Skip to content

Settings

Experimental Settings

List of Machines

Intel

Name CPU (GFlops) Inst.(Up to) Threads RAM GPU Cores
sable-tigger i7-8700K @ 3.70GHz (320GF) AVX2 12 (1x12) 32 GB GeForce GTX 1080 Ti (c6.1) 3584
sable-beaver i7-9400F @ 2.90GHz (306GF) AVX2 6 (1x6) 32 GB (Two cards, see below) ...
sable-intel E7-4850 @ 2.00GHz (240GF) SSE4.2 80 (4x20) 128 GB N/A N/A
sable-cheetah i7-3930K @ 3.20GHz (135GF) AVX 12 (1x12) 16 GB GeForce GT 720 (c3.5) 192
sable-lynx i7-3820 @ 3.60GHz (105GF) AVX 8 (1x8) 8 GB GeForce GTX 560 Ti (c2.1) 384
sable-deer i7-3820 @ 3.60GHz (105GF) AVX 8 (1x8) 16 GB GeForce GTX 560 Ti (c2.1) 384
sable-cougar i7-3820 @ 3.60GHz (105GF) AVX 8 (1x8) 16 GB GeForce GTX 560 Ti (c2.1) 384
chf-pc i7-3610QM @ 2.30GHz (87 GF) AVX 8 (1x8) 8 GB GeForce GT 630M (c2.1) 96
chf-mac i5-5257U @ 2.70GHz (70 GF) AVX2 4 (1x4) 8 GB Intel Iris Graphics 6100 Void
sable-venus i5-660 @ 3.33GHz (25 GF) SSE4.2 4 (1x4) 12 GB GeForce 8400 GS Void

AMD

Name CPU (GFlops) Inst.(Up to) CPUs * RAM GPU Cores
sable-amd1 Opteron 6274 @ 2.20 GHz 64 (4x8x2) 64 GB AMD ATI RD890
sable-amd2 Opteron 6274 @ 2.20 GHz 64 (4x8x2) 64 GB AMD ATI RD890

*Note: AMD CPUs = socket*core*thread

New GPUs We Already Purchased

Note: From Apple support, we know that chf-mac can work with OpenGL 4.1 and OpenCL 1.2.

GCC Optimizations

Native optimization configuration (generated with gcc -Q --help=target -march=native)

Hardware

GPU

The memory bandwidth is determined by the memory clock, the memory type, and the memory width.

Name Cores Clock (MHz) M. Speed M. Type M. Width M. Bandwidth (GB/s)
1080Ti 3584 1582 11 Gbps GGDR5X 352-bit 484

Software

MonetDB

Configurations on sable-intel

- Data: /mnt/local/datafarm
- Script: /mnt/local/script
  + ./runsql 6  // run tpch query 6
- TPC-H: /mnt/local/tpch_2_17_0

Query timing

call sys.querylog_enable();
\<../tpch_2_17_0/chf/myqueries/6.sql
call sys.querylog_disable();
select id, query, avg(run) as run_time from sys.querylog_history group by id,query;

OpenMP

OpenMP affinity (link)

OMP_PLACES=cores
OMP_PROC_BIND=spread
// or
GOMP_CPU_AFFINITY="0-37:4 1-38:4 2-39:4 3-40:4"