Power Marketing

I thought I would avoid controversial subjects but I read the IBM's Power7 pitch deconstructed on The Register and they did not deconstruct enough for my taste.
While looking at the IBM slides the article showed I got the impression that some alien technology was feature as there were claims that the Power7 allowed for solutions that were 30 times better...
The article gave a few explanations but the comparison with the HP DL380 G5 really left me with the impression that something seriously wrong was going on. It was basically explaining that you could replace 71 of those servers (credited with a 36 specintrate 2006) with one Power 7 based 750 (credited with 1060).

As I was really puzzled I went to spec.org to check some numbers and did some fact checking. I indeed found a HP DL380 G5 with a specintrate of 36 but that was a 1 slot server when these machines make more sense for 4 slots (they can use 2KW after all...). The 4 CPU slot one had a score other 100...

So, obviously nobody would have 71 of DL380 with only one slot occupied.
Then the article also mentioned that the 71 machines were assumed to have a 20% load while the 750 was supposed to run at 60% thanks to the consolidation.

So basically, the slide compares the worst possible architecture to the new Power 7 to make it look like it offers out of this world performance.

With a decent architecture and a subset of the hardware envisioned, it's quite easy to see that the comparison would not be as dramatic. You could fill the 4 slots of the machines and consolidate some load to 50% load (very reasonable) and you would consolidate the bad architecture down to 7 servers.

So basically, the new Power 7 is 7 times better than some 2007 hardware on the specintrate...
It's not bad but it's not quite as impressive.

Then there is the fact that there is no specint (single threaded general benchmark) nor specfp (single thread floating point benchmark) information for the Power 7. I consider that to be very telling as IBM stopped publishing these results years ago when it was clear that their Power processors could not deliver the single thread performance of even a low end Intel or AMD processor. They had used specfp for a while but they stopped that too.
I have often used the specint to predict the performance of applications at work with good success. I can't say the same for the specintrate, though. So, I would tend to trust the specint first and get x86 hardware over Power any day.

In the end, it really looked like the slide was so artificial and tweaked that it was completely uninformative. I'll take note and know that next time I am invited by IBM sales reps I don't need to bother to attend...