• Home

  • Documents

  • Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity...
23
Turning software into computer chips - Hastlayer Zoltán Lehóczky @ Lombiq GPU Day 11.07.2019

Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

  • Upload
    others

  • View
    15

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

Turning software into computer chips - Hastlayer

Zoltán Lehóczky @ Lombiq

GPU Day 11.07.2019

Page 2: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

2

Page 3: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

3

Page 4: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

Warning

Hastlayer is currently in alpha stage!

4

Page 5: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

What’s Hastlayer?

5

Page 6: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

computer program → computer chip (hardware)

6

Page 7: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

logic expressed as software → logic expressed as hardware

7

Page 8: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

…with FPGAs

• Field-Programmable Gate Array

• Can behave like any other chip (with limitations)

• Can dynamically be „re-wired”

8

Image by SparkFun Electronics, Boulder, USA

Page 9: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

FPGAs?

• Found in routers, X-ray machines, self-driving cars…

• Compute accelerator

• You need to be a hardware engineer to utilize them

9

Page 10: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

computer program → FPGA logic

10

Page 11: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

.NET (C#, VB, C++, F#, Python, PHP, JavaScript…) → FPGA logic

11

Page 12: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

What’s the point?

12

Page 13: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

CPU vs GPU vs FPGA

13

Parallelism Program complexity Power efficiency

CPU GPU FPGA

Page 14: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

But!

14

CPU GPU FPGA FPGA with Hastlayer

How hard to learn?

Page 15: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

The benefits of FPGAs for us all

• Performance increase for parallel compute-bound algorithms

• Higher power efficiency

• Still only software development

15

Page 16: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

Demo: Hands-on Hastlayer

16

Page 17: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

What’s next?

17

Page 18: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

Project Catapult support

• Microsoft’s FPGA platform in TACC and Azure

• Hastlayer is in the Project Catapult Academic Program

• 8GB vs 100MB RAM, PCIe vs USB, ~10x capacity

• Azure networking, Project Brainwave, direct FPGA access?

• Let me know if you’d like a preview!

18

Page 19: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

Posit number format

• https://hastlayer.com/arithmetics

• Better range/accuracy than IEEE float

• We already have a posit „processor”

19

Page 20: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

And more!

• More use-cases

• Partnership with the Wigner Research Centre for Physics

• FPGAs in every datacenter (AWS, Azure, Intel)

20

Page 21: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

Wrapping up

21

Page 22: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

I like this, how do I start?

• Check out the SDK: https://github.com/Lombiq/Hastlayer-SDK/

• Be ready for an FPGA-filled future!

22

Page 23: Turning software into hardware - Hastlayer · CPU vs GPU vs FPGA 13 Parallelism Program complexity Power efficiency CPU GPU FPGA. But! 14 CPU GPU FPGA FPGA with Hastlayer How hard

Are you ready to *be* the hardware?

• [email protected]

• https://hastlayer.com

• https://github.com/Lombiq/Hastlayer-SDK/

• https://lombiq.com

23


LabVIEW FPGA and GPU

LabVIEW FPGA and GPU

Documents
Agenda CPU Threads Flip Queue CPU Queues GPU Hardware Queue

Agenda CPU Threads Flip Queue CPU Queues GPU Hardware Queue

Documents
Extending Unified Parallel C for GPU Computing · PGAS Programming Model for Hybrid Multi-Core Systems Computer Node CPU Memory GPU GPU Memory CPU CPU GPU GPU Memory Computer Node

Extending Unified Parallel C for GPU Computing · PGAS Programming Model for Hybrid Multi-Core Systems Computer Node CPU Memory GPU GPU Memory CPU CPU GPU GPU Memory Computer Node

Documents
Speedup Altair RADIOSS Solvers Using NVIDIA GPU - GTC ......0 250 500 750 1000 Xeon Nehalem CPU Nehalem CPU + Tesla GPU Lower is better 4 core CPU 1 core + 1 GPU CPU 2 core + 2 GPU

Speedup Altair RADIOSS Solvers Using NVIDIA GPU - GTC ......0 250 500 750 1000 Xeon Nehalem CPU Nehalem CPU + Tesla GPU Lower is better 4 core CPU 1 core + 1 GPU CPU 2 core + 2 GPU

Documents
High Performance Embedded Computing Module Enabled by GPU, CPU and FPGA …ieee-hpec.org/2013/index_htm_files/fritz_CEI_IEEE_HPEC13.pdf · 2013. 11. 20. · Enabled by GPU, CPU and

High Performance Embedded Computing Module Enabled by GPU, CPU and FPGA …ieee-hpec.org/2013/index_htm_files/fritz_CEI_IEEE_HPEC13.pdf · 2013. 11. 20. · Enabled by GPU, CPU and

Documents
GPUfs: Integrating a file system with GPUs · CPU-GPU CPU GPUs Accelerated applications OS. Mark Silberstein - UT Austin 4 Software-hardware gap is widening Manycore processors FPGA

GPUfs: Integrating a file system with GPUs · CPU-GPU CPU GPUs Accelerated applications OS. Mark Silberstein - UT Austin 4 Software-hardware gap is widening Manycore processors FPGA

Documents
HetExchange: Encapsulating heterogeneous CPU-GPU

HetExchange: Encapsulating heterogeneous CPU-GPU

Documents
Design GPU Systems for Hyperscalers ,Diverse AI ... Y… · Inspur AI Market Share Vertical AI Solutions Caffe-MPI AIStation T-Eye GPU Server CPU Server FPGA Accelerator End-to-End

Design GPU Systems for Hyperscalers ,Diverse AI ... Y… · Inspur AI Market Share Vertical AI Solutions Caffe-MPI AIStation T-Eye GPU Server CPU Server FPGA Accelerator End-to-End

Documents
REDEFINING THE ROLE OF THE CPU IN ERA OF CPU ...cseweb.ucsd.edu/~marora/files/papers/cpu-gpu-ieeemicro...REDEFINING THE ROLE OF THE CPU IN THE ERA OF CPU-GPU INTEGRATION

REDEFINING THE ROLE OF THE CPU IN ERA OF CPU ...cseweb.ucsd.edu/~marora/files/papers/cpu-gpu-ieeemicro...REDEFINING THE ROLE OF THE CPU IN THE ERA OF CPU-GPU INTEGRATION

Documents
GPU Computing April 2009. GPU Outpacing CPU in Raw Processing GPU NVIDIA GTX 285 240 cores 1.04 TFLOPS CPU GPU CUDA Architecture Introduced DP HW Introduced

GPU Computing April 2009. GPU Outpacing CPU in Raw Processing GPU NVIDIA GTX 285 240 cores 1.04 TFLOPS CPU GPU CUDA Architecture Introduced DP HW Introduced

Documents
GPU Analysis and Optimisation - Peoplepeople.maths.ox.ac.uk/gilesm/cuda/lecs/NV_Profiling_lowres.pdf · inter-node/inter-process communication CPU-GPU communication CPU/GPU performance

GPU Analysis and Optimisation - Peoplepeople.maths.ox.ac.uk/gilesm/cuda/lecs/NV_Profiling_lowres.pdf · inter-node/inter-process communication CPU-GPU communication CPU/GPU performance

Documents
Get moving with CMC FPGA/GPU Cluster · 2019. 12. 12. · FPGA GPU CPU CPU FPGA GPU CPU CPU FPGA GPU CPU CPU FPGA GPU CPU 2 x CPU 2 x CPU 2 x CPU 2 x CPU 8 x CPU 2 x CPU & Large memory

Get moving with CMC FPGA/GPU Cluster · 2019. 12. 12. · FPGA GPU CPU CPU FPGA GPU CPU CPU FPGA GPU CPU CPU FPGA GPU CPU 2 x CPU 2 x CPU 2 x CPU 2 x CPU 8 x CPU 2 x CPU & Large memory

Documents
Cpu Fpu Gpu Apu

Cpu Fpu Gpu Apu

Documents
Simplifying Computer Vision Application Development · (lighting noise, etc.) • Consideration of HW platform • FPGA? CPU? DSP? GPU? • Speed and resource requirement • Resolution,

Simplifying Computer Vision Application Development · (lighting noise, etc.) • Consideration of HW platform • FPGA? CPU? DSP? GPU? • Speed and resource requirement • Resolution,

Documents
Making the most out of Heterogeneous Chips with CPU, GPU and FPGA

Making the most out of Heterogeneous Chips with CPU, GPU and FPGA

Engineering
FPGA Acceleration of LFRic - EMiT · • “GPU vs FPGA Performance Comparison”, Berton White Paper •GPU: 0.07-0.12 vs. FPGA: 0.23 €/Gflop/s/W •GPU: 20 vs. FPGA: 70 Gflops/W

FPGA Acceleration of LFRic - EMiT · • “GPU vs FPGA Performance Comparison”, Berton White Paper •GPU: 0.07-0.12 vs. FPGA: 0.23 €/Gflop/s/W •GPU: 20 vs. FPGA: 70 Gflops/W

Documents
GPU Design on FPGA

GPU Design on FPGA

Design
5g, gpu and fpga

5g, gpu and fpga

Mobile
AMD’s Uni ed CPU & GPU Processor Concept€¦ · Uni ed CPU & GPU Processor Concept Sven Nobis Introduction Background CPU vs. GPU OpenCL & CUDA Related Work The way to HSA Heterogeneous

AMD’s Uni ed CPU & GPU Processor Concept€¦ · Uni ed CPU & GPU Processor Concept Sven Nobis Introduction Background CPU vs. GPU OpenCL & CUDA Related Work The way to HSA Heterogeneous

Documents
Computing Performance Benchmarks among CPU, GPU, and FPGA

Computing Performance Benchmarks among CPU, GPU, and FPGA

Documents
Gpu vs fpga

Gpu vs fpga

Technology
cpu and GPU

cpu and GPU

Documents
Build GPU Cluster Hardware for Efficiently Accelerating ... · Hardware GPU dense HPC Cluster CPU Host RAM GPU #5 PHB IB Card Host RAM CPU GPU #0 PHB IB Card GPU #1 GPU #4 GPU #2

Build GPU Cluster Hardware for Efficiently Accelerating ... · Hardware GPU dense HPC Cluster CPU Host RAM GPU #5 PHB IB Card Host RAM CPU GPU #0 PHB IB Card GPU #1 GPU #4 GPU #2

Documents
GPU Architecture and Programming. GPU vs CPU

GPU Architecture and Programming. GPU vs CPU

Documents
Selective GPU Caches to Eliminate CPU–GPU HW Cache Coherence

Selective GPU Caches to Eliminate CPU–GPU HW Cache Coherence

Documents
Altera Cyclone V: The Marriage of CPU and FPGA ... · FPGA Devices +CPU Memory FPGA CPU Devices Devices Memory Memory FIGURE 3 Integrating the processor and FPGA into a single SoC

Altera Cyclone V: The Marriage of CPU and FPGA ... · FPGA Devices +CPU Memory FPGA CPU Devices Devices Memory Memory FIGURE 3 Integrating the processor and FPGA into a single SoC

Documents
Accelerated ML on cloud FPGAs · CPU vs GPU vs FPGA 23 A GPU is effective at processing the same set of operations in parallel – single instruction, multiple data (SIMD). An FPGA

Accelerated ML on cloud FPGAs · CPU vs GPU vs FPGA 23 A GPU is effective at processing the same set of operations in parallel – single instruction, multiple data (SIMD). An FPGA

Documents
Accelerating Robot Dynamics Gradients on a CPU, GPU, and FPGA

Accelerating Robot Dynamics Gradients on a CPU, GPU, and FPGA

Documents
The Evolution of Modern Parallel Computing · The Evolution of Modern Parallel Computing. 2 GPU Computing Is not vs. It is + CPU CPU GPU GPU. 3 GPU Computing Application Code + GPU

The Evolution of Modern Parallel Computing · The Evolution of Modern Parallel Computing. 2 GPU Computing Is not vs. It is + CPU CPU GPU GPU. 3 GPU Computing Application Code + GPU

Documents
Computing Performance Benchmarks among CPU, GPU, and FPGA · 2.2.2. GPU Architecture and Parallelism..... 12 2.3. FPGA Background ... astonishing performance using the hundreds of

Computing Performance Benchmarks among CPU, GPU, and FPGA · 2.2.2. GPU Architecture and Parallelism..... 12 2.3. FPGA Background ... astonishing performance using the hundreds of

Documents