VASP performance on Intel vs AMD
Posted: Fri Sep 14, 2012 11:15 pm
Hello everyone,
We are trying to buy a new machine for running vasp calculations and we couldn't decide on the type of configuration (especially on the number of cores) we should choose to buy.
We have the following options
1) AMD Opteron 6276 2.3GHz (4x16-Core 16MB Cache)
2) Intel Xenon E5-4610, 2.4GHz (4x6-Core, 15MB Cache)
Both of them come at nearly the same price (keeping the other components nearly the same)
What we have noticed for "Gaussian" and "Molpro" calculations is that if we use same number of cores on both INTEL and AMD processors INTEL seem to do nearly twice as well than AMD in terms of calculation time.
For VASP we are not sure if the calculation does well with with more cores with less scaling (which is AMD) or less cores with better scaling (which is INTEL).
In the the two CPUs provided above AMD comes with (4x16=64 cores) and INTEL comes with (4x6=24 cores).
Since intel processors are scaling twice as much as AMD (for gaussian and molpro), we would like to know if they do the same with VASP.
Is it better we go with more cores on AMD or better scaling on INTEL.
We would like to run systems containing 250-350 heavy atoms like gold at both normal (LDA/GGA) and Hybrid (HSE/B3LYP) DFT.
Thank you in advance
<span class='smallblacktext'>[ Edited ]</span>
We are trying to buy a new machine for running vasp calculations and we couldn't decide on the type of configuration (especially on the number of cores) we should choose to buy.
We have the following options
1) AMD Opteron 6276 2.3GHz (4x16-Core 16MB Cache)
2) Intel Xenon E5-4610, 2.4GHz (4x6-Core, 15MB Cache)
Both of them come at nearly the same price (keeping the other components nearly the same)
What we have noticed for "Gaussian" and "Molpro" calculations is that if we use same number of cores on both INTEL and AMD processors INTEL seem to do nearly twice as well than AMD in terms of calculation time.
For VASP we are not sure if the calculation does well with with more cores with less scaling (which is AMD) or less cores with better scaling (which is INTEL).
In the the two CPUs provided above AMD comes with (4x16=64 cores) and INTEL comes with (4x6=24 cores).
Since intel processors are scaling twice as much as AMD (for gaussian and molpro), we would like to know if they do the same with VASP.
Is it better we go with more cores on AMD or better scaling on INTEL.
We would like to run systems containing 250-350 heavy atoms like gold at both normal (LDA/GGA) and Hybrid (HSE/B3LYP) DFT.
Thank you in advance
<span class='smallblacktext'>[ Edited ]</span>