Talkback
E-mail
Printer Friendly
Reprints
The past few years have seen some major changes in Sun hardware. The return of Andy Bechtolsheim has brought forth an impressive array of new server hardware, and reinvented Sun as an x86 server vendor. But where does that leave the SPARC?
|
|||||||||||||||||||||||||||||||||||||||||||||||||
For many years the SPARC processor was lauded as the only high-end business platform worth investing in, and sales of SPARC-based servers and workstations were nearly as high as the prices. In the eighties and nineties, a tour through any financial company was roughly the same as a tour through a Sun showroom. In the late nineties that began to change, as Linux and even Solaris x86 began to surge in popularity due to the extremely low cost and reasonable performance. Since then, the performance game has left the SPARC platform bereft.
But all is not lost. The SPARC platform still has some kick left, and while that kick may not be in clock speed, it’s certainly in the threads.
I’ve had a Sun SPARC Enterprise T5120 server in the lab for a few months now, and I’ve found it to be a very solid platform, as expected, and much like its predecessor based on the “Niagara” T1 processor, it’s very, very good at some tasks, and not so good at others.
[ See my review of Sun’s original, UltraSPARC T1-based server, the Sun Fire T2000. ]
The reasons for this performance lie in the CPU itself. The T5120 runs the UltraSPARC T2 processor, which is a single CPU running four, six, or eight cores. This architecture is a significant departure from the rest of the industry, and positions these servers to fill specific, highly transactional roles, foregoing the rest. In that capacity, they do quite well.
The tale of T2
The basis of the UltraSPARC T2 design is to fit as much onto the processor die as possible, decreasing latency and increasing
bandwidth to mainline I/O, such as 10G Ethernet and RAM. In order to achieve this, Sun has basically produced a “system on
a chip” with core-specific memory controllers, FPU, cryptographic accelerator (SPU, or Stream Processing Unit), and 4MB L2
cache, along with x8 PCIe and two 10G Ethernet controllers -- all residing on the CPU die. Each of the eight discrete cores
is positioned to support eight concurrent threads. All together, this equals a single CPU that uses less than 95W, and appears
to the OS as 64 CPUs – a neat trick.
This represents the second iteration of Sun’s play in this game, supplanting the UltraSPARC T1. The T2 has many benefits over the T1, such as the integrated per-core FPU and crypto units. These two additions erase some of the math and crypto performance concerns of the T1, while bumping the top clock speed to 1.4GHz.
But what does this mean in the real world? Easy — the T2 is a great choice for highly threaded applications, and a miserable choice for single-threaded applications. The low clock speed causes single-threaded apps to chug along, while the eight cores give multi-threaded apps plenty of room to stretch out. Each thread is still bound to the 1.4GHz clock speed, but depending on the application, that doesn’t necessarily equal slower performance.
My T5120 evaluation unit came with the single, eight-core T2 CPU, 64GB of RAM, four copper gigabit Ethernet ports, two USB 2.0 ports, three PCI-E slots (two doubling as 10G Ethernet slots), serial and Ethernet service processor connections, and redundant power supplies.
Paul Venezia is senior contributing editor of the InfoWorld Test Center and writes The Deep End blog.
