Skip to main content

Table 1 GPUs used in this work. The first column indicates the GPU, followed by three columns that show the memory properties. The clock-speed in Mhz in the second, the bus width in bits in the third and the product of the two, the bandwidth in GB/s in the fourth. The fifth column contains the number of compute cores and the sixth their clock-speed in Mhz. The next two columns indicate the theoretical performance in TFlop/s, the single precision performance is in the seventh column and the double precision in the eight column. The next two columns gives the relative performance of each GPU where we set the GTX480 to 1. For the ninth column these numbers are determined using the theoretical peak single precision performance (TPP) of the chips. The tenth column indicates the relative practical single precision peak performance (PPP) which is determined using a simple embarrassingly parallel N -body code

From: Sapporo2: a versatile direct N-body library

 

Memory

Cores

SP

TFlop/s

DP

TFlop/s

TPP

PPP

Mhz

bus

bw

#

Mhz

GTX480

3,696

384

133.9

480

1,401

1.35

0.17

1

1

GTX680

6,008

256

192.2

1,536

1,006

3.09

0.13

2.3

1.7

K20m

5,200

320

208

2,496

706

3.5

1.17

2.6

1.8

GTX Titan

6,144

384

288.4

2,688

837

4.5

1.5

3.35

2.2

HD7970

5,500

384

264

2,048

925

3.8

0.94

2.8

2.3