Oracle will reduce the number of cores in processors SPARC T4
The next generation of server processors SPARC T4 will be with eight cores instead of 16 in the current SPARC T3. This decision was made by Oracle, which now owns the microprocessor division of Sun.
SPARC T3 after its appearance in September 2010 was the first server processor with 16 cores. A large number of threads are better suited for processing web requests and other easily parallelized tasks, and not for the ERP systems Oracle specializes in. Therefore, it was decided that in SPARC T4 (expected in 2011) the number of cores will be reduced in order to improve the performance of each individual thread.
A hint of a reduction in the number of cores was made during the presentation of Sparc Superclusterlast week. In one of the presentation slides there was information about plans for SPARC T4. Later information about the return to eight cores was confirmed.
Larry Ellison said that in SPARC T4, the performance of a single thread will be tripled, so that even 8-core processors in Oracle servers will be much more efficient than SPARC T3.
At the same time, Sun (before the takeover of Oracle) always positioned SPARC T3, which was then called UltraSPARC T3, as a processor for "network" tasks. However, Oracle has canceled this positioning and describes it as a “general purpose processor,” and Sparc Supercluster introduced last week as a “general purpose computer”.
Sparc Supercluster even managed to overtake IBM in the officialTPC-C 's top-performing clusters with 30 million transactions per minute, although Oracle’s methods have been criticized by IBM. They believe that “combining 27 systems is an extremely inefficient approach that is impractical in the real world, and is caused only by the desire to win the test.”
In addition to Sun chips, Oracle also uses Fujitsu's Sparc64 VII + (3 GHz, 12 MB cache) processors in its M-series servers. This is another implementation of the same Sparc v9 architecture, and “sharpening” for single-thread performance is already clearly visible here.
According to some analysts, in the future, Oracle wants to standardize processors and use only one implementation.
SPARC T3 after its appearance in September 2010 was the first server processor with 16 cores. A large number of threads are better suited for processing web requests and other easily parallelized tasks, and not for the ERP systems Oracle specializes in. Therefore, it was decided that in SPARC T4 (expected in 2011) the number of cores will be reduced in order to improve the performance of each individual thread.
A hint of a reduction in the number of cores was made during the presentation of Sparc Superclusterlast week. In one of the presentation slides there was information about plans for SPARC T4. Later information about the return to eight cores was confirmed.
Larry Ellison said that in SPARC T4, the performance of a single thread will be tripled, so that even 8-core processors in Oracle servers will be much more efficient than SPARC T3.
At the same time, Sun (before the takeover of Oracle) always positioned SPARC T3, which was then called UltraSPARC T3, as a processor for "network" tasks. However, Oracle has canceled this positioning and describes it as a “general purpose processor,” and Sparc Supercluster introduced last week as a “general purpose computer”.
Sparc Supercluster even managed to overtake IBM in the officialTPC-C 's top-performing clusters with 30 million transactions per minute, although Oracle’s methods have been criticized by IBM. They believe that “combining 27 systems is an extremely inefficient approach that is impractical in the real world, and is caused only by the desire to win the test.”
In addition to Sun chips, Oracle also uses Fujitsu's Sparc64 VII + (3 GHz, 12 MB cache) processors in its M-series servers. This is another implementation of the same Sparc v9 architecture, and “sharpening” for single-thread performance is already clearly visible here.
According to some analysts, in the future, Oracle wants to standardize processors and use only one implementation.