Thread: Quad Quad-cores
View Single Post
Old 2008-03-21, 18:43   #1
SlashDude
 
SlashDude's Avatar
 
Aug 2002
Minneapolis, MN

22×3×19 Posts
Default Quad Quad-cores

FYI-
I've had the opportunity to evaluate an HP DL580 G5 and an IBM System x3850 M2.

Both systems had Windows 2003 server Enterprise x64 with 4 x quad-core CPU's. (16 cores) The HP server was 2.4GHz, 1066MHz FSB and 32GB ram, and the IBM server was a 2.98GHz 1333MHz FSB and 64GB ram. (Note - IBM states a ~30% memory speed increase)

I ran manual tests using exponents in the 47,000,000 range.
Here are the results:
I used 0-F (hex) for the CPU names. 0-3, 4-7, 8-B, and C-F were on the same chip.

HP DL 580:
Code:
Cores running                         Time per iteration
0                                         0.067			
0,8                                       0.067			
2,6,A                                     0.068			
0,4,8,C                                   0.069			
2,6,A,C                                   0.069			
0,2                                       0.072			
0,2,8,A                                   0.074			
0,2,4,6,8,A                               0.075			
2,6,8,A,C                                 0.075	.069 on 2,6,C		
0,1                                       0.076			
2,4,6,8,A,C                               0.076	.070 on 2,C		
0,2,4,6,8,A,C                             0.08	.072 on C		
0,2,4,6,8,A,C,E                           0.085			
0,1,2                                     0.09	.077 on 2		
0,1,4,5,8,9,C,D                           0.093			
0,1,2,3                                   0.103			
0,1,2,3,4,5,6,7                           0.114			
0,1,2,3,8,9,A,B                           0.114			
0,2,4,5,6,8,A,C,E                         0.12	.93 on 6		.86 on 0,2,8,A,C,E
0,2,4,5,6,8,9,A,C,E	                  0.124	.96 on 6,A		.89 on 0,2,C,E
0,1,2,3,4,5,6,7,8,9,A,B                   0.125			
0,1,2,4,5,6,8,9,A,C,E                     0.13	.98 on 2,6,A,E		.091 on C
0,1,2,4,5,6,8,9,A,C,D,E                   0.134	.1 on 2,6,A and E		
0,1,2,4,5,6,7,8,9,A,C,D,E                 0.155	.135 on 1,2,9,A,C,D		.101 on 2, A and E
0,1,2,4,5,6,7,8,9,A,B,C,D,E               0.156	.138 on 1,2,C,D		.102 on 2 and E
0,1,2,3,4,5,6,7,8,9,A,B,C,D,E             0.158	.141 on C,D		.103 on E
All                                       0.16
IBM System x3850 M2:
Code:
Cores running                         Time per iteration
0                                         0.061		
0,8                                       0.061		
2,6,A                                     0.061		
0,4,8,C                                   0.061		
2,6,A,C                                   0.061		
0,2                                       0.064		
0,2,8,A                                   0.065		
0,2,4,6,8,A                               0.066		
2,6,8,A,C                                 0.066	0.061	
0,1                                       0.068		
2,4,6,8,A,C                               0.066	0.061	
0,2,4,6,8,A,C                             0.066	0.061	
0,2,4,6,8,A,C,E                           0.066		
0,1,2                                     0.077	0.07	
0,1,4,5,8,9,C,D                           0.079		
0,1,2,3                                   0.087		
0,1,2,3,4,5,6,7                           0.088		
0,1,2,3,8,9,A,B                           0.088		
0,2,4,5,6,8,A,C,E                         0.079	0.072	0.067
0,2,4,5,6,8,9,A,C,E	                  0.08	0.073	0.067
0,1,2,3,4,5,6,7,8,9,A,B                   0.091		
0,1,2,4,5,6,8,9,A,C,E                     0.83	0.074	0.068
0,1,2,4,5,6,8,9,A,C,D,E                   0.85	0.075	
0,1,2,4,5,6,7,8,9,A,C,D,E                 0.1	0.087	0.077
0,1,2,4,5,6,7,8,9,A,B,C,D,E               0.103	0.09	0.078
0,1,2,3,4,5,6,7,8,9,A,B,C,D,E             0.107	0.093	0.08
All                                       0.111
Now the true reason for these tests was a VMware ESX server evaluation.
Here are the numbers running the same tests within a "virtual machine" running on the above hardware (Times averaged between all running clients.) Each guest OS was configured with a single CPU and 512MB RAM. I didn't have an x64 OS image, so these tests were done using the x86 client on Windows Server 2003 standard.
Code:
HP	IBM	Faster%
0.068	0.062	9.68%
0.069	0.062	11.29%
0.070	0.062	12.90%
0.074	0.062	19.35%
0.075	0.063	18.35%
0.076	0.068	11.76%
0.081	0.068	18.45%
0.090	0.071	26.67%
0.102	0.079	29.94%
0.101	0.083	21.65%
0.110	0.084	30.31%
0.119	0.088	35.42%
0.129	0.09	43.36%
0.139	0.097	42.57%
0.144	0.103	39.50%
0.155	0.112	38.62%
Hopefully this will be of interest to someone else too.

-SD
LLR Note:
I had a chance to run LLR on the HP, and found that a single test in the x*2^~540000-1 range took 668 seconds to complete. With 16 tests running, it took 673 seconds. (a 5 second slowdown per test - less then 1% per test!)

<Edit - Fixed code boxes>

Last fiddled with by SlashDude on 2008-03-21 at 18:58
SlashDude is offline   Reply With Quote