Benchmark Turing Pi 2 cluster with 4x RK1 Nodes
Closed this issue · 5 comments
geerlingguy commented
I am going to see how well 4x RK1 nodes (each running an 8-core RK3588 SoC) will run on the Turing Pi 2 cluster.
geerlingguy commented
I was having some trouble getting the test to run (I am using an ssh_user of ubuntu), then realized avahi-daemon isn't running on this install, so mDNS / .local discovery doesn't work (I was testing with DNS names like turing1.local, turing2.local, etc.
So I installed it on all four nodes: sudo apt-get install avahi-daemon
And now DNS resolution for .local hostnames on each of the cluster nodes is working. I'll add a note in the README.
geerlingguy commented
Benchmark is running now. First test with 4/8 Ps and Qs: 228.94 Gflops
================================================================================
HPLinpack 2.3 -- High-Performance Linpack benchmark -- December 2, 2018
Written by A. Petitet and R. Clint Whaley, Innovative Computing Laboratory, UTK
Modified by Piotr Luszczek, Innovative Computing Laboratory, UTK
Modified by Julien Langou, University of Colorado Denver
================================================================================
An explanation of the input/output parameters follows:
T/V : Wall time / encoded variant.
N : The order of the coefficient matrix A.
NB : The partitioning blocking factor.
P : The number of process rows.
Q : The number of process columns.
Time : Time in seconds to solve the linear system.
Gflops : Rate of execution for solving the linear system.
The following parameter values will be used:
N : 100009
NB : 256
PMAP : Row-major process mapping
P : 4
Q : 8
PFACT : Right
NBMIN : 4
NDIV : 2
RFACT : Crout
BCAST : 1ringM
DEPTH : 1
SWAP : Mix (threshold = 64)
L1 : transposed form
U : transposed form
EQUIL : yes
ALIGN : 8 double precision words
--------------------------------------------------------------------------------
- The matrix A is randomly generated for each test.
- The following scaled residual check will be computed:
||Ax-b||_oo / ( eps * ( || x ||_oo * || A ||_oo + || b ||_oo ) * N )
- The relative machine precision (eps) is taken to be 1.110223e-16
- Computational tests pass if scaled residuals are less than 16.0
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR11C2R4 100009 256 4 8 2912.76 2.2894e+02
HPL_pdgesv() start time Fri Feb 23 16:15:13 2024
HPL_pdgesv() end time Fri Feb 23 17:03:46 2024
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 2.41638360e-03 ...... PASSED
================================================================================
Finished 1 tests with the following results:
1 tests completed and passed residual checks,
0 tests completed and failed residual checks,
0 tests skipped because of illegal input values.
--------------------------------------------------------------------------------
End of Tests.
================================================================================
geerlingguy commented
Testing with one node, while measuring power consumption: 59.810 Gflops, 18.1W, for 3.30 Gflops/W
================================================================================
HPLinpack 2.3 -- High-Performance Linpack benchmark -- December 2, 2018
Written by A. Petitet and R. Clint Whaley, Innovative Computing Laboratory, UTK
Modified by Piotr Luszczek, Innovative Computing Laboratory, UTK
Modified by Julien Langou, University of Colorado Denver
================================================================================
An explanation of the input/output parameters follows:
T/V : Wall time / encoded variant.
N : The order of the coefficient matrix A.
NB : The partitioning blocking factor.
P : The number of process rows.
Q : The number of process columns.
Time : Time in seconds to solve the linear system.
Gflops : Rate of execution for solving the linear system.
The following parameter values will be used:
N : 50004
NB : 256
PMAP : Row-major process mapping
P : 1
Q : 4
PFACT : Right
NBMIN : 4
NDIV : 2
RFACT : Crout
BCAST : 1ringM
DEPTH : 1
SWAP : Mix (threshold = 64)
L1 : transposed form
U : transposed form
EQUIL : yes
ALIGN : 8 double precision words
--------------------------------------------------------------------------------
- The matrix A is randomly generated for each test.
- The following scaled residual check will be computed:
||Ax-b||_oo / ( eps * ( || x ||_oo * || A ||_oo + || b ||_oo ) * N )
- The relative machine precision (eps) is taken to be 1.110223e-16
- Computational tests pass if scaled residuals are less than 16.0
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR11C2R4 50004 256 1 4 1393.69 5.9810e+01
HPL_pdgesv() start time Fri Feb 23 18:28:10 2024
HPL_pdgesv() end time Fri Feb 23 18:51:24 2024
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 3.65051815e-03 ...... PASSED
================================================================================
Finished 1 tests with the following results:
1 tests completed and passed residual checks,
0 tests completed and failed residual checks,
0 tests skipped because of illegal input values.
--------------------------------------------------------------------------------
End of Tests.
================================================================================
geerlingguy commented
Full cluster again, with Ps/Qs being 4/8: 224.60 Gflops, 73 W, for 3.08 Gflops/W
================================================================================
HPLinpack 2.3 -- High-Performance Linpack benchmark -- December 2, 2018
Written by A. Petitet and R. Clint Whaley, Innovative Computing Laboratory, UTK
Modified by Piotr Luszczek, Innovative Computing Laboratory, UTK
Modified by Julien Langou, University of Colorado Denver
================================================================================
An explanation of the input/output parameters follows:
T/V : Wall time / encoded variant.
N : The order of the coefficient matrix A.
NB : The partitioning blocking factor.
P : The number of process rows.
Q : The number of process columns.
Time : Time in seconds to solve the linear system.
Gflops : Rate of execution for solving the linear system.
The following parameter values will be used:
N : 100009
NB : 256
PMAP : Row-major process mapping
P : 4
Q : 8
PFACT : Right
NBMIN : 4
NDIV : 2
RFACT : Crout
BCAST : 1ringM
DEPTH : 1
SWAP : Mix (threshold = 64)
L1 : transposed form
U : transposed form
EQUIL : yes
ALIGN : 8 double precision words
--------------------------------------------------------------------------------
- The matrix A is randomly generated for each test.
- The following scaled residual check will be computed:
||Ax-b||_oo / ( eps * ( || x ||_oo * || A ||_oo + || b ||_oo ) * N )
- The relative machine precision (eps) is taken to be 1.110223e-16
- Computational tests pass if scaled residuals are less than 16.0
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR11C2R4 100009 256 4 8 2969.05 2.2460e+02
HPL_pdgesv() start time Fri Feb 23 19:03:56 2024
HPL_pdgesv() end time Fri Feb 23 19:53:25 2024
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 2.41638360e-03 ...... PASSED
================================================================================
Finished 1 tests with the following results:
1 tests completed and passed residual checks,
0 tests completed and failed residual checks,
0 tests skipped because of illegal input values.
--------------------------------------------------------------------------------
End of Tests.
================================================================================
geerlingguy commented
With 1/32 Ps/Qs, the result is: 171.50 Gflops at 68W average, so 2.52 Gflops/W.
I think 4/8 is the most efficient layout on this cluster.