Testing the new GeForce GTX Titan
We recently tested the new NVIDIA Kepler boards for our GPGPU flow solver elbe. After initial tests on the Kepler K20 boards were very promising and led to speed-up factors of up to 2.5x (without any code modifications), we decided to purchase the new GeForce GTX Titan board to see if the consumer hardware can keep up with the professional boards. This consumer card features 2688 streaming processors, 6 GB device memory and single- and double-precision compute capability.
Running the elbe code for both 2D/3D and single/double precision calculations, we found a performance of up to 1.5 GNUPS (giga node updates per second), 2.5x more than on our recent Tesla C2075 GPGPUs, for a D2Q9 singlephase LBM calculation.