1

It’s Time To ROCm!

1

AMD INNOVATIONS AND SUPERCOMPUTING

\ AMD has been a leading innovator in high-performance systems‒ First with 64-bit x86 architecture, HyperTransport™, integrated memory controllers, multi-core socket compatible

upgrades, SSE FMAC extensions

\ AMD processors have been in many leading DOE supercomputers‒ In 2008, seven of the top ten supercomputers used AMD Opteron processors

\ AMD will lead again with systems incorporating advanced CPUs, GPUs, memory, and networks

2

>$50M fundingMemory,

Programming, APUs, Network,

more2012-2016

HBM High Bandwidth

Memory 14 NM

Radeon RX480June 2016

Driving Innovation Not Waiting For Miracles

Extreme Perf/Watt

Efficacy

Radeon R9 Nano46 GFLOP/Watt

Sept 2015

Open GPU ComputingPlatform

ROCm 1.0 April 2016

DOE Investsin AMD

The first GPU devices to use HBM and CoWoS

Radeon R9 Fury X, June 2015

3

4

INTRODUCING ROCm SOFTWARE PLATFORM

Graphics Core Next Headless Linux® 64-bit Driver• Multi-GPU Shared Virtual Memory• Large Memory Single Allocation• Peer to Peer Multi-GPU• Peer to Peer with RDMA• Systems Management API and Tools

HSA drives rich capabilities into the ROCm hardware and software• User Mode Queues • Architected Queuing Language • Flat memory Addressing • Atomic Memory Transactions • Process Concurrency & Preemption

Rich Compiler Foundation for HPC Developer• LLVM Native GCN ISA Code Generation• Offline Compilation Support• Standardized loader and Code Object• GCN ISA Assembler and Disassembler

A New Fully Open Source Foundation for HPC Class GPU computing

Open Source Tools and Libraries• Rich Set of Open Source Math Libraries• Tuned Deep Learning Library • Optimized Parallel Programing Frameworks• CodeXL Profiler and GDB Debugging

5

HIP

Convert CUDA to portable C++

• Single-source Host+Kernel• C++ Kernel Language• C Runtime (CUDA-like)• Platforms: AMD GPU, NVIDIA

(same perf as native CUDA)When to use it?• Port existing CUDA code• Developers familiar with CUDA • New project that needs

portability to AMD and NVIDIA

ROCm PROGRAMMING MODEL OPTIONS

HCC

True single-source C++ accelerator language

• Single-source Host+Kernel• C++ Kernel Language• C++ Runtime• Platforms: AMD GPU

When to use it?• New projects where true C++

language preferred• Use features from latest ISO

C++ standards

OpenCL

Khronos Industry Standard accelerator language

• Split Host/Kernel • C99-based Kernel Language• C Runtime• Platforms: CPU, GPU, FPGA

When to use it?• Port existing OpenCL code• New project that needs

portability to CPU,GPU,FPGA

6

Where you get your “Song Sheets” • Overview, Install, Docs, Code, Links: https://radeonopencompute.github.io/install.html

• Blogs: http://gpuopen.com/

7

WE ARE LOOKING TO BUILD OUT A WORLDWIDE BAND

� Get started today developing with ROCm - GPUOpen ROCm Getting Started http://bit.ly/1ZTlk82

� Engage In the develop of ROCm @ GitHub RadeonOpenCompute

� Show case your applications, libraries and tools on to ROCm via GPUOpen

How to Join

“The power of one, if fearless and focused, is formidable, but the power of many

working together is better.”

– Gloria Macapagal Arroyo

8

9

CUDA

PORTING WITH HIPIFY

hipify

Developer Cleanup and

Tuning

` 95%+ Automatic Conversion

PortableHIP C++

` Developer maintains HIP port` Resulting C++ code runs on NVIDIA

(same performance) or AMD GPUs

10

HIP COMPILATION PROCESS

HIP->HC

Header

HCC

Portable HIP C++(Kernels + HIP API)

HCC C++(Kernels + HC)

` HIP API implemented with lightweight HIP runtime

` Uses HCC’s hc::accelerator, hc::accelerator_view, hc::completion_future

` Some calls directly into ROCR` Compute kernels mostly unchanged

` Code compiled with HCC` Can use CodeXL

HIP->CUDA

Header

NVCC

CUDA(Kernels + CUDA API)

` HIP API implemented as inlined calls to CUDA Runtime

` Compute kernels mostly unchanged

` Code compiled with NVCC (same as CUDA)

` Can use nvprof, CUDA debugger, other tools

AMDNVIDIA