Hi, I wonder if guys you can help me setup a decent server that is capable calculating molecular dynamics calculations pretty well. So here is the problem:
I used Gromacs (software that does these kinds of simulations) to run a water+ion+protein system with about 70k atoms on a 24 core server (not sure what kind of CPUs) and it was able to simulate about 10 ns of the 70k water+ion+protein system in a about a day's worth of time.
For these kinds of simulations to yield biologically relevant data, I need to run the system for about at least a microsecond. That means that I need to run the simulation for about 100 day on that 24 core system. It is not worth it.
Now the Gromacs team have written code that can make use of Nvidia's GPU together with CPUs. See this presentation if you are interested.
To give you an idea of how much this speeds up these simulation, I have run the same 70k water+ion+protein system on my desktop with a Xeon 1230v2 chip and an evga 660ti GPU and it yielded about 25 ns per day. But this is still not enough.
So I am thinking of applying for a research grant that allows for purchases of assets as I would like to train post-graduate students (still looking for them
) on these desktop servers.
I was thinking of the following build:
CPU: 2x Intel Xeon™ E5-2630 - 2.30GHz Six Core (12 Thread), Socket 2011, 15MB L3 Cache
GPU: 4x GIGABYTE GV-N760OC-2GD, GTX760
MB: Asus Z9PE-D8 WS LGA2011 (Dual Xeon)
RAM: 2GB per core so --> 3x Kingston KVR13E9K3/24I ValueRam 3x8GB DDR3-1333 1.5V CL9 ECC
It comes in at R37500 and should do membrane proteins with 120k atoms at about 100ns per day, so a ten day simulation would be worth it. If I can get closer to about R30K it would be great (perhaps a cheaper GPU such as a 660ti? or cheaper RAM?)
Any suggestions would be great. Thanks
.
I used Gromacs (software that does these kinds of simulations) to run a water+ion+protein system with about 70k atoms on a 24 core server (not sure what kind of CPUs) and it was able to simulate about 10 ns of the 70k water+ion+protein system in a about a day's worth of time.
For these kinds of simulations to yield biologically relevant data, I need to run the system for about at least a microsecond. That means that I need to run the simulation for about 100 day on that 24 core system. It is not worth it.
Now the Gromacs team have written code that can make use of Nvidia's GPU together with CPUs. See this presentation if you are interested.
To give you an idea of how much this speeds up these simulation, I have run the same 70k water+ion+protein system on my desktop with a Xeon 1230v2 chip and an evga 660ti GPU and it yielded about 25 ns per day. But this is still not enough.
So I am thinking of applying for a research grant that allows for purchases of assets as I would like to train post-graduate students (still looking for them
I was thinking of the following build:
CPU: 2x Intel Xeon™ E5-2630 - 2.30GHz Six Core (12 Thread), Socket 2011, 15MB L3 Cache
GPU: 4x GIGABYTE GV-N760OC-2GD, GTX760
MB: Asus Z9PE-D8 WS LGA2011 (Dual Xeon)
RAM: 2GB per core so --> 3x Kingston KVR13E9K3/24I ValueRam 3x8GB DDR3-1333 1.5V CL9 ECC
It comes in at R37500 and should do membrane proteins with 120k atoms at about 100ns per day, so a ten day simulation would be worth it. If I can get closer to about R30K it would be great (perhaps a cheaper GPU such as a 660ti? or cheaper RAM?)
Any suggestions would be great. Thanks
Last edited: