Earlier this week, Advanced Micro Devices finally introduced the first true quad-core microprocessor in the PC market, code-named "Barcelona" and launched as part of AMD's Opteron line. David Kanter provided an excellent technical analysis of Barcelona on his Real World Technologies site.
Barcelona is not the absolutely fastest processor on the market. For single-core performance, both IBM and Intel offer faster chips. With multiple cores working, Intel and Sun can claim higher performance per socket--Intel because it can put two high-frequency dual-core chips in one socket, and Sun because it has an eight-core processor (the UltraSparc T2).
Although I'm sure AMD wishes it could claim those titles, there's another metric that matters even more to some customers. Barcelona delivers high performance per watt of power consumption-- at the chip level, and even more so at the system level. (To be fair to Sun, the T2 is also very good at this, but it can't run Windows and doesn't support dual-processor configurations, so its market niche is very narrow.)
As CNET's Tom Krazit described in a blog post ahead of the Barcelona launch (AMD's Barcelona is not a savior, yet), and AMD is trying to draw attention to Barcelona's improved power-management features by creating a new way to express processor power consumption: Average CPU Power (ACP). This is not a terribly useful idea, since it presupposes a certain workload. If there's one thing the server market doesn't have, it's one standard workload.
But the underlying truth here is that Barcelona does a good job of not using much power when it's not doing much work. This sounds like such a simple thing, but it's often overlooked by CPU and server makers. Since servers almost never run at 100 percent all the time, power efficiency under less than full load is very important.
In his Barcelona review, Anandtech's Johan De Gelas gives some power-consumption figures for servers based on Barcelona and Intel's competing Xeon processors. Although the two systems he compared were essentially identical other than the CPU and chipset on the motherboard, the AMD system consumed 27 percent less power when idle and 14 percent less when busy. The AMD system was also faster at typical server tasks (but Intel offers faster--and hotter--chips than the one tested here).
That performance/power advantage will be enough to sell Barcelona into many data centers. Krazit reports that "The four major server vendors in the world--Dell, Hewlett-Packard, IBM and Sun Microsystems--all plan to use Barcelona in their servers."
I think that's proof enough of Barcelona's commercial potential.