This document discusses high performance computing with accelerators. It introduces accelerators like GPUs, FPGAs, and IBM's Cell processor that are used in supercomputers and computer clusters to accelerate computations. GPUs are well-suited for graphics and floating-point work, FPGAs for embedded and low-bit applications, and ClearSpeed's accelerators for matrix operations. While accelerators provide improved performance and efficiency over CPUs, programming them can be more difficult due to their specialized architectures.

1. HIGH PERFORMANCE COMPUTING WITH
ACCELARATORS

2. CONTENTS
 INTRODUCTION
 STATE OF AFFAIRS
 ACCELERATORS
 TYPES OF ACCELERATORS
 GPU
 FPGAs
 IBM CELL
 CONCLUSION
 REFERENCE

3. INTRODUCTION
 High performance computing (HPC) uses
supercomputers and computer clusters to solve
advanced computing problems.
 Applied in business uses : data warehouses, transaction
process etc.
 A synonym for supercomputers.
 Supercomputer is used to refer a more powerful subset
of HPC.
 Hardware acceleration.

4. STATE OF AFFAIRS
 In the past few years, a new class of HPC systems has
emerged.
 Unconventional processor —such as IBM's Cell
processor and graphics processing units (GPUs)—for
heavy computations
 Conventional CPU mostly for non-compute-intensive
tasks, such as I/O and communication.
 e.g. : Roadrunner and Tianhe-1.

5.  Alternative technologies includes:
 General purpose Graphical Processing Units (GPGPUs)
 Field Programmable Gate Arrays (FPGAs) boards.
 Clear Speed Floating Point boards
 IBM cell processors

6. ACCELERATOR
 Accelerators are computing components containing
functional units, together with memory and control
systems.
 An ideal accelerator has the following characteristics:
1. Much faster and easy to program.
2. Improves price/performance .
3. Improves performance/watt.

7. FEATURES OF
ACCELERATORS
 A slow clock period compared to CPUs.
 Aggregate high performance is achieved through
parallelism.
 Transferring data between the accelerators and CPUs is
slow compared to the memory bandwidth available for
the primary processors.
 Needs lots of data reuse for good performance.
 The fewer the bits, the better the performance.

8.  Integer is faster than 32-bit floating-point which is faster
than 64-bit floating-point.
 Learning the theoretical peak is difficult.
 Software tools lacking.
 Requires programming in languages designed for the
particular technology.

9. TYPES OF
ACCELERATORS
1. GPU (Graphics Processing Unit)
 A specialized microprocessor that offloads and
accelerates 3D or 2D graphics rendering from the
microprocessor.
 Used in embedded systems, mobile phones, personal
computers, workstations, and game consoles.
 It is a processor attached to a graphics card dedicated to
calculating floating point operation.

10.  HPC GPU consumes more energy than CPU.
 GPU doing floating point code use only half the power
of one doing pure graphics code.
 Max memory per GPU is about 1 GB.

11. 2. FPGAs( Field Programmable Gate Arrays)
 They are programmable.
 FPGAs chips seem primitive compared to standard
CPUs.

12.  Compared to modern CPUs, FPGAs run at very modest
speeds.
 Running code on FPGAs is difficult as it involves some
steps that are not necessary for CPUs.
 HARDWARE:-Some are designed to perform integer
calculations and others for floating point calculations.

13.  SOFTWARE:- There is a spectrum of FPGA software
development tools.
 At one end is the popular hardware design language.
 Verilog
 The smaller the number of bits, the less space required.
 So more can be contained on a chip and hence better
performance.

14.  3.CLEAR SPEED’S FLOATING POINT ACCELERATORS
 Two ways to use : one is to make a call to a routine from
their math library.
 And other way to access it is to write routines using the
Clear speed accelerator language Cn.
 Some are designed to perform integer calculations and
others are for floating-point calculation.
 Produces a board that is designed to accelerate floating-
point calculations

15. 4. IBM CELL:
 Cell is shorthand for Cell Broadband Engine
Architecture.
 It is a microprocessor designed to bridge the gap
between conventional desktop processors.

17.  CELL ARCHITECTURE:
 Cell has a total of nine cores and is a heterogeneous
multiprocessor with a unique design.
 First major commercial application of Cell was in
Sony’s PlayStation 3.

18. CONCLUSION
 There are multiple families of accelerators
suitable for executing applications from portions
of HPC space
 These include GPGPUs, FPGAs, ClearSpeed and
the Cell processor.
 They are best at the following:
 GPGPUs: graphics, 32-bit floating-point .
 FPGAs: embedded applications, applications that
require a small number of bits.
 Clear speed: matrix-matrix multiplication, 64-bit
floating-point .

19. REFERENCE
 http://www.computer.org/portal/web/csdl/maga
zines/cise#4
 http://www.hp.com/techservers/hpccn/hpccolla
boration/ADCatalyst/downloads/accel erators.pdf
 http://en.wikipedia.org/wiki/Graphics_processin
g_unit
 http://www.scientificcomputing.com/HPC-
Future.aspx
 http://www.nvidia.com/object/fermi_architectur
e.html
 http://www.top500.o

20. THANK YOU