Dave Rowell
We tested NT workstations at the start of this year. At that time, IBM and Motorola had agreed on the PReP (PowerPC Reference Platform) hardware that will run operating systems like Windows NT and OS/2, but systems hadn't shown up in retail channels. More important, Microsoft wanted to do more testing on systems based on 603 and 604 PowerPC chips, and thus delayed adding PowerPC support to Windows NT. That will happen by mid-year, if not sooner, with a minor upgrade called NT 3.51.
In spite of the late start, PowerPC systems are expected to do well relative to other RISC-based NT platforms. That's due to IBM's clout and the possibility that Apple, IBM, Motorola, and others will prod
uce systems based on a CHRP (Common Hardware Reference Platform) that will run NT, OS/2, and an upcoming version of Apple's Mac operating system (see ``New PowerPC Hardware Standard to Support Macs'' in News & Views).
IBM, Motorola Computer Group, Tatung, Bull, DTK, FirePower, and other companies have demonstrated NT systems based on 601, 603, and 604 PowerPC processors. There is a PowerPC Windows NT 3.5 beta SDK (Software Development Kit) available, and several applications have been ported to PowerPC, including Excel, Word for Windows NT, SQL Server, WordPerfect, and PhotoMorph 2. By the time the first NT-running PowerPC systems ship this year, there may be as many NT applications for it as there are for Alpha-based systems.
Not wanting to leave PowerPC out of our NT workstation performance testing, we acquired a 601-based PowerPC prototype system from IBM. Though not as fast as 604-based systems will be, the 120-MHz PowerPC system we tested is representative of the first PowerPC systems you'l
l see. It arrived with 256 KB of L2 cache, 48 MB of DRAM, a PCI-based Diamond Stealth 64 graphics card, a 540-MB SCSI hard drive, and the requisite CD-ROM drive.
We compiled and ran BYTE's Native Mode tests on the PowerPC system using new hand-tuned floating-point libraries (beta) from Motorola. We also ran the PhotoMorph test, compiled with the beta NT 3.5 and freshly linked with the same Motorola libraries. The Native Mode test results (an integer index of 2.01 and a floating-point index of 1.87) put the performance of the 120-MHz 601 just below the 200-MHz R4400 Netpower system. Likewise, with the PhotoMorph 2 application test, which mirrors floating-point performance, the PowerPC system came in just below the Netpower system with an index result of 0.96. That puts our PowerPC test system at the bottom of the performance pile with this particular test, only because the systems with slower processors had two of them.
The PowerPC 601 chip appears to give similar performance to Mips chips runnin
g at higher clock speeds. The 604 should be significantly faster. Like Mips chips, PowerPC processors are relatively inexpensive, providing roughly twice the performance per dollar as Intel's Pentium chips. It's likely that PowerPC 601 workstation prices will be in the same range as are Pentium-based systems. Then the decision will come down to the PowerPC's faster performance versus Pentium's compatibility with legacy applications.
Dave Rowell is a BYTE technical editor in charge of hardware reviews. You can reach him on BIX or the Internet at
drowell@bix.com
.
PowerPC: Late to the Party
