Recent News about Amber and AmberTools

• Kepler K20 and K20X Benchmarks published + Updated Recommended Hardware (14 Jan 13)

From Ross Walker:

Updated benchmarks have been published for the AMBER GPU Code including Kepler K20 and K20X GPUs: K20 Benchmarks

Updates have been made to the recommended hardware for running GPU AMBER: Recommended GPU Hardware

• AMBER MD Workstations Announced - Recommended Hardware for GPU Acceleration (2 Nov 12)

From Ross Walker:

In response to feedback from a number of users we have worked with Exxact Corporation to design a series of machines termed AMBER MD Workstation and AMBER MD SimCluster that are optimized for both price and performance when running GPU accelerated AMBER Simulations. The following page provides details of recommended machine configurations and benchmarks. This includes opportunities for free test drives where you can try out your own simulations on these machines.

  Recommended GPU Hardware & Test Drives page for more information.
 

AMBER MD Workstations		AMBER MD SimCluster
Kepler K10 MD Workstation	Kepler GTX680 MD Workstation	Kepler MD SimCluster (K10 or GTX680)
• 2 Intel Xeon E5-2660 8-Core CPUs • 4 Tesla K10 8GB GPUs (8 x 4GB GPUS!) • 64 GB 1600MHz system memory • AMBER12 preinstalled • CentOS • 3 year warranty	• 2 Intel Xeon E5-2660 8-Core CPUs • 4 GeForce GTX 680 4GB GPUs • 64 GB 1600MHz system memory • AMBER12 preinstalled • CentOS • 3 year warranty	• 8U cluster (4 x 2U) • 2 Intel Xeon E5-2660 8-Core CPUs per node • 4 Tesla Kepler GPUs (K10 or GTX680) per node • 64 GB 1600MHz system memory per node • AMBER12 preinstalled23,558a • Optional InfiniBand QDR Switch & HCAs • Bright Computing Cluster Manager • full warranty (even with GTX680s)
($5000 - $18000) As shown ~$15,400	($3000 - $8000) As shown ~$7,600	Price varies

 

• Major Update Patch for AMBER 12 GPU Released (2 Sep 12)

  From Ross Walker:

  A major update to the GPU support (bugfix.9, GPU support revision 12.1, Aug 2012) has been released for AMBER 12. This patch includes a number of major improvements, bugfixes and refinements including:

      i) Support for Kepler I (K10), Kepler II (K20) and GTX6XX GPUs
      ii) New high performance, lower memory footprint hybrid fixed precision model: SPFP
      iii) Support for REMD using 1 GPU per replica.
      iv) Various bugfixes for aMD and IPS simulations.
      v) Tweaks to igb=5 GB model to bring it in line with original paper and NAB implementation.
      vi) Support for GPU accelerated NMR restraints.
      vii) Improved serial performance.
      viii) Reduced memory footprint.
      ix) More comprehensive test suite.
      x) Adds deterministic operation irrespective of GPU core count.

  This bugfix will be automatically applied when $AMBERHOME/configure is run. You should rebuild the GPU code in order to activate this update.
   

• NVIDIA Announces MD SimCluster - GPU Accelerated AMBER Preinstalled (14 Sep 11)

  From Ross Walker:

  NVIDIA have just announced the MD SimCluster Program. A collaboration between AMBER developer Ross Walker and NVIDIA engineers to design a GPU accelerated mini cluster optimized specifically for running GPU accelerated AMBER. These machines are offered by NVIDIA's partners and come with AMBER pre-configured and ready to run.
   

  GPU Computing Enabled Cluster Designed for GPU Accelerated AMBER 4 Nodes x 8 Tesla M2090 GPUs with CUDA 4.0. Dual 6-core CPUs + Infiniband QDR /node. Preconfigured With AMBER 11 (AMBER license required). No Setup necessary. Buy cluster loaded with Linux OS and cluster management software.

  Try the Latest AMBER GPU Code on the Tesla MD SimCluster for Free

  NVIDIA is collaborating with several system integrator partners to provide a trial program allowing people to gain remote access to the Tesla MD SimCluster, an 8 GPU computing system for free. This will allow you to see the exceptional performance offered by the latest version of GPU accelerated AMBER (GPU v2.2 [AMBER 11+ Bugfix 17]) and try it with your own simulations. All you need to do is load your models and start simulating.

  Details of the MD SimCluster project along with instructions for signing up for a free trial are available by following this link or by email to mdsimcluster@nvidia.com

  Additional information on recommended hardware for running AMBER on GPUs can also be found on the Main AMBER GPU webpage.

   

  Major performance update for AMBER GPU code released (18 Aug 11)

  Thanks to funding from the NSF SI2-SSE program (NSF1047875 & NSF1047919) and close collaboration with NVIDIA a major performance update is released August 18th 2011 for the GPU accelerated PMEMD package. This update, released as bugfix 17, doubles performance across the board for PME calculations while also fixing a number of outstanding bugs. A single GTX580 now runs the JAC Production NVE benchmark at over 50ns/day.

  

  Detailed benchmarks are available on the AMBER GPU Page. The update can be obtained from the Amber 11 bugfix page.
  

• Binary version of pmemd for Windows

  We are now distributing pre-built binaries for pmemd for Microsoft Windows.