Fujitsu/Fujitsu Siemens Computers PRIMEPOWER Enterprise Servers
© 2003 Gartner, Inc. and/or its Affiliates. All Rights Reserved.
DPRO-97941
17 April 2003
11
processors than the Sun server, it also supports more than four times as many I/O slots. While few
customers are likely to require such high-level I/O—most large servers are connected to storage area
networks (SANs) rather than huge amounts of direct attached storage—the PRIMEPOWER XA servers
offer a balanced design, with high I/O throughput to match the high throughput of the system interconnect.
One key point about the PRIMEPOWER’s OLTP capabilities is that with its more powerful SPARC64
processors and lower intersystem board latencies, the PRIMEPOWER should be able to comfortably do
the same amount of work as the Sun Fire servers, but with fewer processors. The PRIMEPOWER XA
models will not achieve quite such reduced configurations compared with the Sun servers as can be
achieved with IBM’s POWER4 and HP’s forthcoming dual-core PA-8800 processors, but we estimate that
similar performance can be achieved with a good 25 percent fewer processors than the Sun F15K for a
similar workload. That can mean less expensive licenses for software priced on a per-processor basis,
such as Oracle.
Table 4: Fujitsu SPARC64 vs. Sun UltraSPARC III Cu
SPARC64 V
UltraSPARC III Cu
Clockspeed
1.35GHz
1.05GHz or 1.2GHz
SPECint2000
858 (PRIMPOWER 900)
550 est. (1.05GHz)
590 est. (1.2GHz)
Technology
0.13 micron copper
17 million transistors
0.13 micron copper
16 million transistors
Cache Sizes
L1: 128KB (instruction) +
128KB (data)
L2: 2MB on-chip (two-way);
fully tagged
L1: 32KB (instruction) +
64KB (data)
L2: 8MB off-chip (one-
way); only eight sublines
Out-of-Order Execution
Yes
No
End-to-End Data Protection
ECC for caches and buses;
parity protection on
execution units and
registers
ECC for caches and buses;
no protection on execution
units and registers
Power Consumption
50 Watts
53 Watts (1.2GHz)
The technology is impressive in terms of power and capacity. The previous generation of PRIMEPOWER
servers, which included the PRIMEPOWER 2000, was already the single-system benchmark-leader on
Transaction Processing Council’s TPC-C and SAP R/3 benchmarks. The PRIMEPOWER 2500 with its
more powerful processors, 2x faster crossbar and greater I/O bandwidth looks set to maintain that
leadership. Fujitsu estimates that the 2500 should offer 80 percent better OLTP performance than its
predecessor, the PRIMEPOWER 2000.
Fujitsu has not been shy of submitting its servers to the benchmarks and publishing the results, though
the TPC-C benchmarks were undertaken using its obscure proprietary database software called
SymfoWare. We believe the PRIMEPOWER 2500 will maintain its lead on OLTP benchmarks through
2003, but Fujitsu should use Oracle if it really wants to prove itself against the competition. A 128-way
server running Oracle is not likely to look so good in price/performance terms, however. Nevertheless,
only a very small number of clients will ever require such high levels of scalability.
High Availability (HA)
The PRIMEPOWER 2500 doesn’t only excel in scalability and capacity terms. Its hardware RAS features
are also advanced and set the standard for the Unix competition. No other Unix vendor has designed
