NVIDIA GeForce RTX 2080 Ti SLI Review with NVLink


SiSoftware Sandra GPGPU Testing

We fired up SiSoftware Sandra Titanium 2018 SP2 (version 28.28) to look at GPGPU performance on the new NVIDIA Turing GPU and to see how the RTX 2080 Ti performs in SLI. This benchmark isn’t really aimed at gamers, but some will likely find the results interesting.

The first test that we ran was the Scientific Analysis benchmark and it should be noted that used CUDA for the test. We set the Floating-Point precision to ‘normal’ and we charted the results for N-Body Simulations (NBDY), Fast Fournier Transformation (FFT) and General Matrix Multiply (GEMM) operations.

We got some pretty interesting results as the performance doubled on the N-Body Simulation and General Matrix Multiply operation tests.

We also ran the GPGPU Memory Bandwidth test and noticed that SLI didn’t improve the interface transfer bandwidth at all, but did manage to more than double the internal memory bandwidth. A single GeForce RTX 2080 Ti FE was found to have 489.3 GB/s of internal memory bandwidth and when two are run in SLI that number jumps up to 979.5 GB/s. We aren’t sure exactly what this benchmark is testing, but the results look really good for a multi-GPU setup like the RTX 2080 Ti in SLI.