Intel flaunts Xeon mark two

Dell and Intel this week showed off servers using the chipmaker's forthcoming high-end "Tulsa" Xeon, a chip Intel has begun shipping but not formally announced.

The chips were used in an unlabeled Intel server and a Dell PowerEdge 6850 on display at the LinuxWorld Conference and Expo in San Francisco. Linux is popular chiefly on servers, so AMD, Dell, HP, Intel and others with server products are prime sponsors of the show.

Tulsa, a dual-core processor, is the last of the ill-fated NetBurst lineage of x86 chips from Intel. The NetBurst design in recent years was more notable for increases in power consumption than in performance but it's now been largely replaced by the Core micro-architecture that performs better and uses less electricity.

Intel released a Core-based Xeon called Woodcrest in June but that processor is used only in dual-processor servers. Tulsa is for more powerful four-processor machines.

Intel's NetBurst fumble left room for rival AMD to enter the x86 server market which Intel once had to itself. AMD this week announced its new "Rev F" line of dual-core Opteron processors, which in 2007 will be supplanted by quad-core models that plug into the same "Socket F".

Intel began shipping Tulsa chips for revenue this quarter, chief executive Paul Otellini said in July, bringing the products to market a quarter earlier than planned.

Each Tulsa processor core has 1MB of level-two cache memory, and the two cores share a whopping 16MB of level-three cache memory, more than any other x86 processor. Cache memory stores instructions and data so the information can be retrieved more quickly than that in main memory.

Tulsa uses 1.3 billion transistors - nearly as many as the 1.6 billion in Intel's new "Montecito" version of the Itanium processor, according to an Intel representative at the show. Intel can afford to build large caches better than its rivals, because it has moved more quickly to a new manufacturing process with 65-nanometer circuitry elements, letting more transistors be crammed into a given surface area. Most rivals are still building chips with a 90-nanometer process.

The processor runs at a top speed of 3.4GHz, according to an Intel presentation at the show. Tulsa comes in two variants: one high-performance model that consumes 150 watts of power and one geared for rack-mounted servers that consumes 95 watts.

Tulsa chips fit into servers using Intel's "Truland" platform, a server design that also accommodates the single-core "Potomac" Xeon processors introduced in early 2005 and the dual-core "Paxville" models introduced in late 2005.

Compared with Paxville, Tulsa boosts transaction processing performance by a factor of 1.7, ERP software by a factor of 1.4, and ecommerce software by a factor of two, Intel said.

Stephen Shankland writes for CNET News.com