Craig, yes 6 cores a processor. It's Dunnington not the older Xeons they compared...

So no, Sylvie hasn't got her maths wrong. However some of this seems a bit unfounded with a lack of proof, however despite the mismatch it is obvious AMD has still got a lot of catching up to do despite it's recent successful launch.

The two tests are using differnt OS and different databases. Win2003 and DB2 vs Suse and Oracle.

Actually launching *early*, which is an even bigger improvement over launching later than late :)

So this article is to try and confirm that the scores were based on dodgy mismatch of two non-identical machines, but you guys dont do yourselves any favours when it comes to clearing it up - i quote:

" IBM System x3950 M2, 8 Processors / 48 Cores / 48 Threads"

6 cores per processor??? you mean to say 12 processors and 48 threads???

Have you followed the link to the SAP benchmark ? It says the XEON's result of 9200 was achieved on 48 core system (8-way x 6 cores/chip), while Operon 8384' result of 7010 was achieved on 32 core system (8-way x 4 cores/chip). You call that oranges-vs-oranges comparison ? The XEON system of equivalent 32 core configuration (8-way x 4 cores/chip) was around 6600 ...

Hey, but what do you expect of George Ou, the famous AMD basher ?

I think this article makes too many lame excuses for AMD, it is perfectly fair to compare Dunnington to Shanghai for the simple reason that those two products will be in direct competition with each other in the server market. Its really AMD's own fault that they haven't come out with a 6-core processor yet to compete and quite frankly, the customers of these machines aren't going to care.

Less barking and more action is all I have to say.

You probably mean to say that nehalem seems to consume 322 watts of power compared to amd's 264. Oh, as usual, a poorly done article with little to no facts to back it up.

AMD still leads by huge margin in FP, almost 50% if compare clock-by-clock and 26% at the clocks mentioned above. That's like one generation ahead. For my science only FP matters

I suspect that EPA's problem with Java floating point development may explain some of the differences. This paper from Lawrence Berkeley/NERSC and the heavy weighting of the Java benchmarks http://www.cs.berkeley.edu/~wkahan/JAVAhurt.pdf may explain some of the SPEC Power differential. EPA has proposed a different benchmark for inclusion in the EnergyStar 5.0 certification. The intent is to give a more representative evaluation of floating point performance. There is no question that Intel's performance in integer based apps is superior to AMD. Which cpu is superior is going to really depend on the application Int vs FP. The other thing that SPEC Power doesn't address are scaling issues when you interconnect 8 or more CPU's. I was quite surprised to see that the results are sensitive to power factor, altitude and line voltage quality in addition to temperature. The number of variables that can affect the testing are greater than the number of tests to be run.

I have to note that the comparison of the Intel Xeon L5430 with the AMD 2384 Shanghai Opteron is again not 100% apples to apples.

While the Intel based system was equipped with a 500 W power supply, the AMD based system only had 350 W power supply. The latter seems a bit too small in terms of efficiency, or am i wrong?

Second, the AMD based system had 16 GB of RAM using 4x4 GB DDR2-800 DIMMs, the Intel based system is listed with 2x4 GB DDR2-667 DIMMs. Allthough these might be constraints of the different architectures it is still a difference.

mrc