CARMA Update

CUDA on ARM Architecture

A Platform for Developing Accelerated

Applications on ARM

2

CARMA is a forward-looking development platform for

high performance, energy efficient hybrid computing

It’s a platform for the next generation of HPC, leveraging

commodity driven improvements from the most rapidly evolving

compute markets.

3

The next revolution: Power Efficiency

Look at the market for the next generation of HPC components

Power-effective computing driven by phones and tablets

ARM, with architectural and experience advantages

System-level software complexity is high

HPC driven by accelerated computing

All major vendors have switched to accelerators

GPUs have an architectural efficiency advantage

Titan gets 90% of its performance from the accelerator

4

Possible Obvious Power-efficient Future

Power-efficient general purpose cores

combined with Compute Accelerators Power control shared with mobile products

Ultra-focused on power efficiency

Competition forces rapid improvement

Technology evolution driven by commodity market

Bulk of compute power provided by inherently efficient GPUs

Increase to over 50% of chip power for flops.

5

GPU 225 pJ/flop

Optimized for throughput and power efficiency

Explicit management of on-chip memory

CPU 1700 pJ/flop

Optimized for latency

Caches

Fermi

40 nm

Westmere

32 nm

6

GPU and Computing

NVIDIA has these elements

VGX

ARM SoCs

7

Why CARMA?

Have a development platform for these future HPC systems

Explore the efficiency and performance trade-offs for existing

ARM+GPU systems

Check, tune and evaluate accelerator enabled applications

8

CARMA Hardware Overview

Ultra low power host Tegra CPU module

T30 “Kal-El” w/ 4 ARM A9 cores, NEON and VFPv3

2GB DRAM, and 4GB or 8GB embedded flash

On a Seco Q7 module for easy upgrade

NVIDIA MXM GPU module

Quadro1000m (Fermi class) on 4 lanes of PCIe

96 CUDA cores with 200GFLOPS SP peak

Seco carrier board

Provides I/O connectors, power supplies, etc

9

Moving from prototype to development systems

CARMA clusters

10

CARMA Software Overview

ARM Linux distribution

Ubuntu 11.04 for ARM

Linux 3.1.10 kernel

Enhancements to support Tegra features

CUDA 4.2 run-time and libraries

Host x86 system support for cross development

CUDA cross-compiler

11

Developer Information

For support and questions, register on the CUDA DevZone

http://www.nvidia.com/carmadevkit

http://www.nvidia.com/devzone

Future enhancements

Native (ARM hosted) compile support

Updated CUDA versions e.g. CUDA 5.0

Long term plans for the CARMA platform

Updated modules

ARMv8 64 bit platform support