Performance measure for selection of GPU for GPGPU purposes

[petric3]

I need to buy a workstation for solving numerical problems with GPUs (GPGPU). I have two options to choose from (1) 2xGeforce GTX 1080ti or (2) 3xGeforce GTX 1070. Looking at benchmark results I do not think that the performance measures like 'G3D Mark' and 'G2D Rating' are relevant for my case, but rather the number of CUDA cores and its clock value. So if I do the math for (1) 1080ti: 2GPUs x 3584cores x 1582Mhz = 11339Ghz and for (2) 1070: 3GPUs x 1920cores x 1683Mhz = 9694Ghz, I find out that the 2xGTX1080ti option is about 17% faster and I should go for this one. Kindly comment if my thinking is correct or advice/suggest which option would be better?

[David (PassMark)]

Have a a look at the GPU compute benchmark here.

[petric3]

I saw it, thank you. According to my first post that relates purely to GHz potential, the 2xGTX1080ti is the winner whereas according to GPU compute benchmark, the 1070 is the winner, namely (1) 1080ti: 2GPUs x 9.483 = 18.966 and (2) 1070: 3 x 6.687 = 20.061. Any comments on that?