New CPU GRO_A8 core for FAH with p16810
Posted: Tue Sep 01, 2020 4:42 pm
We have just released CPU FahCore_a8 0.0.6 to FAH!
This new CPU FahCore_a8 updates the FahCore from GROMACS 5 (FahCore_a7) to GROMACS 2020. This will provide substantial gains in efficiency of WU production! Specifically, we have been observing that FahCore_a8 runs noticeably faster than FahCore_a7; in tests, FahCore_a8 runs about 50% faster on recent CPUs that support AVX2 and FMA, and still 5-15% on older ones, with the bigger boosts seen in Windows. These are big optimizations that should make a significant difference for our research while better utilizing donor's existing hardware!
Testing this core will help new future CPU Projects such that researcher's science output will be increased significantly on existing hardware. It is a project based on the SARS-CoV-2 E protein which is an ion channel for which we will computationally study its electrophysiology.
Features:
- Improvements in MD algorithm with going to GROMACS 2020: update groups for neighbour lists, optimized SIMD, dual list dynamic pruning, use of OMP threading...
- Production of xtc rather than trrs to save space and potentially change to tng in the future.
- Boosts in TPF ranging from 7%-60% depending on project and system.
- Improvements in hardware accelerators: New FahCore_a8 has been updated to latest GROMACS code which now makes use of AVX2 instructions on modern CPUs.
- New way of generating gens in project.xml.
Bugfixes:
- Problems with pause/unpause.
- The viewer only showed a single chain, now it solves them all.
- For the moment, the core only supports ntmpi=1. We have found some problems with ntmpi > 1 which we hope to fix in the near future.
In addition to this, the hybrid CPU/GPU MD strategy of GROMACS could potentially enable a mixed CPU/GPU core in the future.
Please post here if you notice any issues!
This new CPU FahCore_a8 updates the FahCore from GROMACS 5 (FahCore_a7) to GROMACS 2020. This will provide substantial gains in efficiency of WU production! Specifically, we have been observing that FahCore_a8 runs noticeably faster than FahCore_a7; in tests, FahCore_a8 runs about 50% faster on recent CPUs that support AVX2 and FMA, and still 5-15% on older ones, with the bigger boosts seen in Windows. These are big optimizations that should make a significant difference for our research while better utilizing donor's existing hardware!
Testing this core will help new future CPU Projects such that researcher's science output will be increased significantly on existing hardware. It is a project based on the SARS-CoV-2 E protein which is an ion channel for which we will computationally study its electrophysiology.
Features:
- Improvements in MD algorithm with going to GROMACS 2020: update groups for neighbour lists, optimized SIMD, dual list dynamic pruning, use of OMP threading...
- Production of xtc rather than trrs to save space and potentially change to tng in the future.
- Boosts in TPF ranging from 7%-60% depending on project and system.
- Improvements in hardware accelerators: New FahCore_a8 has been updated to latest GROMACS code which now makes use of AVX2 instructions on modern CPUs.
- New way of generating gens in project.xml.
Bugfixes:
- Problems with pause/unpause.
- The viewer only showed a single chain, now it solves them all.
- For the moment, the core only supports ntmpi=1. We have found some problems with ntmpi > 1 which we hope to fix in the near future.
In addition to this, the hybrid CPU/GPU MD strategy of GROMACS could potentially enable a mixed CPU/GPU core in the future.
Please post here if you notice any issues!