OpenPower architectures will challenge x86 dominance as accelerators become more common. The growth of big data will require new frameworks and appliances beyond Hadoop.

Uploaded by

xiso2507

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

50 views

HPC Trends For 2015

Uploaded by

xiso2507

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 19

HPC Trends for 2015

Addison Snell
[email protected]

A New Era in HPC

Supercomputing for U.S. Industry

Report by U.S. Council on
Competitiveness
Justifications for new
levels of supercomputing
for U.S. industry
Free download from
www.compete.org, under
Publications tab
Free webinar recording at
www.intersect360.com,
under Presentations

Need for Scalability

Two-thirds of
respondents say
they could use
10x performance
increase over the
next five years

68%

37%

Over one-third
could use 1,000x

Barriers to Scalability
Scalability of
software is
viewed as the #1
barrier for 10x
greater scalability
This remains a top
barrier (with cost
of hardware) for
1,000x scalability

Software in Commercial HPC

Source: Solve. The Exascale

Effect: Benefits of Supercomputing
Investment for U.S. Industry, new
report pending publication, with
U.S. Council on Competitiveness

SC14: Vendor News

CORAL win boosts prospects for OpenPower, GPU
adherents
Nvidia, AMD, Intel advance accelerator roadmap
Bull reveals exascale platform
Lenovo hits the ground running with 1TF server
SGI extends UV portfolio
Dell intros GPU-heavy PowerEdge, Lustre storage
Huawei gets in the game with big memory server
Mellanox closes loop on 100Gbps
Cavium demos 48-core ARM chip

SC14: Trends
Intel vs. the world
IBM, Nvidia, AMD, Mellanox, and ARM line up against Intel
hegemony
More server vendors could back away from Intel dependency

FLOPS give way to data centricity

IBM Power architecture
SGI UV 300 and 30EX shared memory systems for in-memory
analytics
Cray Urika-XA and Urika GD analytics appliances
Data Vortex system optimized for data communications;
A3CUBE has data-centric architecture

The Year Ahead in HPC

Server vendors reshuffle market share
IBM will drop overall market share in its first year, post-Lenovo
HP and Dell will vie for HPC server market leadership
Lenovo will likely end up as number 3 server vendor in HPC

Standard x86 Beowulf cluster yields to more specialized

systems
ARM, GPU, Xeon Phi, and DSP accelerators in dense, powerefficient designs; AMD APU-based systems
OpenPower systems with or without accelerators
Systems with integrated flash for fast, nonvolatile data caching
Shared memory systems for in-memory analytics

The Year Ahead in HPC

ARM and OpenPower poised to chip away at x86
dominance
64-bit ARM experimentation continues, more servers, newer
processors

3D memory enters the mix

Micron commercializing Hybrid Memory Cube
Intel will integrate into Knights Landing processor
Toshiba, Intel to ship 3D NAND flash devices in 2015; Samsung
shipping today

Accelerator adoption reaching critical mass; more than

half of new HPC systems will contain some type of
accelerator in 2015

Solutions Evolving for Big Data

Commercialized frameworks
More than just Hadoop
Requires understanding of the specific problem

Big-memory appliances for in-memory database

The return of shared memory!
Amplified by changes in processing architectures

Data-centric computing
Move the computation, not the data (?)
Focus on complete workflows / data flows, not flops

Important Technologies for Big Data

Parallel file systems

Windows and mobile clients
Programming environments
Middleware (job scheduling, optimization, )
(Okay, sure ) Flash
On the other hand:
Keep Hadoop in its appropriate place
Not so much public cloud
Sorry, theres still no killer app

Growth in the Big Data Market

First off, what does that mean, specifically?
Certainly there are more organizations adopting
Big Data initiatives, but usually with:
Infrastructure they already have
People they already have
Maybe new software, but much of it open-source or
in-house

Most real spending is coming from planned

upgrades in normal refresh cycles
Big Data could help drive enterprise growth soon

Help Wanted
New applications and architectures create a
sudden need for new skills. From where?
Even without Big Data, new processor
architectures cause reinvention in skills for
normal titles like administrator
Programming skills and services will be in
particularly short supply
Warning: A bad analyst can do a lot of bad
analysis with good data

The Other Issue (Time Permitting)

QUIZ: Who said it, and in what year?
You have zero privacy anyway. Get over it.

The Other Issue (Time Permitting)

QUIZ: Who said it, and in what year?
You have zero privacy anyway. Get over it.

Answer: Scott McNealy,

CEO, Sun Microsystems
1999

Remember This?

What is your right to privacy?

Does it matter if you are famous?
in public office? a minor?

Conclusions
The demand for HPC is undiminished, and
the market will continue to grow, but
Specialization is now winning out over
standard solutions
Software is a bigger deal than hardware
Skill sets are a bigger deal than facilities
Regulation will play an increasing role