Relative efficiency of XLA for numerical models (hollow = f64; filled = f32; circles = HEAT1D; triangles = HEAT2D; squares = NS2D; black = PC CPU; red = PC GPU; blue = HPC CPU; magenta = HPC GPU).

Optimal implementation of HEAT2D and NS2D on the GPU platform (double precision; solid black line = benchmarked bandwidth; lines + symbols = effective bandwidth; black = HEAT2D; red = NS2D; blue = XPXPY12_2D; filled circles = CUDA utilising L2 cache; hollow triangles = CUDA utilising shared memory; filled squares = XLA with slice and concatenation; hollow diamonds = XLA with convolution; hollow pentagons = XLA with roll).

Xuzhen He (14676114)

February 2023

Optimal implementation of HEAT2D and NS2D on the GPU platform (double precision; solid black line = ...

Comparison of GPU and CPU efficiency while solving heat conduction problems

Semenenko, Julija
Kolesau, Aliaksei
Starikovičius, Vadimas
Mackūnas, Artūras
Šešok, Dmitrij

November 2020

Overview of GPU usage while solving different engineering problems, comparison between CPU and GPU c...

Optimal implementation of HEAT1D on the GPU platform (double precision; solid black line = benchmarked bandwidth; lines + symbols = effective bandwidth; red = VEC COPY; black = HEAT1D; filled circles = CUDA utilising L2 cache; hollow triangles = CUDA utilising shared memory; filled squares = XLA with slice and concatenation; hollow diamonds = XLA with convolution; hollow pentagons = XLA with roll).

Xuzhen He (14676114)

February 2023

Optimal implementation of HEAT1D on the GPU platform (double precision; solid black line = benchmark...

Relative efficiency of XLA for element-wise operations (hollow = vector operations; filled = matrix operations; circles = f64; triangles = f32; black = PC CPU; red = PC GPU; blue = HPC CPU; magenta = HPC GPU).

Xuzhen He (14676114)

February 2023

Relative efficiency of XLA for element-wise operations (hollow = vector operations; filled = matrix ...

Information of the tested GPU platforms.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

The examined operations.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

Information of the tested CPU platforms.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

Hardware parameters of optimal implementations.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

S1 Data -

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

The GPU computing efficiency of the three different thread arrangements in comparison with the original MATLAB implementation.

Wei-Jen Wang (427148)
I-Fan Hsieh (427149)
Chun-Chuan Chen (427150)

June 2013

The GPU computing efficiency of the three different thread arrangements in comparison with the or...

Pseudocodes for the operations.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

Comparison of computational performance between the FA-MLR using CPU and GPU.

Lauro C. M. de Paula (672575)
Anderson S. Soares (672576)
Telma W. de Lima (672577)
Alexandre C. B. Delbem (672578)
Clarimar J. Coelho (672579)
Arlindo R. G. Filho (672580)

December 2014

Comparison of computational performance between the FA-MLR using CPU and GPU.</p

Model for the maximal effective bandwidth of numerical models (lines = model prediction by Eq 12; symbols = measured maximal effective bandwidth; filled = f64; hollow = f32; black = CPU medium; red = CPU large; blue = GPU medium; magenta = GPU large).

Xuzhen He (14676114)

February 2023

Model for the maximal effective bandwidth of numerical models (lines = model prediction by Eq 12; sy...

Model for the effective bandwidth of element-wise operations (lines = model prediction by Eq 11; symbols = measured maximal effective bandwidth; filled symbols = matrix operations; hollow symbols = vector operations; circles = CPU; triangles = GPU; black = medium f64; red = medium f32; blue = large f64; magenta = large f32).

Xuzhen He (14676114)

February 2023

Model for the effective bandwidth of element-wise operations (lines = model prediction by Eq 11; sym...

Relative Performance Speedup.

Paul Richmond (352031)
Lars Buesing (337191)
Michele Giugliano (227795)
Eleni Vasilaki (33913)

February 2013

Figure shows the relative performance improvement of our GPU model with ...

Optimal implementation of HEAT2D and NS2D on the GPU platform (double precision; solid black line = benchmarked bandwidth; lines + symbols = effective bandwidth; black = HEAT2D; red = NS2D; blue = XPXPY12_2D; filled circles = CUDA utilising L2 cache; hollow triangles = CUDA utilising shared memory; filled squares = XLA with slice and concatenation; hollow diamonds = XLA with convolution; hollow pentagons = XLA with roll).

Xuzhen He (14676114)

February 2023

Optimal implementation of HEAT2D and NS2D on the GPU platform (double precision; solid black line = ...

Comparison of GPU and CPU efficiency while solving heat conduction problems

Semenenko, Julija
Kolesau, Aliaksei
Starikovičius, Vadimas
Mackūnas, Artūras
Šešok, Dmitrij

November 2020

Overview of GPU usage while solving different engineering problems, comparison between CPU and GPU c...

Optimal implementation of HEAT1D on the GPU platform (double precision; solid black line = benchmarked bandwidth; lines + symbols = effective bandwidth; red = VEC COPY; black = HEAT1D; filled circles = CUDA utilising L2 cache; hollow triangles = CUDA utilising shared memory; filled squares = XLA with slice and concatenation; hollow diamonds = XLA with convolution; hollow pentagons = XLA with roll).

Xuzhen He (14676114)

February 2023

Optimal implementation of HEAT1D on the GPU platform (double precision; solid black line = benchmark...

Relative efficiency of XLA for element-wise operations (hollow = vector operations; filled = matrix operations; circles = f64; triangles = f32; black = PC CPU; red = PC GPU; blue = HPC CPU; magenta = HPC GPU).

Xuzhen He (14676114)

February 2023

Relative efficiency of XLA for element-wise operations (hollow = vector operations; filled = matrix ...

Information of the tested GPU platforms.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

The examined operations.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

Information of the tested CPU platforms.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

Hardware parameters of optimal implementations.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

S1 Data -

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

The GPU computing efficiency of the three different thread arrangements in comparison with the original MATLAB implementation.

Wei-Jen Wang (427148)
I-Fan Hsieh (427149)
Chun-Chuan Chen (427150)

June 2013

The GPU computing efficiency of the three different thread arrangements in comparison with the or...

Pseudocodes for the operations.

Xuzhen He (14676114)

February 2023

The recent dramatic progress in machine learning is partially attributed to the availability of high...

Comparison of computational performance between the FA-MLR using CPU and GPU.

Lauro C. M. de Paula (672575)
Anderson S. Soares (672576)
Telma W. de Lima (672577)
Alexandre C. B. Delbem (672578)
Clarimar J. Coelho (672579)
Arlindo R. G. Filho (672580)

December 2014

Comparison of computational performance between the FA-MLR using CPU and GPU.</p

Model for the maximal effective bandwidth of numerical models (lines = model prediction by Eq 12; symbols = measured maximal effective bandwidth; filled = f64; hollow = f32; black = CPU medium; red = CPU large; blue = GPU medium; magenta = GPU large).

Xuzhen He (14676114)

February 2023

Model for the maximal effective bandwidth of numerical models (lines = model prediction by Eq 12; sy...

Model for the effective bandwidth of element-wise operations (lines = model prediction by Eq 11; symbols = measured maximal effective bandwidth; filled symbols = matrix operations; hollow symbols = vector operations; circles = CPU; triangles = GPU; black = medium f64; red = medium f32; blue = large f64; magenta = large f32).

Xuzhen He (14676114)

February 2023

Model for the effective bandwidth of element-wise operations (lines = model prediction by Eq 11; sym...

Relative Performance Speedup.

Paul Richmond (352031)
Lars Buesing (337191)
Michele Giugliano (227795)
Eleni Vasilaki (33913)

February 2013

Figure shows the relative performance improvement of our GPU model with ...

Optimal implementation of HEAT2D and NS2D on the GPU platform (double precision; solid black line = benchmarked bandwidth; lines + symbols = effective bandwidth; black = HEAT2D; red = NS2D; blue = XPXPY12_2D; filled circles = CUDA utilising L2 cache; hollow triangles = CUDA utilising shared memory; filled squares = XLA with slice and concatenation; hollow diamonds = XLA with convolution; hollow pentagons = XLA with roll).

Xuzhen He (14676114)

February 2023

Optimal implementation of HEAT2D and NS2D on the GPU platform (double precision; solid black line = ...

Comparison of GPU and CPU efficiency while solving heat conduction problems

Semenenko, Julija
Kolesau, Aliaksei
Starikovičius, Vadimas
Mackūnas, Artūras
Šešok, Dmitrij

November 2020

Overview of GPU usage while solving different engineering problems, comparison between CPU and GPU c...

Optimal implementation of HEAT1D on the GPU platform (double precision; solid black line = benchmarked bandwidth; lines + symbols = effective bandwidth; red = VEC COPY; black = HEAT1D; filled circles = CUDA utilising L2 cache; hollow triangles = CUDA utilising shared memory; filled squares = XLA with slice and concatenation; hollow diamonds = XLA with convolution; hollow pentagons = XLA with roll).

Xuzhen He (14676114)

February 2023

Optimal implementation of HEAT1D on the GPU platform (double precision; solid black line = benchmark...

Relative efficiency of XLA for numerical models (hollow = f64; filled = f32; circles = HEAT1D; triangles = HEAT2D; squares = NS2D; black = PC CPU; red = PC GPU; blue = HPC CPU; magenta = HPC GPU).

Abstract

Extracted data

Relative efficiency of XLA for numerical models (hollow = f64; filled = f32; circles = HEAT1D; triangles = HEAT2D; squares = NS2D; black = PC CPU; red = PC GPU; blue = HPC CPU; magenta = HPC GPU).

Abstract

Extracted data

Related items

Related items