AMBER 14 NVIDIA GPU
ACCELERATION SUPPORT

| Background | Authorship & Support | Features | Supported GPUs | System Size Limits |
| Accuracy Considerations | Installation and Testing | Running GPU Accelerated Simulations |
| Considerations for Maximizing GPU Performance | Benchmarks |
| Recommended Hardware & Test Drives |
| Return to Main Amber Page |


News and Updates

  • Sept 2014:New GTX980 and 970 Maxwell Cards - NVIDIA has just release the GTX980 and GTX970 cards based on the new Maxwell GM204 chip - this is brand new hardware design and it will take us approximately a month to properly test, port and optimize AMBER 14 on these cards. At present they are thus NOT supported. Please check this page regularly where we will post updates as soon as we can.
     

  • Sept 2014: New recommended hardware designs announced. High Performance - 8 GPU systems and High Density (2U) 4 GPU systems added to the list of Exxact AMBER Certified GPU Computing solutions.
     

  • Apr 2014: AMBER 14 released. >30% performance improvement, peer to peer parallel support and many new GPU features added.
     

  • Aug 2013: In collaboration with Exxact Corp we have put together an updated version of the AMBER Certified MD Workstation program. This provides a simple way to purchase machines for running GPU accelerated AMBER. The systems come with AMBER 12 or 14 pre-installed (AMBER license required) and fully tested and are ready to run out of the box. A range of AMBER optimized specifications are available for different budget levels and all are fully customizable and include a full warranty. Free test drives of these machines are also available. (More details are available on the Recommended Hardware Page.)


Background

This page provides background on running AMBER v14 (PMEMD) with NVIDIA GPU acceleration. If you are using AMBER v12 please see the archived AMBER version 12 GPU pages.

One of the new features of AMBER, introduced with version 11, was the ability to use NVIDIA GPUs to massively accelerate PMEMD for both explicit solvent PME and implicit solvent GB simulations. This was further extended in AMBER 12 and the performance envelope has been pushed even further with AMBER 14. This work is by Ross Walker at the San Diego Supercomputer Center and Scott Le Grand at Amazon Web Services in collaboration with NVIDIA. While this GPU acceleration is considered to be production ready, and has been heavily used since it's release as part of AMBER 11, it is still evolving and thus has not been tested to the same extent as the CPU code has over the years and so users should still exercise caution when using this code. The error checking is not as verbose in the GPU code as it is on the CPU. In particular simulation failures, such as atom collisions or other simulation instabilities will manifest themselves as CUDA launch errors or GPU download failures and not as informative error messages. If you encounter problems during a simulation on the GPU you should first try to run the identical simulation on the CPU to ensure that it is not your simulation setup which is causing problems.

If you encounter problems during a simulation on the GPU you should first try to run the identical simulation on the CPU to ensure that it is not your simulation setup which is causing problems. Feedback and questions should be posted to the Amber mailing list.

New in AMBER 14

AMBER 14 includes full GPU support in PMEMD and is a major update over AMBER 12. Focus for AMBER 14 has been on extending the performance envelope, supporting multi-GPU runs efficiently and further increasing the features available in the GPU accelerated code. Key new features in the AMBER 14 GPU version of PMEMD include:

  1. ~30% performance improvement for single GPU runs.
  2. Addition of peer to peer support for multi-GPU runs providing enhanced multi-GPU scaling.
  3. Hybrid bitwise reproducible fixed point precision model as standard (SPFP)
  4. Support for Extra Points in Multi-GPU runs.
  5. Jarzynski Sampling
  6. GBSA support
  7. Support for off-diagonal modifications to VDW parameters.
  8. Multi-dimensional Replica Exchange (Temperature and Hamiltonian)
  9. Support for CUDA 5.0, 5.5, 6.0 and 6.5
  10. Support for latest generation GPUs.
  11. Monte Carlo barostat support providing NPT performance equivalent to NVT.
  12. ScaledMD support.
  13. Improved accelerated (aMD) MD support.
  14. Explicit solvent constant pH support.
  15. NMR restraint support on multiple GPUs.
  16. Improved error messages and checking.
  17. Hydrogen mass repartitioning support (4fs time steps).

Support is provided for single GPU and multiple GPU runs. Employing multiple GPUs in a single simulation requires MPI and the pmemd.cuda.MPI executable. If you have multiple simulations to run then the recommended method is to use one GPU per job. The pmemd GPU code has been developed in such a way that for single GPU runs the PCI-E bus is only used for I/O. This sets AMBER apart from other MD packages since it means the CPU specs do not feature in the GPU code performance. As such low end economic CPUs can be used. Additionally it means that in a system containing 4 GPUs 4 individual calculations can be run at the same time without interfering with each other's performance.

Authorship & Support

NVIDIA CUDA Implementation:

Scott Le Grand (NVIDIA)
Ross C. Walker (SDSC)*

*Corresponding author.

Further information relating to the specifics of the implementation, methods used to achieve performance while controlling accuracy, and details of validation are available or will be shortly from the following publications:

  • Romelia Salomon-Ferrer; Andreas W. Goetz; Duncan Poole; Scott Le Grand;  & Ross C. Walker* "Routine microsecond molecular dynamics simulations with AMBER - Part II: Particle Mesh Ewald" , J. Chem. Theory Comput., 2013, 9 (9), pp 3878-3888 DOI: 10.1021/ct400314y
     
  • Andreas W. Goetz; Mark J. Williamson; Dong Xu; Duncan Poole; Scott Le Grand;  & Ross C. Walker* "Routine microsecond molecular dynamics simulations with AMBER - Part I: Generalized Born", J. Chem. Theory Comput., 2012, 8 (5), pp 1542-1555, DOI: 10.1021/ct200909j
     
  • Scott Le Grand; Andreas W. Goetz; & Ross C. Walker* "SPFP: Speed without compromise - a mixed precision model for GPU accelerated molecular dynamics simulations.", Comp. Phys. Comm, 2013, 184, pp374-380, DOI: 10.1016/j.cpc.2012.09.022

Funding for this work has been graciously provided by NVIDIA, The University of California (UC Lab 09-LR-06-117792), The National Science Foundation's (NSF) TeraGrid Advanced User Support Program through the San Diego Supercomputer Center and NSF SI2-SSE grants to Ross Walker (NSF1047875 / NSF1148276) and Adrian Roitberg (NSF1047919 / NSF1147910)

Citing the GPU Code

If you make use of any of this GPU support in your work please use the following citations (in addition to the standard AMBER 14 citation):

Explicit Solvent

  • Romelia Salomon-Ferrer; Andreas W. Goetz; Duncan Poole; Scott Le Grand;  & Ross C. Walker* "Routine microsecond molecular dynamics simulations with AMBER - Part II: Particle Mesh Ewald" , J. Chem. Theory Comput., 2013, 9 (9), pp 3878-3888. DOI: 10.1021/ct400314y

Implicit Solvent

  • Andreas W. Goetz; Mark J. Williamson; Dong Xu; Duncan Poole; Scott Le Grand;  & Ross C. Walker* "Routine microsecond molecular dynamics simulations with AMBER - Part I: Generalized Born", J. Chem. Theory Comput., 2012, 8 (5), pp 1542-1555 , DOI: 10.1021/ct200909j

^

Cuda Zone

 


Supported Features

The GPU accelerated version of PMEMD 14, supports both explicit solvent PME or IPS simulations in all three canonical ensembles (NVE, NVT and NPT) and implicit solvent Generalized Born simulations. It has been designed to support as many of the standard PMEMD v14 features as possible, however, there are some current limitations that are detailed below. Some of these may be addressed in the near future, and patches released, with the most up to date list posted on the web page. The following options are NOT supported (as of the Apr 2014 AMBER GPU v14.0 release):

1) ibelly /= 0 Simulations using belly style constraints are not supported.
2) icfe /= 0 Support for TI is not currently implemented. It is hoped that an update will be released in a few months that adds support for TI.
3) if (igb/=0 & cut<systemsize) GPU accelerated implicit solvent GB simulations do not support a cutoff.
4) nmropt > 1 Support is not currently available for nmropt > 1. In addition for nmropt = 1 only features that do not change the underlying force field parameters are supported. For example umbrella sampling restraints are supported as well as simulated annealing functions such as variation of Temp0 with simulation step. However, varying the VDW parameters with step is NOT supported.
5) nrespa /= 1 No multiple time stepping is supported.
6) vlimit /= -1 For performance reasons the vlimit function is not implemented on GPUs.
7) es_cutoff /= vdw_cutoff Independent cutoffs for electrostatics and van der Waals are not supported on GPUs.
8) order > 4 PME interpolation orders of greater than 4 are not supported at present.
9) imin=1 (in parallel) Minimization is only supported in the serial GPU code.
10) netfrc is ignored The GPU code does not currently support the netfrc correction in PME calculations and the value of netfrc in the ewald namelist is ignored.
11) emil_do_calc /= 0 Emil is not supported on GPUs.
12) lj264 /= 0 The 12-6-4 potential is not supported on GPUs.
13) isgld > 0 Self guided langevin is not supported on GPUs.
14) nmropt = 1 .and. For n=1,4; iat(n) < 0 The GPU code does not currently support the use of COM simulations.

Additionally there are some minor differences in the output format. For example the Ewald error estimate is NOT calculated when running on a GPU. It is recommended that you first run a short simulation using the CPU code to check the Ewald error estimate is reasonable and that your system is stable. With the exception of item 10 the above limitations are tested for in the code, however, it is possible that there are additional simulation features that have not been implemented or tested on GPUs.

^


Supported GPUs

GPU accelerated PMEMD has been implemented using CUDA and thus will only run on NVIDIA GPUs at present. Due to accuracy concerns with pure single precision the code uses a custom designed hybrid single / double / fixed precision model termed SPFP. This places the requirement that the GPU hardware supports both double precision and integer atomics meaning only GPUs with hardware revision 2.0 and later can be used. Support for hardware revision 1.3 was present in previous versions of the code but for code complexity and maintenance reasons has been deprecated in AMBER 14. At the time of Amber's release the following cards are supported (* = untested):

Hardware Version 3.0 / 3.5 (Kepler I / Kepler II)
Hardware Version 2.0 (Fermi)
  • Tesla M2090
  • Tesla C2050/C2070/C2075 (and M variants)
  • GTX560 / 570 / 580 / 590
  • GTX465 / 470 / 480
  • Quadro cards supporting SM2.0*

Caution:
The GTX-780Ti cards are NOT recommended at this time due to the high failure rate, believed to be due to insufficient CAS refresh voltage for the GPU DIMMS, of these models.

GTX-Titan and GTX-780 cards require NVIDIA Driver version >= 319.60.

GTX-Titan-Black Edition cards require NVIDIA Driver version >= 337.09 .or. 331.79 or later.

Other cards not listed here may also be supported as long as they correctly implement the Hardware Revision 2.0, 3.0 or 3.5 specifications. Due to the larger graphics memories and extended testing offered by the Tesla series GPUs these are the recommended models although GeForce cards will work fine.

Note that you should ensure that all GPUs on which you plan to run PMEMD are connected to PCI-E 2.0 x 16 lane slots or better, especially when running in parallel across multiple GPUs which really needs PCI-E 3.0 x16 with peer to peer support for reasonable scaling. If this is not the case then you will likely see degraded performance, although this effect is lessened in serial if you write to the mdout or mdcrd files infrequently (e.g. every 2000 steps or so). Scaling over multiple GPUs within a single node is possible, if all are in x16 or better slots. It is also possible to run over multiple nodes using infiniband but not recommended except for loosely coupled runs such as REMD. The main advantage of AMBER's approach to GPU implementation over other implementations such as NAMD and Gromacs is that it is possible to run multiple single GPU runs on a single node with little or no slow down. For example a node with 4 GTX-Titan-Black cards could run 4 individual AMBER DHFR 4fs NPT calculations all at the same time without slowdown providing an aggregate throughput in excess of 1000ns/day.

To simplify the process of selecting supported hardware we have teamed with Exxact Corp to provide AMBER Certified GPU computing solutions. For more information please refer to the Recommended Hardware section.

Optimum Hardware Designs for Running GPU Amber are provided on the Recommended Hardware page.

^


System Size Limits

In order to obtain the extensive speedups that we see with GPUs it is critical that all of the calculation take place on the GPU within the GPU memory. This avoids the performance hit that one takes copying to and from the GPU and also allows us to achieve extensive speedups for realistic size systems. This avoids the need to create systems with millions of atoms to show reasonable speedups even when sampling lengths are unrealistic. This unfortunately means that the entire calculation must fit within the GPU memory. Additionally we make use of a number of scratch arrays to achieve high performance. This means that the GPU memory usage can actually be higher than a typical CPU run. It also means, due to the way we had to initially implement parallel GPU support that the memory usage per GPU does NOT decrease as you increase the number of GPUs. This is something we hope to fix in the future but for the moment the atom count limitations imposed on systems by the GPU memory is roughly constant whether you run in serial or in parallel.

Since, unlike CPUs it is not possible to add more memory to a GPU (without replacing it entirely) and there is no concept of swap as there is on the CPU the size of the GPU memory imposes hard limits on the number of atoms supported in a simulation. Early on within the mdout file you will find information on the GPU being used and an estimate of the amount of GPU and CPU memory required:

|------------------- GPU DEVICE INFO --------------------
|
| CUDA Capable Devices Detected: 1
| CUDA Device ID in use: 0
| CUDA Device Name: Tesla C2070
| CUDA Device Global Mem Size: 6143 MB
| CUDA Device Num Multiprocessors: 14
| CUDA Device Core Freq: 1.15 GHz
|
|--------------------------------------------------------

...

| GPU memory information:
| KB of GPU memory in use: 4638979
| KB of CPU memory in use: 790531

The reported GPU memory usage is likely an underestimate and meant for guidance only to give you an idea of how close you are to the GPU's memory limit. Just because it is less than the available Device Global Mem Size does not necessarily mean that it will run. You should also be aware that the GPU's available memory is reduced by 1/9th if you have ECC turned on.

Memory usage is affected by the run parameters. In particular the size of the cutoff, larger cutoffs needing more memory, and the ensemble being used. Additionally the physical GPU hardware affects memory usage since the optimizations used are non-identical for different GPU types. Typically, for PME runs, memory usage runs:

NPT > NVT > NVE
NTT=3 > NTT=1 == NTT=2 > NTT=0
Barostat=1 > Barostat=2

Use of restraints etc will also increase the amount of memory in use. As will the density of your system. The higher the density the more pairs per atom there are and thus the more GPU memory will be required. The following table provides an approximate UPPER BOUND to the number of atoms supported as a function of GPU model. These numbers were estimated using boxes of TIP3P water (PME) and solvent caps of TIP3P water (GB). These had lower than optimum densities and so you may find you are actually limited for dense solvated proteins to around 20% less than the numbers here. Nevertheless these should provide reasonable estimates to work from.

All numbers are for SPFP precision and are approximate limits. The actual limits will depend on system density, simulation settings etc. These numbers are thus designed to serve as guidelines only.

Explicit Solvent (PME)

8 angstrom cutoff. Cubic box of TIP3P Water, NTT=3, NTB=1, NTP=1,NTF=2,NTC=2,DT=0.002.

GPU PME (max atoms)

type

memory

SPFP

GTX580 3.0 GB 1,240,000
M2090 (ecc off) 6.0 GB 2,680,000
GTX680 2.0 GB 920,000
K10/GTX680 (ecc off) 4.0 GB 1,810,000
K20X/GTX-Titan (ecc off) 6.0 GB 2,710,000
K40 (ecc off) 12.0 GB 5,021,000

Implicit Solvent (GB)

No cutoff, Sphere of TIP3P Water (for testing purposes only), NTT=3, NTB=0, NTF=2, NTC=2, DT=0.002, IGB=1

GPU GB (max atoms)

type

memory

SPFP

GTX580 3.0 GB 1,195,000
M2090 (ecc off) 6.0 GB 1,740,000
GTX680 2.0 GB 960,000
K10/GTX680 (ecc off) 4.0 GB 1,310,000
K20X/GTX-Titan (ecc off) 6.0 GB 1,750,000
K40 (ecc off) 12.0 GB 3,350,000

^


Accuracy Considerations

The nature of current generation GPUs is such that single precision arithmetic is considerably faster (>23x for K10, >2x for K20) than double precision arithmetic. This poses an issue when trying to obtain good performance from GPUs. Traditionally the CPU code in Amber has always used double precision throughout the calculation. While this full double precision approach has been implemented in the GPU code it gives very poor performance. Thus the AMBER GPU implementation was initially written to use a combination of single and double precision, termed hybrid precision (SPDP), that is discussed in further detail in the references provided at the beginning of this page. This approach used single precision for individual calculations within the simulation but double precision for all accumulations. It also used double precision for shake calculations and for other parts of the code where loss of precision was deemed to be unacceptable. Tests have shown that energy conservation is equivalent to the full double precision code and specific ensemble properties, such as order parameters, match the full double precision CPU code. The user should understand though that this approach leads to divergence between GPU and CPU simulations, similar to that observed when running the CPU code across different processor counts in parallel but occurring more rapidly. For this reason the GPU test cases are more sensitive to rounding differences caused by hardware and compiler variations and will likely require manual inspection of the test case diff files in order to verify that the installation is providing correct results. There can also be differences between CPU and GPU runs and between GPU and GPU on different GPU models for runs that rely on the random number stream. For example NTT=2 and NTT=3. This is because the random number stream is different between CPUs and GPUs.

With AMBER 14 the single and double precision models (SPSP, SPDP and DPDP) have been deprecated and replaced with a new hybrid model that combines single precision calculation with fixed precision accumulation. This is described in detail in the following manuscript and tests have shown that it provides accuracy that is as good or better than the original SPDP model. This new precision model is supported on v2.0 hardware and newer and is especially beneficial on Kepler I (3.0) hardware.

Scott Le Grand; Andreas W. Goetz; & Ross C. Walker* "SPFP: Speed without compromise - a mixed precision model for GPU accelerated molecular dynamics simulations.", Comp. Phys. Comm, 2013, 184, pp374-380, DOI: 10.1016/j.cpc.2012.09.022

While the default precision model is currently the new SPFP model there is also a DPFP model available which provides full double precision calculation with bitwise reproducible fixed precision accumulation to facilitate advanced testing and comparison. The precision models supported in AMBER 14, and determined at compile time as described later, are:

  • SPFP - (Default) Uses a combination of single precision for calculation and fixed (integer) precision for accumulation. This approach provides optimum speedup on Kepler cards, minimizes memory overhead and provides greater net precision than the original SPDP model. It is designed for optimum performance on hardware revision 2.0 or later.
     
  • DPFP - Uses double precision (and double precision equivalent fixed precision) for the entire calculation. This provides for careful regression testing against the CPU code. It makes no additional approximations above and beyond the CPU implementation and would be the model of choice if performance was not a consideration. On v2.0 NVIDIA hardware (e.g. M2090) the performance is approximately half that of the SPFP model while on v3.0 NVIDIA hardware (e.g. K10) the performance is substantially less than the SPFP model.

Recommendation for Minimization

One limitation of the SPFP precision model is that force can be truncated if they overflow the fixed precision representation. This should never be a problem during MD simulations for any well behaved system. However, for minimization or very early in the heating phase it can present a problem. This is especially true if two atoms are close to each other and thus have large VDW repulsions. It is recommended therefore that for minimization you use the CPU version of the code. Only in situations where you are confident the structure is reasonable, for example if it was a snapshot from dynamics, should you use the GPU code (SPFP) for minimization.

^


Installation and Testing

The single GPU version of PMEMD is called pmemd.cuda while the multi-GPU version is called pmemd.cuda.MPI. These are built separately from the standard serial and parallel installations. Before attempting to build the GPU versions of PMEMD you should have built and tested at least the serial version of Amber and preferably the parallel version as well. This will help ensure that basic issues relating to standard compilation on your hardware and operating system do not lead to confusion with GPU related compilation and testing problems. You should also be reasonably familiar with Amberís compilation and test procedures. The minimum requirements for building the GPU version of PMEMD are, as of the AMBER 14 release:

  • NVIDIA Toolkit v5.0, 5.5, 6.0 or 6.5. (v5.0 recommended for performance reasons)
     
  • NVIDIA GPU supporting Hardware Revision 2.0, 3.0, 3.5 and later.
     
  • NVIDIA CUDA Driver v319.60 or later (v337.09 or later required for GTX-780Ti and GTX-Titan-Black)
     
  • MPI library for parallel GPU. (MVAPICH2 v1.8 or later / MPICH2 v1.4p1 or later recommended)

It is assumed that you have already correctly installed and tested CUDA support on your GPU. Before attempting to build pmemd.cuda you should make sure you have correctly installed both the NVIDIA Toolkit (nvcc compiler) and a CUDA supporting NVIDIA driver. You may want to try downloading the NVIDIA CUDA SDK (available from http://www.nvidia.com/ or included as 'samples' in the toolkit v5.0 or later) and see if you can build that. Additionally the environment variable CUDA_HOME should be set to point to your NVIDIA Toolkit installation and $CUDA_HOME/bin/ should be in your path. At the time of release CUDA 5.0 or later is required with 5.0, 5.5, 6.0 and 6.5 having been tested and officially supported. For performance reasons, at the time of writing, CUDA 5.0 is currently the recommended version to use.

Building and Testing the Default SPFP Precision Model

Single GPU (pmemd.cuda)

Assuming you have a working CUDA installation you can build the single GPU version, pmemd.cuda, using the default SPFP precision model as follows:

cd $AMBERHOME
make clean
./configure -cuda gnu    (or intel)
make install

Next you can run the tests using the default GPU (the one with the largest memory) with:

make test.cuda

The majority of these tests should pass. However, given the parallel nature of GPUs, meaning the order of operation is not well defined, the limited precision of the SPFP precision model, and variations in the random number generator on different GPU hardware, it is not uncommon for there to be several possible failures. You may also see some tests, particularly the cellulose test, fail on GPUs with limited memory. You should inspect the diff file created in the $AMBERHOME/logs/test_amber_cuda/ directory to manually verify any possible failures. Differences which occur on only a few lines and are minor in nature can be safely ignored. Any large differences, or if you are unsure, should be posted to the Amber mailing list for comment.

Multiple GPU (pmemd.cuda.MPI)

Once you have built and tested the serial GPU version you can optionally build the parallel version (if you have multiple GPUs of the same model). Unlike the CPU code it is not necessary to build the parallel version of the GPU code in order to access specific simulation options (except REMD). Thus you only need to build the parallel GPU code if you plan to run a single calculation across multiple GPUs. AMBER 12 had poor multi-GPU scaling for GPUs later than the Fermi (C2050/2070/2090) generation. This scaling has been improved significantly in AMBER 14 via the use of peer to peer communication over the PCI-E bus. While the underlying communication between GPUs is peer to peer if supported the code still requires MPI in order to run.

No special software beyond CUDA 5.0 and an MPI library is required to support peer to peer. For details on determining which GPUs can operate in peer to peer mode please see the multi-GPU section of the running on GPUs section below. The instructions here assume that you can already successfully build the MPI version of the CPU code. If you cannot, then you should focus on solving this before you move on to attempting to build the parallel GPU version.

The parallel GPU version of the code works using MPI v1 or later.

You can build the multi-GPU code as follows:

cd $AMBERHOME
make clean
./configure -cuda -mpi gnu    (or intel)
make install

Next you can run the tests using GPUs enumerated sequentially within a node (if you have multiple nodes or more complex GPU setups within a node then you should refer to the discussion below on running on multiple GPUs):

export DO_PARALLEL='mpirun -np 2' # for bash/sh
setenv DO_PARALLEL 'mpirun -np 2' # for csh/tcsh
./test_amber_cuda_parallel.sh

Building non-standard Precision Models

You can build different precision models as described below. However, be aware that this is meant largely as a debugging and testing issue and NOT for running production calculations. Please post any questions or comments you may have regarding this to the Amber mailing list. You select which precision model to compile as follows:

cd $AMBERHOME
make clean
./configure -cuda_DPFP gnu
make install

This will produce executables named pmemd.cuda_XXXX where XXXX is the precision model selected at configure time (SPFP or DPFP). You can then test this on the GPU with the most memory as follows:

cd $AMBERHOME/test/
./test_amber_cuda.sh DPFP (to test the DPFP precision model)

Testing Alternative GPUs

Should you wish to run the serial GPU tests on a GPU different from the one with the most memory (and lowest GPU ID if more than one identical GPU exists) then you can provide this by setting the CUDA_VISIBLE_DEVICES environment variable. For example, to test the GPU with ID = 2 and the default SPFP precision model you would specify:

cd $AMBERHOME
export CUDA_VISIBLE_DEVICES=2
make test.cuda

^


Running GPU Accelerated Simulations

Single GPU

In order to run a single GPU accelerated MD simulation the only change required is to use the executable pmemd.cuda in place of pmemd. E.g.

$AMBERHOME/bin/pmemd.cuda -O -i mdin -o mdout -p prmtop \
-c inpcrd -r restrt -x mdcrd

This will automatically run the calculation on the GPU with the most memory even if that GPU is already in use (see below for system settings to have the code auto select unused GPUs). If you have only a single CUDA capable GPU in your machine then this is fine, however if you want to control which GPU is used, for example you have a Tesla C2050 (3GB) and a Tesla C2070 (6GB) in the same machine and want to use the C2050 which has less memory, or you want to run multiple independent simulations using different GPUs then you manually need to specify the GPU ID to use using the CUDA_VISIBLE_DEVICES environment variable. The environment variable CUDA_VISIBLE_DEVICES lists which devices are visible as a comma-separated string. For example, if your desktop has two tesla cards and a Quadro:

$ ./deviceQuery -noprompt | egrep "^Device"
Device 0: "Tesla C2050"
Device 1: "Tesla C2070"
Device 2: "Quadro FX 3800"

By setting CUDA_VISIBLE_DEVICES you can make only a subset of them visible to the runtime:

$ export CUDA_VISIBLE_DEVICES="0,2"
$ ./deviceQuery -noprompt | egrep "^Device"
Device 0: "Tesla C2050"
Device 1: "Quadro FX 3800"

Hence if you wanted to run two pmemd.cuda runs, with the first running on the C2050 and the second on the C2070 you would run as follows:

$ export CUDA_VISIBLE_DEVICES="0"
nohup $AMBERHOME/bin/pmemd.cuda -O -i mdin -o mdout -p prmtop \
-c inpcrd -r restrt -x mdcrd </dev/null &

$ export CUDA_VISIBLE_DEVICES="1"
nohup $AMBERHOME/bin/pmemd.cuda -O -i mdin -o mdout -p prmtop \
-c inpcrd -r restrt -x mdcrd </dev/null &

In this way you only ever expose a single GPU to the pmemd.cuda executable and so avoid issues with the running of multiple runs on the same GPU. This approach is the basis of how you can control GPU usage in parallel runs.

If you want to know which GPU a calculation is running on the value of CUDA_VISIBLE_DEVICES and other GPU specific information is provided in the mdout file.

To check whether a GPU is in use or not you can use the nvidia-smi command. For Tesla series GPUs (K20, K40 etc) this provides % load information and process information for each GPU. For GeForce series cards this information is not available via nvidia-smi so it is better to just check the GPU memory usage and temperature. You can do this with:

+------------------------------------------------------+                       
| NVIDIA-SMI 337.12     Driver Version: 337.12         |                       
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX TIT...  Off  | 0000:02:00.0     N/A |                  N/A |
| 26%   32C  N/A     N/A /  N/A |     15MiB /  6143MiB |     N/A      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX TIT...  Off  | 0000:03:00.0     N/A |                  N/A |
|  0%   35C  N/A     N/A /  N/A |     15MiB /  6143MiB |     N/A      Default |
+-------------------------------+----------------------+----------------------+

In this case both GPUs are idle (temp is < 50C and Memory usage is < 50MiB). If only GPU 0 was in use it might look something like this:

+------------------------------------------------------+
| NVIDIA-SMI 337.12     Driver Version: 337.12         |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX TIT...  Off  | 0000:02:00.0     N/A |                  N/A |
| 67%   80C  N/A     N/A /  N/A |    307MiB /  6143MiB |     N/A      Default |
+-------------------------------+----------------------+----------------------+
|   1  GeForce GTX TIT...  Off  | 0000:03:00.0     N/A |                  N/A |
|  0%   35C  N/A     N/A /  N/A |     15MiB /  6143MiB |     N/A      Default |
+-------------------------------+----------------------+----------------------+

Note running X11 can confuse things a little bit, showing slightly higher memory usage on one of the GPUs, but it should still be possible to use this approach to determine which GPU is in use.

An alternative approach, and the one recommended if you have multiple GPUs in a single node, and only want to run single GPU jobs, is to set them to Persistence and Compute Exclusive Modes. In this mode a GPU will reject more than one job. pmemd.cuda is capable of detecting this and will automatically run on a free GPU. For example suppose you have 4 GPUs in your node and you run 5 calculations:

$AMBERHOME/bin/pmemd.cuda -O -i mdin -o mdout.1 -x mdcrd.1 -inf mdinfo.1 &
$AMBERHOME/bin/pmemd.cuda -O -i mdin -o mdout.2 -x mdcrd.2 -inf mdinfo.2 &
$AMBERHOME/bin/pmemd.cuda -O -i mdin -o mdout.3 -x mdcrd.3 -inf mdinfo.3 &
$AMBERHOME/bin/pmemd.cuda -O -i mdin -o mdout.4 -x mdcrd.4 -inf mdinfo.4 &
$AMBERHOME/bin/pmemd.cuda -O -i mdin -o mdout.5 -x mdcrd.5 -inf mdinfo.5 &

In this situation calculation 1 will run on GPU 0, 2 on GPU 1, 3 on GPU 2 and 4 on GPU 3. The 5th job will quit immediately with an error stating that no free GPUs are available.

cudaMemcpyToSymbol: SetSim copy to cSim failed all CUDA-capable devices are busy or unavailable

This approach is useful since it means you do not have to worry about setting CUDA_VISIBLE_DEVICES and you do not have to worry about accidentally running multiple jobs on the same GPU. The code will automatically look for GPUs that are not in use and will use them automatically and will quit if sufficient free GPUs are not available. You set Persistence and Compute Exclusive Modes by running the following as root:

$ nvidia-smi -pm 1
$ nvidia-smi -c 3

The disadvantage of this is that you need to have root permissions to set it and the setting is also lost on a reboot. We recommend that you add these settings to your system's startup scripts. In Redhat or Centos this can be accomplished with:

echo "nvidia-smi -pm 1" >> /etc/rc.d/rc.local
echo "nvidia-smi -c 3" >> /etc/rc.d/rc.local

Note this approach also works on clusters where queuing systems understand GPUs as resources and thus can keep track of total gpus allocated but do not control which GPU you see on a node.

Multi GPU

The way in which a single calculation runs across multiple GPUs has been changed in AMBER 14. When originally designed the PCI-E bus speed was gen 2 x16 and the GPUs were C1060 or C2050s. Since then we have seen GPUs advance to GTX-Titan-Black and K40s which are ~ 8x faster than the C1060s (more if we include that AMBER 14 is >30% faster than AMBER 12) while the PCI-E bus speed has only increased by 2x (PCI Gen2 x16 to PCI Gen3 x16) and Infiniband interconnects by about the same amount. This unfortunate situation is all too common in parallel machines and yet many machine designers do not seem to appreciate the problems this places on software designers. The situation is now such that for PME calculations using traditional MPI it is not possible for AMBER to scale over multiple GPUs. In AMBER 14 we have chosen to address this problem by focusing on the use of Peer to Peer communication within a node. This mode of operation allows GPUs to communicate directly through the PCI-E bus without going through the CPU chipset which adds too much latency. Assuming all your GPUs have PCI-E gen 3 x16 bandwidth and support non-blocking peer to peer communication it is possible to scale to 4, or perhaps even 8 GPUs within a single node. Determining which GPUs can communicate by peer to peer is discussed below.

It should be noted that while the legacy MPI and GPU-Direct methods of multi-GPU communication are still supported, and will be used by the code automatically if peer to peer communication is not available, you are very unlikely to see any speedup by using multiple GPUs for a single job if the GPUs are newer than C2050s. Multi-node runs are almost impossible to get to scale. We suggest saving money by skipping the expensive infiniband interconnect and instead going for GPU dense nodes. Note the 4 x GPU Exxact systems are designed to support 2 x 2 GPU runs at once using peer to peer. We will shortly be releasing a system design that supports 4 way peer to peer and will post details on the recommended hardware page once available.

When running in parallel across multiple GPUs the selection of which GPU to run on becomes a little more complicated. We have attempted to simplify things for you by writing a short CUDA program that checks which GPUs in your node can support peer to peer communication. You should download and build this as follows:

  1. Download check_p2p.tar.bz2 and save it to your home directory.
  2. Untar with tar xvjf check_p2p.tar.bz2.
  3. cd check_p2p
  4. make    (assuming you have the cuda toolkit correctly installed this should build you the executable gpuP2PCheck.

On most of the Exxact Amber certified machines this will be pairs of GPUs in the order 0+1 and 2+3. In other words on a 4 GPU machine you can run a total of two by two GPU jobs, one on GPUs 0 and 1 and one on GPUs 2 and 3. Running a calculation across more than 2 GPUs will result in peer to peer being switched off which will likely mean the calculation will run slower than if it had been run on a single GPU. To see which GPUs in your system can communicate via peer to peer you can run the 'gpuP2PCheck' program you built above. This reports which GPUs can talk to each other. For example from an Exxact Quantum TXR411-064R 4xGPU system:

CUDA_VISIBLE_DEVICES is unset.
CUDA-capable device count: 4
   GPU0 "GeForce GTX TITAN"
   GPU1 "GeForce GTX TITAN"
   GPU2 "GeForce GTX TITAN"
   GPU3 "GeForce GTX TITAN"

Two way peer access between:
   GPU0 and GPU1: YES
   GPU0 and GPU2: NO
   GPU0 and GPU3: NO
   GPU1 and GPU2: NO
   GPU1 and GPU3: NO
   GPU2 and GPU3: YES

So in this case GPUs 0 and 1 can talk to each other and GPUs 2 and 3 can talk to each other. Thus to run a 2 GPU job on GPUs 0 and 1 you would run with:

export CUDA_VISIBLE_DEVICES=0,1
mpirun -np 2 $AMBERHOME/bin/pmemd.cuda.MPI -O -i mdin -o mdout -p ...

If peer to peer communication is working you will see the following reported in your mdout file.

|---------------- GPU PEER TO PEER INFO -----------------
|
|   Peer to Peer support: ENABLED
|
|--------------------------------------------------------

Since peer to peer communication does not involve the CPU chipset so it is possible, unlike in previous versions of AMBER, to run multiple multi-GPU runs on a single node. In our 4 GPU example above we saw that GPUs 0 and 1 can communicate via peer to peer and GPUs 2 and 3 can communicate. Thus we can run the following options without the jobs interfering with each other performance wise:

Option 1: All single GPU
export CUDA_VISIBLE_DEVICES=0
cd run1
nohup $AMBERHOME/bin/pmemd.cuda -O -i ... &

export CUDA_VISIBLE_DEVICES=1
cd run2
nohup $AMBERHOME/bin/pmemd.cuda -O -i ... &

export CUDA_VISIBLE_DEVICES=2
cd run3
nohup $AMBERHOME/bin/pmemd.cuda -O -i ... &

export CUDA_VISIBLE_DEVICES=3
cd run4
nohup $AMBERHOME/bin/pmemd.cuda -O -i ... &

Option 2: One dual GPU and 2 single GPU runs
export CUDA_VISIBLE_DEVICES=0,1
cd run1
nohup mpirun -np 2 $AMBERHOME/bin/pmemd.cuda.MPI -O -i ... &

export CUDA_VISIBLE_DEVICES=2
cd run2
nohup $AMBERHOME/bin/pmemd.cuda -O -i ... &

export CUDA_VISIBLE_DEVICES=3
cd run3
nohup $AMBERHOME/bin/pmemd.cuda -O -i ... &

Option 3: Two dual GPU runs
export CUDA_VISIBLE_DEVICES=0,1
cd run1
nohup mpirun -np 2 $AMBERHOME/bin/pmemd.cuda.MPI -O -i ... &

export CUDA_VISIBLE_DEVICES=2,3
cd run2
nohup mpirun -np 2 $AMBERHOME/bin/pmemd.cuda.MPI -O -i ... &


Considerations for Maximizing GPU Performance

There are a number of considerations above and beyond those typically used on a CPU for maximizing the performance achievable for a GPU accelerated PMEMD simulation. The following provides some tips for ensuring good performance.

  1. Avoid using small values of NTPR, NTWX, NTWV, NTWE and NTWR. Writing to the output, restart and trajectory files too often can hurt performance even on CPU runs, however, this is more acute for GPU accelerated simulations because there is a substantial cost in copying data to and from the GPU. Performance is maximized when CPU to GPU memory synchronizations are minimized. This is achieved by computing as much as possible on the GPU and only copying back to CPU memory when absolutely necessary. There is an additional overhead in that performance is boosted by only calculating the energies when absolutely necessary, hence setting NTPR or NTWE to low values will result in excessive energy calculations. You should not set any of these values to less than 100 (except 0 to disable them) and ideally use values of 500 or more. >100000 for NTWR is ideal.
     
  2. Avoid setting ntave /= 0. Turning on the printing of running averages results in the code means needing to calculate both energy and forces on every step. This can lead to a performance loss of 8% or more when running on the GPU. This can also affect performance on CPU runs although the difference is not as marked. Similar arguments apply to setting the value of ene_avg_sampling to small values.
     
  3. Avoid using the NPT ensemble (NTB=2) when it is not required; if needed make use of the Monte Carlo barostat (barostat=2). Performance will generally be NVE>NVT>NPT. (NVT ~ NPT for barostat=2).
     
  4. Avoid the use of GBSA in implicit solvent GB simulations unless required. The GBSA term is calculated on the CPU and thus requires a synchronization between GPU and CPU memory on every MD step as opposed to every NTPR or NTWX steps when running without this option.
     
  5. Use the Berendsen Thermostat (ntt=1) or Anderson Thermostat (ntt=2) instead of the Langevin Thermostat (ntt=3). Langevin simulations require very large numbers of random numbers which slows performance slightly.
     
  6. Do not assume that for small systems the GPU will always be faster. Typically for GB simulations of less than 150 atoms and PME simulations of less than 9,000 atoms it is not uncommon for the CPU version of the code to outperform the GPU version on a single node. Typically the performance differential between GPU and CPU runs will increase as atom count increases. Additionally the larger the non-bond cutoff used the better the GPU to CPU performance gain will be.
     
  7. When running in parallel across multiple GPUs you should NOT attempt to share nodes and thus interconnects. For example you should avoid running 2 separate MPI jobs on individual nodes. For example if you have 2 nodes, each with a QDR IB card in, 1 C2050 and 1 C1060 you will likely get very poor performance if you attempt to run a dual GPU job on the 2 C2050's and a second dual GPU job on the 2 C1060's. It is also not advisable to mix GPU models when running in parallel. In this situation you are advised to physically place both C2050's in one node and both C1060's in the other. You could of course run a dual C2050 job across the two nodes and then 2 single GPU jobs on each of the C1060's. It should be noted that for more modern GPUs such as K40 and GTX-780 it is recommended to only run multi-GPU calculations in the peer to peer intranode mode. Infiniband is simply not fast enough to cope with modern GPUs and so it is more efficient to save money on the interconnect and have multi-GPU nodes.
     
  8. Turn off ECC (C2050 and later). ECC can cost you up to 10% in performance and hurts parallel scaling. You should verify that your GPUs are working correctly, and not giving ECC errors for example before attempting this. You can turn this off on Fermi based cards and later by running the following command for each GPU ID as root, followed by a reboot:

    nvidia-smi -g 0 --ecc-config=0   (repeat with -g x for each GPU ID)

    Extensive testing of AMBER on a wide range of hardware has established that ECC has little to no benefit on the reliability of AMBER simulations. This is part of the reason it is acceptable (see recommended hardware) to use the GeForce gaming cards for AMBER simulations.
     
  9. Turn on boost clocks if supported. Newer GPUs from NVIDIA, such as the K40, support boost clocks which allow the clock speed to be increased if there is power and temperature headroom. This must be turned on to obtain optimum performance with AMBER. If you have a K40 or newer GPU supporting boost clocks then run the following:

    sudo nvidia-smi -i 0 -ac 3004,875       which puts device 0 into the highest boost state.

    To return to normal do: sudo nvidia-smi -rac

    To enable this setting without being root do: nvidia-smi -acp 0

 

^


Recommended Hardware

For examples of hardware that has been optimized for running GPU AMBER simulations, in terms of both price and performance please see the following page.

^


Benchmarks

Benchmarks are available on the following page.

^