Inventing the Future of Computing

Kickstarting the Transition to Parallel Computing With Open Hardware

Andreas Olofsson

andreas@adapteva.com Linux Collaboration Summit 2013

April 15th-17th, 2013 – San Francisco, CA

1

What is Adapteva?

2

Possibly the World’s Smallest Semiconductor

Company

64-core 28nm 100 GFLOPS

Coprocessor @2W

Now Also a System Company…

#1 in Processor Energy Efficiency at

50 GFLOPS/Watt

The Computing Energy Crisis: IT’S REAL!!

3

0

1

10

100

1,000

10,000

100,000

1,000,000

10,000,000

100,000,000

1990 1995 2000 2005 2010 2015 2020 2025 2030

System ProcessingNeedsLegacy ProcessingEfficiency

“The Efficiency Gap” Von Neumann Saturation

The Ideal World

4

Inf B RAM

CPU (Inf Hz)

Inf B/sec

The Real World: Performance is Saturating!

5

4. Add Cores

3. SIMD

1. Scale Frequency

Now What?

2. Multi Issue

Power Consumption

Thermal Density

Memory Bottlenecks

Latency Wall

Wiring

Frequency Wall

Yield Issues

Time to Market

Software Complexity

Amdahl’s Law

10 Trends that Will Shape the Future of Computing

6

Nature is Massively Parallel

7

Efficient

Robust Hetero-geneous

Parallel

The Heterogeneous Computing Vision

8

SYSTEM-ON-CHIP

BIG CPU

FPGA

BIG CPU

BIG CPU

BIG CPU

100’s of small RISC CPUs GPU Analog

The Current State of Parallel Programming

9

How To Make Every Programmer a Parallel Programmer?

Source: Github

10

Industry Challenges Before Us

• Rebuild the computer ecosystem

• Rewrite billions of lines of code

• Re-educate millions of programmers

• Rewrite the education curriculum

11

No Computing Parallel Computing

“Von Neumann Age” Serial Computing

1943-2013?

No Electronic Computing

-1943

Parallel Computing 2013-??

Any Reason to Think the Future of Computing is NOT Parallel?

What is Parallella?

12

“A $99 credit card sized parallel computing platform”

Parallella Principles

13

• PARALLEL: • Heterogeneous and scalable parallel hardware

• OPEN: • Open source O/S (Linux) • Open chip documentation • Open source drivers and SDK • Open source hardware (board schematics, docs, layout) • Open standards (OpenCL, OpenMP, MPI, ...)

• ACESSIBLE: • $99 starting point • Easy to use novice progammers

Parallella Kickstarter Stats

14

• ”$99 Linux supercomputer” • 5,000 backers • 6,300 boards ”pre-sold” in 4 weeks • 67 countries, all 50 US states • 50-75% of backers are developers • 5,000 more signups since Jan 1st • Customer Application Classes:

• SDR (GNU Radio) • Ray tracing/rendering • Image processing • Robotics • Gaming

• Cryptography • Media Server • Distributed Computing • Signal processing • HPC

The Parallella Architecture

15

Dual Core ARM A9

AXI BUS

MIO

SHARED DRAM

“O/S” DRAM

USB OTG USB 2.0

UART Ethernet

SD-CARD I2C

DAC/ADC IF HDMI

Controller

AXI-MASTER AXI-SLAVE

“Glue-Logic”

Daughter Card

AXI-MASTER

Zynq FPGA

Zynq “Hard”

Off-Chip

MEM-CTRL

“Sandbox” EPIPHANY

For the first time in public…

16

PARALLELLA-16 PARALLELLA-64

• Zynq Dual Core ARM A9 • 16-core Epiphany Accelerator • 1GB RAM • GbE, USB, HDMI, uSD • 6 GB/s expansion connectors • $99 (long term goal)

• Same as PARALLELLA-16, with • 64-core Epiphany Accelerator

First Parallella-16 Power-Up was at 1:30pm on April 11

Parallella Bring-up Day 1

17

Parallella Bring-up Day 2

18

Parallella Bring-up Day 3

19

Epiphany Coprocessor Read/Write

Success

Parallella – What’s Next?

20

Ship 6,300 Boards ASAP

Start the Parallella Academic Program

Start working on Parallella-1024

Build a sustainable supply model