Intel will unveil its "Tulsa" chip for multiprocessor servers August 29, finally talking publicly about a chip it has been shipping to vendors for several weeks now.
The company hopes the new chip will stem its loss of market share to the Opteron chip from Advanced Micro Devices (AMD). Server vendors including IBM and Dell have recently announced they would build more computers powered by Opteron instead of Intel's Xeon chips.
Now Intel will add the new chip to its Xeon family of dual-core processors, slotting it above the Dempsey chip launched in May and Woodcrest chip launched in June. Tulsa is built with the same 65-nanometer process as those chips, but uses Intel's aging Netburst architecture and Truland platform, instead of the new Core Microarchitecture and Bensley platform.
Tulsa is designed for servers with four or more processors, as an upgrade from Intel's current product in that space, the "Paxville-MP" chip, Intel engineers said in a presentation at the Hot Chips trade show in Palo Alto, California, this week.
Compared to that chip, Tulsa's cores run 13 percent faster while using 20 to 40 percent fewer watts. Intel builds each Tulsa chip by combining two 3.4GHz Pentium 4 cores on a single die, and will deliver both 150-watt and 95-watt versions. Technically, the main advantage of Tulsa is that it supports four threads per processor and has a large, 16M-byte cache, compared to Woodcrest's 4M-byte cache.
Intel will target Tulsa at customers with demanding business applications such as enterprise resource planning and e-commerce, according to spokesman Nick Knupffer.