In August 2007 Sun Microsystem announced the UltraSPARC T2, code-named "Niagara 2". The processor contains up to eight processor cores, which are able to execute 8 threads simultaneously each.
With each processor core equally accessing a common large L2 cache and the shared main memory via an internal crossbar the UltraSPARC T2 processor is a shared memory machine on a single chip with a flat memory (UMA = uniform memory architecture) from the programmer's perspective.
While the chip’s peak floating point performance of 11,2 GFlop/s quite moderate, the high potential of the Niagara 2 is revealed, when many threads are active and the high memory bandwidth of some 60 GB/s (theoretically) can be exploited - a frequent bottleneck of standard architectures when executing technical applications.
In October 2007 Sun Microsystems announced the first UltraSPARC T2–based Sun SPARC Enterprise T5120 and T5220 Servers. At Aachen we had the opportunity to test a pre-production system equipped with 8 cores at 1.4 GHz and 32 GB of main memory early on. After experimenting with the predecessor, the UltraSPARC T1 processor, which was not at all designed for HPC but still offered a similar innovative chip multi-threading architecture, ( see “What if the Niagara could count using floating point numbers ...”), we were highly interested in getting hands on this brand new machine.
Two characteristics of the UltraSPARC T2 are particularly interesting from the HPC perspective: the high memory bandwidth and the low power consumption. The question arises: How does this new chip perform in comparison to other recent processors?
For readability we use abbreviations for nomination of the machines in our comparions:
SFE2900 / SFE6900 / SFE25K : Sun Fire E2900 / E6900 / E25K, our UltraSPARC IV based servers
SFV40z : Sun Fire V40z equipped with 4 dualcore Opteron processors
DPE1950W / DPE1950C : Dell PowerEdge 1950 with 2 dualcore Woodcrest or 2 quadcore Clovertown processors resp.
ST5x20 : Sun SPARC Enterprise T5120 / T5220 beta with 1 UltraSPARC T2 processor
You can find more details in the machine descriptions.
All performance measurements have been made on a pre-production system. Therefore they are preliminary and subject to change with upcoming updates of hardware, firmware or software.