TheOtherPhil
01-18-2005, 04:57 AM
Just for something to do, I decided to bench both the 64bit and 32bit d.net clients for RC5 and OGR performance. I am using an AMD FX51 @ 2.3GHz on the Asus SK8V mobo. OS is Suse 9.2 Pro 64bit. The results are quite interesting....the fastest core is highlighted in bold:
Client: [AMD64/ELF] v2.9009.494
[Jan 18 09:36:49 UTC] RC5-72: using core #0 (SNJL 3-pipe).
[Jan 18 09:37:08 UTC] RC5-72: Benchmark for core #0 (SNJL 3-pipe)
0.00:00:16.84 [8,372,188 keys/sec]
[Jan 18 09:37:08 UTC] RC5-72: using core #1 (ANSI 4-pipe).
[Jan 18 09:37:27 UTC] RC5-72: Benchmark for core #1 (ANSI 4-pipe)
0.00:00:16.64 [3,724,355 keys/sec]
[Jan 18 09:37:27 UTC] RC5-72: using core #2 (ANSI 2-pipe).
[Jan 18 09:37:46 UTC] RC5-72: Benchmark for core #2 (ANSI 2-pipe)
0.00:00:16.42 [3,375,296 keys/sec]
[Jan 18 09:37:46 UTC] RC5-72: using core #3 (ANSI 1-pipe).
[Jan 18 09:38:04 UTC] RC5-72: Benchmark for core #3 (ANSI 1-pipe)
0.00:00:16.60 [2,188,762 keys/sec]
[Jan 18 09:38:23 UTC] OGR-P2: Benchmark for core #0 (GARSP 6.0-64)
0.00:00:16.09 [30,145,118 nodes/sec]
Client: [x86/ELF] v2.9009.494
[Jan 18 09:38:40 UTC] RC5-72: using core #0 (SES 1-pipe).
[Jan 18 09:38:59 UTC] RC5-72: Benchmark for core #0 (SES 1-pipe)
0.00:00:16.40 [3,569,952 keys/sec]
[Jan 18 09:38:59 UTC] RC5-72: using core #1 (SES 2-pipe).
[Jan 18 09:39:17 UTC] RC5-72: Benchmark for core #1 (SES 2-pipe)
0.00:00:16.43 [6,670,670 keys/sec]
[Jan 18 09:39:17 UTC] RC5-72: using core #2 (DG 2-pipe).
[Jan 18 09:39:37 UTC] RC5-72: Benchmark for core #2 (DG 2-pipe)
0.00:00:17.05 [8,278,917 keys/sec]
[Jan 18 09:39:37 UTC] RC5-72: using core #3 (DG 3-pipe).
[Jan 18 09:39:56 UTC] RC5-72: Benchmark for core #3 (DG 3-pipe)
0.00:00:16.88 [8,360,228 keys/sec]
[Jan 18 09:39:56 UTC] RC5-72: using core #4 (DG 3-pipe alt).
[Jan 18 09:40:14 UTC] RC5-72: Benchmark for core #4 (DG 3-pipe alt)
0.00:00:16.41 [8,590,932 keys/sec]
[Jan 18 09:40:14 UTC] RC5-72: using core #5 (SS 2-pipe).
[Jan 18 09:40:33 UTC] RC5-72: Benchmark for core #5 (SS 2-pipe)
0.00:00:16.45 [8,582,452 keys/sec]
[Jan 18 09:40:33 UTC] RC5-72: using core #6 (GO 2-pipe).
[Jan 18 09:40:52 UTC] RC5-72: Benchmark for core #6 (GO 2-pipe)
0.00:00:16.37 [9,197,893 keys/sec]
[Jan 18 09:40:52 UTC] RC5-72: using core #7 (SGP 3-pipe).
[Jan 18 09:41:11 UTC] RC5-72: Benchmark for core #7 (SGP 3-pipe)
0.00:00:17.19 [8,200,968 keys/sec]
[Jan 18 09:41:11 UTC] OGR-P2: using core #0 (GARSP 6.0-A).
[Jan 18 09:41:30 UTC] OGR-P2: Benchmark for core #0 (GARSP 6.0-A)
0.00:00:16.33 [18,249,519 nodes/sec]
[Jan 18 09:41:30 UTC] OGR-P2: using core #1 (GARSP 6.0-B).
[Jan 18 09:41:48 UTC] OGR-P2: Benchmark for core #1 (GARSP 6.0-B)
0.00:00:16.45 [12,850,946 nodes/sec]
Notice how the 32bit client is slightly faster than the 64bit client for RC5-72 - the fastest 32bit core (GO 2-pipe) is 825705 keys/sec faster than the fastest 64bit core (SNJL 3-pipe).
The 64bit client running OGR (GARSP 6.0-64) blitzes the 32bit client by nearly 12m nodes/sec on the same hardware.
:shocked:
So, run the 64bit client for OGR and the 32bit client for RC5-72 if running x86_64bit Linux.
Client: [AMD64/ELF] v2.9009.494
[Jan 18 09:36:49 UTC] RC5-72: using core #0 (SNJL 3-pipe).
[Jan 18 09:37:08 UTC] RC5-72: Benchmark for core #0 (SNJL 3-pipe)
0.00:00:16.84 [8,372,188 keys/sec]
[Jan 18 09:37:08 UTC] RC5-72: using core #1 (ANSI 4-pipe).
[Jan 18 09:37:27 UTC] RC5-72: Benchmark for core #1 (ANSI 4-pipe)
0.00:00:16.64 [3,724,355 keys/sec]
[Jan 18 09:37:27 UTC] RC5-72: using core #2 (ANSI 2-pipe).
[Jan 18 09:37:46 UTC] RC5-72: Benchmark for core #2 (ANSI 2-pipe)
0.00:00:16.42 [3,375,296 keys/sec]
[Jan 18 09:37:46 UTC] RC5-72: using core #3 (ANSI 1-pipe).
[Jan 18 09:38:04 UTC] RC5-72: Benchmark for core #3 (ANSI 1-pipe)
0.00:00:16.60 [2,188,762 keys/sec]
[Jan 18 09:38:23 UTC] OGR-P2: Benchmark for core #0 (GARSP 6.0-64)
0.00:00:16.09 [30,145,118 nodes/sec]
Client: [x86/ELF] v2.9009.494
[Jan 18 09:38:40 UTC] RC5-72: using core #0 (SES 1-pipe).
[Jan 18 09:38:59 UTC] RC5-72: Benchmark for core #0 (SES 1-pipe)
0.00:00:16.40 [3,569,952 keys/sec]
[Jan 18 09:38:59 UTC] RC5-72: using core #1 (SES 2-pipe).
[Jan 18 09:39:17 UTC] RC5-72: Benchmark for core #1 (SES 2-pipe)
0.00:00:16.43 [6,670,670 keys/sec]
[Jan 18 09:39:17 UTC] RC5-72: using core #2 (DG 2-pipe).
[Jan 18 09:39:37 UTC] RC5-72: Benchmark for core #2 (DG 2-pipe)
0.00:00:17.05 [8,278,917 keys/sec]
[Jan 18 09:39:37 UTC] RC5-72: using core #3 (DG 3-pipe).
[Jan 18 09:39:56 UTC] RC5-72: Benchmark for core #3 (DG 3-pipe)
0.00:00:16.88 [8,360,228 keys/sec]
[Jan 18 09:39:56 UTC] RC5-72: using core #4 (DG 3-pipe alt).
[Jan 18 09:40:14 UTC] RC5-72: Benchmark for core #4 (DG 3-pipe alt)
0.00:00:16.41 [8,590,932 keys/sec]
[Jan 18 09:40:14 UTC] RC5-72: using core #5 (SS 2-pipe).
[Jan 18 09:40:33 UTC] RC5-72: Benchmark for core #5 (SS 2-pipe)
0.00:00:16.45 [8,582,452 keys/sec]
[Jan 18 09:40:33 UTC] RC5-72: using core #6 (GO 2-pipe).
[Jan 18 09:40:52 UTC] RC5-72: Benchmark for core #6 (GO 2-pipe)
0.00:00:16.37 [9,197,893 keys/sec]
[Jan 18 09:40:52 UTC] RC5-72: using core #7 (SGP 3-pipe).
[Jan 18 09:41:11 UTC] RC5-72: Benchmark for core #7 (SGP 3-pipe)
0.00:00:17.19 [8,200,968 keys/sec]
[Jan 18 09:41:11 UTC] OGR-P2: using core #0 (GARSP 6.0-A).
[Jan 18 09:41:30 UTC] OGR-P2: Benchmark for core #0 (GARSP 6.0-A)
0.00:00:16.33 [18,249,519 nodes/sec]
[Jan 18 09:41:30 UTC] OGR-P2: using core #1 (GARSP 6.0-B).
[Jan 18 09:41:48 UTC] OGR-P2: Benchmark for core #1 (GARSP 6.0-B)
0.00:00:16.45 [12,850,946 nodes/sec]
Notice how the 32bit client is slightly faster than the 64bit client for RC5-72 - the fastest 32bit core (GO 2-pipe) is 825705 keys/sec faster than the fastest 64bit core (SNJL 3-pipe).
The 64bit client running OGR (GARSP 6.0-64) blitzes the 32bit client by nearly 12m nodes/sec on the same hardware.
:shocked:
So, run the 64bit client for OGR and the 32bit client for RC5-72 if running x86_64bit Linux.