Dell reveals AMD Barcelona benchmarks

It is good to see AMD cathing up and even overtaking Intel in SPEC-fp ( it stimulates the competition ), the reason for the advantage should worry AMD. If the only reason for this advantage is found in the on chip memory controller, Intel will be ahead again (if ever so slightly) with Nehalem. Then again, if AMD doesn't screw up, we should have them fighting again for top honours, with us the end-users as winners.

Hello,

Interesting news !
Unfortunately, there does not seem to be any benchmark for single-socket systems at heise. It It would have been a nice data point for scaling.

The other interesting point, beside scaling, is that AMD would be very competitive if they could ramp up the clock to Intel's level !
Normalising per GHz gives:

AMD Intel
int, 2 sockets: 18.5 18.9
float, 2 sockets: 16.4 12.6
int, 4 sockets: 16.7 15.1
float, 4 sockets: 15.2 9.2

So the Barcelona core is not that bad, at least from the perspective of the SPEC benchmarks.

Cheers,

Stephane

Clock-for-clock it's either a dead-heat or in the other two cases a win for AMD, albeit a slight one. In this race between the tortoise and the hare, the tortoise is still managing to tread water.

well, nehalem will be out soon with integrated memory controllers and csi. :/

Well, well.... seems like all the Intel Fanboiz will be chewing their nails tonight!
Finally the figures speak for themselves; and it's not just the biazed review sites either. "Dell" all-mighty has produced these numbers without Hector fumbling in someone's pants.
So it's true then: AMD CPUs do scale better than Intel's. Not only that, but they do actually do more work than Intel's per clock... (as usually is the case)
Horah for the competition!
Oh, and from another article, I see that IBM and Unisys announce AMD Barcelona servers,,,,
Sounds like the tide is turning to me. Better start hoisting those Intel sails, Fanboiz, and start sailing back to shore!

In addition to the FSB bottleneck and lack of IMC, one thing to remember that explains why Intel performs so poorly in the 4P benchmarks is that their current quad-core Xeon MP processors ("tigerton" 73xx series) are using previous-gen technology: 65nm core microarch. Whereas the 2P Xeon DP family (54xx) is based on the current-gen 45nm core microarch.

Intel has always been lagging when it comes to bringing their latest tech to the 4P+ market. And it doesn't look like this is going to change; AFAIK the 45nm Xeon MP ("dunnington" 74xx) is only planned for H2 2008 (most likely Q4).

This is one of the reasons why AMD is the leader by volume in the 4P market. In Sep 2006 they had 56% marketshare: http://anandtech.com/IT/showdoc.aspx?i=3091&p=2 I don't have more current numbers but AMD's marketshare is most likely even bigger today because, remember, 5 months ago Intel's ENTIRE 4P offering was based on the NetBurst arch (yep, "tigerton" was officially released in sep 2007 but only became available in quantity around dec 2007). Hard to believe, isn't it ?

"one thing to remember that explains why Intel performs so poorly in the 4P benchmarks is that their current quad-core Xeon MP processors ("tigerton" 73xx series) are using previous-gen technology: 65nm core microarch."

And the Opterons are using what...? 

Anyway, a die shrink by itself doesn't give any speed advantages. It will generally allow higher clock speeds at the same temperature, but there are other issues involved that can limit that. 

And in any case that would affect 2P and 4P equally. The difference in scaling is mainly due to the interconnects.

seems the intel fanboyze are taking a lot for granted believing that intel's nehalem will swamp amds product, seems to me that amd has the most experience and expertise in this area, and also intel is so cheap it will not use soi but will go for something cheaper. That is why they bolt two processors together is that it is cheaper.
Intels to cheap to ever remain ahead very long.

Intel is not cheap they spend Billions on R&D

they had enough mulah to dualy develop on P-M (wich lead to c2d) and P4 

but they are comfortable with thier position right now as AMD was before C2D.....

The tables generally change every few years and while it looks like AMD is ahead in 4P it is not by the margins Intel was and is in the desktop.

me fanboi....actually I prefer AMD ...Americans do tend to fight for the underdog you know. (and rebels)
still I own more Intel based systems then AMD I go for the price/perf at the time...

You COMPLETELY missed my point. Let me reexplain:

It is well known that Intel's 45nm processors perform better than 65nm ones at equal frequency not because of the die shrink, but because of the microarchitectural improvements Intel made. Since their 4P Xeon 73xx series DON'T have these improvements, they are disadavantaged compared to the 54xx series. Therefore, this is one additional reason explaining why Intel's 4P Xeons are less competitive than the 2P Xeons.

Got it now ? :)

By the way, you seem to imply that if AMD is using a 65nm process, then they should be at the same performance level than Intel's 65nm. This is wrong. First this article show the contrary. Second, AMD's 65nm Opterons are based on a 6-month old microarchitecture (Barcelona, released in Dec 2007 or so) whereas Intel's 65nm Xeons are based on a 2-year old microarchitecture (Core released in Jun 2006).