AMD DEMOed Barcelona Native Quad Core

[IntelUser2000]

Estimated Performance 1
Estimated Performance 2

Looks pretty good.

BUMP

Sorry to refresh an old thread, but if any of you guys can correctly interprete the graph, look again. The 40% performance increase in SpecFPRate looks like its over Xeon 5355 Clovertown, while the 70% performance in OLTP is clearly over Opteron/Xeon 5160 Woodcrest(they score similar according to the graph), and not too far from Xeon 5355, but I can't tell exactly the numbers due to the hard to read graph.

You can also take a look at www.spec.org. Dual processor Xeon 5150(2.66GHz) scores around 78, while Single Processor Xeon 5355 scores around 60. Single processor Xeon 5150 scores around 45.

The reason DP Xeon 5150 scores 30% higher than Xeon 5355 with same clock speed and same core count is because DP Xeon has twice the memory bandwidth. So Intel's problem in the multi-processor segment is memory bandwidth. 30% performance increase by fixing memory bandwidth covers most of the 40% advantage Barcelona will have over Clovertown in SpecFPRate, which is also known for K8's to perform well even over Woodcrest(keep in mind Woodcrest has much greater advantage at SpecFP, which is a single threaded version of the benchmark), and is very memory bandwidth sensitive. Single thread?? They might as well be equal.

 

[lopri]

'Direct Connect', or HyperTransport isn't just a marketing term. Its effectiveness was shown most dramatically from Athlon XP -> Athlon 64.