Biowulf Logo

Status
About
Hardware
Applications
Batch queues
Disk storage

MPI
Performance
New Users
User Guide
Documentation
Research
Photos

Biowulf Benchmarks

WU-Blast benchmarks (Jun 2007)

Charmm Benchmarks (Apr 2007)

Blast Benchmarks (Dec 2006)

NAMD benchmarks (August 2006)

Amber 9 benchmarks (July 2006)

GROMACS benchmarks (July 2006)

CNS benchmarks (Aug 2005)

Sequence Analysis benchmark suite (2006)

fastDNAml Benchmarks (June 2002)

CHARMM (c31b2) Benchmarks (Apr 2007)

Model: Myoglobin with CO in a cubic box of 4985 water molecules
Test : dynamics using pme (Particle Mesh Ewald) method, nstep = 2000
CHARMM Benchmark Plot
# proc. Time in minutes
p2800,gige
2.8 GHz Xeon,
1Gb/s ethernet 		o2000,gige
(2.0 GHz Opteron,
1Gb/s ethernet) 		o2200,gige
(2.2 GHz Opteron,
1Gb/s ethernet) 		o2800,gige
(2.8 GHz Opteron,
1Gb/s ethernet) 		o2600,gige
(2.6 GHz Opteron,
dual-core(4p),
1Gb/s ethernet)
1 38.2 35.8 32.4 26.0 27.4
2 21.8 35.8 32.4 13.7 14.4
4 26.0 15.9 15.5 13.4 21.1

CHARMM (c28b3) Benchmarks (7/22/02)

Model: Myoglobin with CO in a cubic box of 4985 water molecules
Test : dynamics using pme (Particle Mesh Ewald) method, nstep = 1000
  • AMD 1.4 GHz vs. Intel 866 MHz
  • AMD 1.4 GHz, Myrinet vs. Ethernet
  • CHARMM scalability

    • CHARMM (c28b3) Benchmarks (12/5/01)

    Model: Myoglobin with CO in a cubic box of 4985 water molecules
    Test : dynamics using pme (Particle Mesh Ewald) method, nstep = 1000
  • Myrinet vs. Ethernet
  • 2.2-tcpfix vs. 2.4 kernels
  • Pentium vs. XP
  • Athlon vs XP
  • 2.4.12 vs 2.4.18 kernels 
    • Test: QM using GAMESS, nstep = 1
  • Pentium vs XP

    • CHARMM MPICH/GM Benchmarks (3/27/01)

    These benchmarks are described in /usr/local/src/mpich-1.2..5/examples/perftest
  • 2 x 866 MHz Pentium III, 256k full speed secondary cache, 512MB memory
  • Myricom Myrinet-2000 PCI64B
  • RedHat Linux 6.x/2.2.16 kernel
  • MPICH-GM 1.2..5
  • GM-1.4pre56
    • Average latency
    (uS)    		 Average transfer rate
    (Mbytes/s)
    p2p-short blocking point-to-point
    message sizes 0 - 1k bytes    		 9.77 37.05
    p2p-long blocking point-to-point
    message sizes 16k - 64k bytes    		 118.33 193.95
    nbpt2pt-long non-blocking point-to-point
    message sizes 16k - 64k bytes    		 119.06 193.44
    bisect8-short bisectional bandwidth
    message sizes 0 - 1k bytes
    8 procs on 8 nodes    		 9.77 148.47
    bisect8-long bisectional bandwidth
    message sizes 16k - 64k bytes
    8 procs on 8 nodes    		 118.32 193.96
    bw Blocking bandwidth 0 - 4MBytes 9.96 218.12

    CHARMM MPICH/GM Benchmarks (3/27/00)

    These benchmarks are described in /usr/local/src/mpich-1.1.2..13/examples/perftest.

  • 2 x 450 MHz Pentium III, 512k secondary caches, 256MB memory
  • Myricom Myrinet PCI32C (1MB)
  • RedHat Linux 6.x/2.2.13 kernel
  • MPICH-GM 1.1.2..13
  • GM-1.1.3
    • latency
    (uS)    		 transfer rate
    (Mbytes/s)    		 variance
    in fit
    pt2ptshort blocking point-to-point
    message sizes 0 - 1k bytes    		 24.55 25.29 0.000034
    pt2ptlong blocking point-to-point
    message sizes 16k - 64k bytes    		 119.77 50.16 0.001753
    nbpt2ptlong non-blocking point-to-point
    message sizes 16k - 64k bytes    		 121.85 50.03 0.001655
    bisectshort8 bisectional bandwidth
    message sizes 0 - 1k bytes
    8 procs on 8 nodes    		 24.53 25.30 0.000034
    bisectlong8 bisectional bandwidth
    message sizes 16k - 32k bytes
    8 procs on 8 nodes    		 125.30 50.63 0.001646

    CHARMM MPICH/GM Benchmarks (10/20/99)

    These benchmarks are described in /usr/local/src/mpich-1.1.2..7/examples/perftest.

  • 2 x 450 MHz Pentium II, 512k secondary caches, 256MB memory
  • Myricom Myrinet PCI32C (1MB)
  • RedHat Linux 6.0/2.2.12 kernel
  • MPICH-GM 1.1.2..7
  • GM-1.086 for Linux-2.2.12smp
    • latency
    (uS)    		 transfer rate
    (Mbytes/s)    		 variance
    in fit
    pt2ptshort blocking point-to-point
    message sizes 0 - 1k bytes    		 32.05 24.47 0.000043
    pt2ptlong blocking point-to-point
    message sizes 16k - 64k bytes    		 143.91 50.73 0.001593
    nbpt2ptlong non-blocking point-to-point
    message sizes 16k - 64k bytes    		 147.85 51.02 0.001584
    bisectshort16 bisectional bandwidth
    message sizes 0 - 1k bytes
    16 procs on 16 nodes    		 32.06 24.37 0.000043
    bisectlong16 bisectional bandwidth
    message sizes 16k - 32k bytes
    16 procs on 16 nodes    		 132.76 49.83 0.001622
    bisectshort32 bisectional bandwidth
    message sizes 0 - 1k bytes
    32 procs on 16 nodes    		 32.06 24.37 0.000043
    bisectlong32 bisectional bandwidth
    message sizes 16k - 32k bytes
    32 procs on 16 nodes    		 142.69 50.73 0.001593

    Performance testing a la Myricom: runex generates basic latency and bandwidth performance data.

    bandwidth latency
    blocking 50.85 Mbytes/s 32.11 usec
    non-blocking 50.47 Mbytes/s 33.34 usec

    Benchmarks published by Myricom for 64-bit and 32-bit interfaces.

    This document is available as http://biowulf.nih.gov/perf.html
    Biowulf home page | Helix Systems | NIH