GPU Benchmarks

First I got some idea about GPU Pricing in last week's post. Then I got some idea about the AI engines' operating environment in this week's post, Stockfish Wins TCEC Season 13. Both the CCCC and the TCEC offer public information about the setups they use in support of the AI/NN engines.

Computer Chess Championship (chess.com):-

GPU: 4 x Tesla V100 (64 GB GPU memory)
CPU: Intel Xeon @ 2.70GHz
RAM: 256 GB

TCEC Season 14 (chessdom.com):-

GPU: 1 x 2080 ti + 1 x 2080
CPU: Quad Core i5 2600k
RAM: 16GB DDR3-2133

I've listed only the hardware that allows a comparison of the two setups. The V100s used by CCCC were first offered by Nvidia in 2017; the 2080s used by TCEC were first offered in 2018. How do they compare? Here's a chart from the same company that provided the numbers I used in 'GPU Pricing'.

October 2018: 2080 Ti TensorFlow GPU benchmarks - 2080 Ti vs V100 vs 1080 Ti vs Titan V (lambdalabs.com) • 'The 2080 Ti comes out on top as the best GPU in 2018 for training neural nets.'

Although the 2080s don't offer the same throughput as the V100, a cost/benefit comparison improves in favor of the 2080s when you factor in the price of the GPUs. Note that these numbers are for training the NNs listed to the right of the top chart. The performance of a system running a specific chess engine with its trained weights would be different.