Quick sanity check please! (GTX 1080 performance, CUDA Nbody sample)

Accelerated Computing CUDA CUDA Programming and Performance
1

CUDA 8 SDK Final, 373.06 drivers, Windows 10 Anniversary update
Intel i7-5930K, 32GB RAM

If you have a GTX 1080 system, could you please confirm whether these performance figures are normal or not? I’m not trying to conduct the world’s most rigorous benchmark, but regardless a quick A/B would be very helpful. Many thanks in advance.

nbody.exe -benchmark -numdevices=1

number of CUDA devices = 1

Windowed mode
Simulation data stored in video memory
Single precision floating point simulation
1 Devices used for simulation
GPU Device 0: “GeForce GTX 1080” with compute capability 6.1

Compute 6.1 CUDA device: [GeForce GTX 1080]
20480 bodies, total time for 10 iterations: 18.530 ms
= 226.348 billion interactions per second
= 4526.967 single-precision GFLOP/s at 20 flops per interaction

nbody.exe -benchmark -numdevices=2

number of CUDA devices = 2

Windowed mode
Simulation data stored in system memory
Single precision floating point simulation
2 Devices used for simulation
GPU Device 0: “GeForce GTX 1080” with compute capability 6.1

Compute 6.1 CUDA device: [GeForce GTX 1080]
Compute 6.1 CUDA device: [GeForce GTX 1080]
40960 bodies, total time for 10 iterations: 40.441 ms
= 414.858 billion interactions per second
= 8297.167 single-precision GFLOP/s at 20 flops per interaction

2

On my system (Win 7 x64, GTX 1080, GTX 690, CUDA 8.0, 16 GB RAM, Intel Core i7-3770K), I get the following results:

number of CUDA devices  = 1
> Windowed mode
> Simulation data stored in video memory
> Single precision floating point simulation
> 1 Devices used for simulation
GPU Device 0: "GeForce GTX 1080" with compute capability 6.1

> Compute 6.1 CUDA device: [GeForce GTX 1080]
20480 bodies, total time for 10 iterations: 19.954 ms
= 210.202 billion interactions per second
= 4204.044 single-precision GFLOP/s at 20 flops per interaction
3

Thank you phw89! Much appreciated.

4 
number of CUDA devices  = 2
> Windowed mode
> Simulation data stored in system memory
> Single precision floating point simulation
> 2 Devices used for simulation
GPU Device 0: "GeForce GTX 1080" with compute capability 6.1

> Compute 6.1 CUDA device: [GeForce GTX 1080]
> Compute 6.1 CUDA device: [GeForce GTX 1080]
40960 bodies, total time for 10 iterations: 7385.681 ms
= 2.272 billion interactions per second
= 45.432 single-precision GFLOP/s at 20 flops per interaction

I’m not getting anywhere near that for some reason :(
Any ideas? Have you guys seen this before?

I’m running on Windows 10

5

rgee27, I accidentally ran this sample in debug mode the other day and got the same performance figure. Recompile in release mode and things should be back to normal.

6

RobbieBC, Thanks! That did the trick.

number of CUDA devices  = 2
> Windowed mode
> Simulation data stored in system memory
> Single precision floating point simulation
> 2 Devices used for simulation
GPU Device 0: "GeForce GTX 1080" with compute capability 6.1

> Compute 6.1 CUDA device: [GeForce GTX 1080]
> Compute 6.1 CUDA device: [GeForce GTX 1080]
40960 bodies, total time for 10 iterations: 46.595 ms
= 360.064 billion interactions per second
= 7201.283 single-precision GFLOP/s at 20 flops per interaction