Performane measurements were based on realistic data production chains as well as

Performance	AMD Opteron(tm) Processor 6380	Intel(R) Xeon(R) CPU E5-2660 v2 @ 2.20GHz	Ratio Intel/AMD	AMD -> Intel %tage gain in performance	averages performance gain
Composite-500 D	818.3	1086.8	1.33	32.81%
Composite-500 O	819.2	1085.1	1.32	32.46%
Composite-5000 D	768	1095	1.43	42.58%
Composite-5000 O	781	1118	1.43	43.15%	37.75%
Fit test (FP ops) D	551.8	1019.1	1.85	84.69%
Fit test (FP ops) O	590.4	1076.1	1.82	82.27%	83.48%
	43.41	26.09
Real Data reco D	230.36	383.36	1.66	66.42%
	29.55	19.06
Real Data reco O	338.41	524.66	1.55	55.04%	60.73%
	500	313
MC full reco D	20	31.95	1.60	59.74%
	362	233
MC full reco O	27.62	42.92	1.55	55.36%	57.55%
Table 1 averages	494.51	746.3

The below table (massaging the number for HT effect estimated to be a 30% gain at most, consistent with past measurements) shows the gain when moving from AMD to Intel based CPU. it is dereived from the table above.

AMD -> Intel %tage gain in performance
Composite-X	24.54%
Fit test	54.26%
Real Data reco	39.47%
MC full reco	37.41%

The improvement number need to be compared to the overall cost loss by moving from AMD to Intel (so much more expensive at this stage). If the performance gain is comparable to the cost increase, numbers need to be inspected more closely (as the overall performance for a bulk pruchase would relate to the number of nodes we can acquire rather than a single CPU performance) ...

cost impact evaluation	AMD Opteron(tm) Processor 6380	Intel(R) Xeon(R) CPU E5-2660 v2 @ 2.20GHz	Ratio rescaled by core Intel/AMD	AMD ->Intel core rescaled %tage increase in cost
1U	6635	13289	1.60	60.23%
2U	10980	17263	1.26	25.78%

... and they are.

• The 1 U would not make it (60% cost increase for at best a 54% performance gain)
• The 2 U cost increase going Intel is ~ 26% while the performance gain be in ~ 38%. This is a one node cost impact.

Folding storage and CPU cost, we arrive at the below table. In that final overview, assuming some budget, we calculate how many nodes we can acquire for the total budget hence a total CPU and storage acquisition. Here, the performance is based on a grand average of all numbers from Table 1 (with the usual rescale for Intel based on the HT and number of virtual cores).

[some budget assumed]	AMD Opteron(tm) Processor 6380	Intel(R) Xeon(R) CPU E5-2660 v2 @ 2.20GHz
Num node 1U	105.5	52.68
Num node 2U	63.75	40.55
Storage 1U (PB)	1.64	0.81
Storage 2U (PB )	1.97	1.25	57.50%	more storage with AMD
CPU power 1U	51923.5	25224.9
CPU power 2U	31154.1	19403.77	60.56%	more CPU with AMD (cost)

Conclusion: we are better off with AMD :-( ...