[gmx-users] Tesla K40 and K80 performance for Gromacs

[Zheng Ruan]

Hi gromacs users,

Our lab decides to buy some exclusive computational resources for MD
using gromacs. The HPC center in our university offers 2 options. I
guess some of you are using K80 GPU right now so I'd like to hear your
suggestions/experiences about it.

option 1 (Tesla K40):
* Dual Intel Xeon E5-2680 v3, 12C, 2.5GHz, 2133MHz RAM, 120W
* 128GB, DDR4-2133 ECC, 2R (8 x 16GB)
* 256GB SSD, 2.5", SATA, 6Gbps, MLC
* NVIDIA Tesla K40m
* InfiniBand HCA, True Scale QLE7340, 1-port QSFP QDR memfree
* dual power supply
* 5-Year Warranty
* 5-Year GPU Support
$14,228.22 total

option 2 (Tesla K80):
* Dual Intel Xeon E5-2680 v3, 12C, 2.5GHz, 2133MHz RAM, 120W
* 128GB, DDR4-2133 ECC, 2R (8 x 16GB)
* 256GB SSD, 2.5", SATA, 6Gbps, MLC
* NVIDIA Tesla K80m
* InfiniBand HCA, True Scale QLE7340, 1-port QSFP QDR memfree
* dual power supply
* 5-Year Warranty
* 5-Year GPU Support
$15,524.18 total

We are now running MD for systems with about 40000 atoms and getting
100ns/day performance using 2 K40 GPU. Will K80 GPU act like 2 K40 as
advertised in the website (http://www.gromacs.org/)?

Thank you!

Ruan