An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

16
An Analytical Model for a GPU

  • date post

    21-Dec-2015

  • Category

    Documents

  • view

    236

  • download

    2

TAGS:

Transcript of An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Page 1: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

An Analytical Model for a GPU

Page 2: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Overview

Page 3: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

SVM Kernel Behavior: Need for other metrics

Page 4: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Degree of Parallelism

GPU Architecture Each SM executes multiple warps in a time-

sharing fashion while one or more are waiting for memory values

Hiding the execution cost of warps that are executed concurrently.

How many memory requests can be serviced and how many warps can be executed together while one warp is waiting for memory values.

Page 5: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

MWP and CWP

Memory Warp: The warp that is waiting for memory values

Memory Warp waiting period: The time period from right after one warp sent memory

requests until all the memory requests from the same warp are serviced.

CWP (Computation Warp Parallelism) represents the number of warps that the SM processor can

execute during one memory warp waiting period plus one. MWP (Memory Warp Parallelism)

represents the maximum number of warps per SM that can access the memory simultaneously during memory warp waiting period

Page 6: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Relationship between MWP and CWP: CWP > MWP

What is getting hidden?

Total execution time (a) 8 warps (b) 4 warps

Page 7: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

What is going on here?

Page 8: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Relationship between MWP and CWP: MWP > CWP

Total execution time (a) 8 warps (b) 4 warps

Page 9: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Relationship between MWP and CWP: MWP > CWP

Total execution time (8 warps)

Page 10: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Not Enough Warps Running

Page 11: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

CPI

PTX instruction set

Page 12: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Model

Page 13: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

ExampleBlocks: 80Threads: 1285 blocks per SM

Page 14: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.
Page 15: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.
Page 16: An Analytical Model for a GPU. Overview SVM Kernel Behavior: Need for other metrics.

Fast and Accurate Refined Nystrom Based Kernel SVM · Fast and Accurate Refined Nystrom Based Kernel SVM¨ Zhe Li,1 Tianbao Yang,1 Lijun Zhang,2 and Rong Jin3 1Department of Computer

Fast and Accurate Refined Nystrom Based Kernel SVM · Fast and Accurate Refined Nystrom Based Kernel SVM¨ Zhe Li,1 Tianbao Yang,1 Lijun Zhang,2 and Rong Jin3 1Department of Computer

SVM and Kernel machine Lecture 1: Linear SVM...A support vector machine (SVM) is a linear classifier associated with the following decision function: D(x) = sign w ⊤x+b where w

SVM and Kernel machine Lecture 1: Linear SVM...A support vector machine (SVM) is a linear classifier associated with the following decision function: D(x) = sign w ⊤x+b where w

Protein-Protein Interaction using SVM based kernel,Jacob Coefficient and Gene Ontology

Protein-Protein Interaction using SVM based kernel,Jacob Coefficient and Gene Ontology

Object Classification of Satellite Images Using Cluster Repulsion Based Kernel Fcm and Svm Classifier

Object Classification of Satellite Images Using Cluster Repulsion Based Kernel Fcm and Svm Classifier

A Realtime GPU Subdivision Kernel - cise.ufl.edu · PDF fileA Realtime GPU Subdivision Kernel ... Computer Graphics—Software Support I.3.5 ... the viewport is initialized to contain

A Realtime GPU Subdivision Kernel - cise.ufl.edu · PDF fileA Realtime GPU Subdivision Kernel ... Computer Graphics—Software Support I.3.5 ... the viewport is initialized to contain

55. IWK - db-thueringen.de · - SVM-Type, here the Types C-SVC, nu-SVC, one-class SVM, epsilon-SVR and can choose nu-SVR, - SVM-Kernel, here the Kernels linear,

55. IWK - db-thueringen.de · - SVM-Type, here the Types C-SVC, nu-SVC, one-class SVM, epsilon-SVR and can choose nu-SVR, - SVM-Kernel, here the Kernels linear,

332 Advanced Computer Architecturephjk/AdvancedComp... · –Serial CPU code –Parallel GPU code (kernels) •GPU kernel is a C function –Each thread executes kernel code –A

332 Advanced Computer Architecturephjk/AdvancedComp... · –Serial CPU code –Parallel GPU code (kernels) •GPU kernel is a C function –Each thread executes kernel code –A

R-SVM+: Robust Learning with Privileged InformationIn[Li et al., 2016], two new algorithms are proposed to ef ciently solve linear SVM+ and kernel SVM+ which uses the L2-loss based

R-SVM+: Robust Learning with Privileged InformationIn[Li et al., 2016], two new algorithms are proposed to ef ciently solve linear SVM+ and kernel SVM+ which uses the L2-loss based

ECE595 / STAT598: Machine Learning I Lecture 20 Support ...Outline Support Vector Machine Lecture 19 SVM 1: The Concept of Max-Margin Lecture 20 SVM 2: Dual SVM Lecture 21 SVM 3: Kernel

ECE595 / STAT598: Machine Learning I Lecture 20 Support ...Outline Support Vector Machine Lecture 19 SVM 1: The Concept of Max-Margin Lecture 20 SVM 2: Dual SVM Lecture 21 SVM 3: Kernel

Object Classification of Satellite Images Using Cluster Repulsion Based Kernel Fcm and Svm Classifi

Object Classification of Satellite Images Using Cluster Repulsion Based Kernel Fcm and Svm Classifi

Support Vector Machines and Kernel Based Learning · 2007. 9. 6. · Kernel based learning: interdisciplinary challenges SVM & kernel methods linear algebra mathematics statistics

Support Vector Machines and Kernel Based Learning · 2007. 9. 6. · Kernel based learning: interdisciplinary challenges SVM & kernel methods linear algebra mathematics statistics

Kernel Weaver: Automatically Fusing Database Primitives for Efficient GPU Computation

Kernel Weaver: Automatically Fusing Database Primitives for Efficient GPU Computation

Support Vector Machines - cedar.buffalo.edusrihari/CSE574/Chap6/Chap6.4-SVMs.pdf · Alternative Names for SVM • Also called Sparse kernel machines • Kernel methods predict based

Support Vector Machines - cedar.buffalo.edusrihari/CSE574/Chap6/Chap6.4-SVMs.pdf · Alternative Names for SVM • Also called Sparse kernel machines • Kernel methods predict based

Kernel Machines - From linear to Kernel SVM€¦ · Road map 1 Intuition and Motivation 2 Kernel Theoretical framework 3 Kernel functions Kernel on vectors Kernels on generic data

Kernel Machines - From linear to Kernel SVM€¦ · Road map 1 Intuition and Motivation 2 Kernel Theoretical framework 3 Kernel functions Kernel on vectors Kernels on generic data

Support Vector Classi cation and Regressioncjlin/talks/itri.pdf · SVM and kernel methods Outline 1 Introduction 2 SVM and kernel methods 3 Dual problem and solving optimization problems

Support Vector Classi cation and Regressioncjlin/talks/itri.pdf · SVM and kernel methods Outline 1 Introduction 2 SVM and kernel methods 3 Dual problem and solving optimization problems

Physics-Motivated Features for Distinguishing Photographic ... · SVM Classification SVM classification with radial basis function (RBF) kernel. Cartoon feature is the conventional

Physics-Motivated Features for Distinguishing Photographic ... · SVM Classification SVM classification with radial basis function (RBF) kernel. Cartoon feature is the conventional

FPGA-GPU Architecture for Kernel SVM Pedestrian Detection · FPGA-GPU Architecture for Kernel SVM Pedestrian Detection ... tion system based on a multi-sensor platform for intersec-

FPGA-GPU Architecture for Kernel SVM Pedestrian Detection · FPGA-GPU Architecture for Kernel SVM Pedestrian Detection ... tion system based on a multi-sensor platform for intersec-

KERNELS, SVM - cse.iitm.ac.invplab/courses/ML/PDF/SVM_intro.pdf · RBFkernels • Squared exponential kernel(SE kernel) or Gaussiankernel • If 𝚺is diagonal, this can be written

KERNELS, SVM - cse.iitm.ac.invplab/courses/ML/PDF/SVM_intro.pdf · RBFkernels • Squared exponential kernel(SE kernel) or Gaussiankernel • If 𝚺is diagonal, this can be written

Support Vector Machine (SVM) and Kernel Methods

Support Vector Machine (SVM) and Kernel Methods

GPU accelerated Arnoldi solver for small batched matrix › content › gtc-kr › part_2_matrix.pdf · - Fermi GPU : 16, Kepler GPU : 32 concurrent kernel run - Maxwell GPU : Unknown(Help!)

GPU accelerated Arnoldi solver for small batched matrix › content › gtc-kr › part_2_matrix.pdf · - Fermi GPU : 16, Kepler GPU : 32 concurrent kernel run - Maxwell GPU : Unknown(Help!)