I have a budget for my computing facilities and I wanted to run some MD simulations. I can afford to either get a computer with the 5th generation of the Intel i7 processor and 16GB RAM, or, the 4th generation Intel i7 processor with 32G RAM. Does anyone know if there would be a large difference in running MD simulations between these two choices?
I would suggest to go for advanced processor. memory you can upgrade later also.
First some general remarks:
I know it is about money but it is not quite effective at all to run MD simulations on mobile or PC CPU architectures - they are designed mainly for laptops and office workstations.
On the model comparison:
Intel i7-4790 might serve you better because it provides more cores. The 5-th generation of Intel CPU-s (i7-5650U, i7-5600U, i7-5557U, i7-5550U, i7-5500U) operates only with two cores, which is absolutely inapplicable for serious MD computations. It is hard even to pin a process to a core. Also you cannot optimize PME there, which is critical (even two-node PME schema will be hard to run).
What to do in the future (if you get more money)?
If you insist to use i7 CPUs it is better to buy i7-5960X. This is the only Intel i7 CPU architecture that could allow you to run effectively MD. But ... I don't think that you should use i7 CPUs for MD at all. PME optimization is a problem there.
In most HPC facilities across EU the computational nodes are based on Xeon CPU E5 family. Two weeks ago we implemented E5-2699 v3 - one of the best CPUs for HPC. It provides 18 cores:
http://www.amazon.com/Xeon-E5-2699-Octadeca-core-Core-Processor/dp/B00PDD1ZX0
http://www.cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5-2699+v3+%40+2.30GHz
If you combine it with GPU like NVidia Kepler it is possible to achieve great performance. And it is cost-effective.
As for the memory:
More is better, but using ECC memory is much more effective.
Please check CPU benchmark at http://www.cpubenchmark.net/. But the speed depends on the calculation you are going to do. For ordinary MD simulation, 2GB memory for 1 core or 1 threading is enough. But if you want to do some other calculation such as entropy calculation. You need much bigger memory, which might be more than 20GB per threading. I think every university should have HPCC, this is a good solution. And you can also apply to use XSEDE. No matter how powerful your personal server is, it is hardly comparable with a HPCC.
I would suggest to go for advanced processor. memory you can upgrade later also.
For the calculations I would rather choose Xeon, even a lower model.
The combination Core i7-5960X with Geforce GTX 980 (or Titan) and overclocking.
You must have a software that can support the advantage of the combination of multicore and GPU.
You have not written what type of MD you want to do. If possible to use GPU, it does not matter how much memory is available for the CPU. 4 GB is sufficient.
However, a purely classical calculation rarity, but quantum requires a lot of memory.
@Th. Homann
"You must have a software that can support the advantage of the combination of multicore and GPU."
What program do you have in mind?
Can you please suggest a suitable configuration of PC required for running MD simulations using GROMACS and NAMD.
Regards
- Badges
- Science topic