HP xw9300 Workstation review
In a world where the raw computing power of the average desktop PC exceeds the supercomputers of the recent past, it’s pertinent to ask whether machines labelled as workstations are worthwhile or simply a way for manufacturers to overcharge for hardware essentially identical to a standard PC. Both HP and Fujitsu Siemens would like you to think otherwise.
The internals of the xw9300 Workstation aren’t quite as heavily engineered as those of the Fujitsu Siemens, and the casing isn’t as deep at 515mm. But the build certainly isn’t lightweight; take off the chassis side panel to access the machine and it feels like it would stop a bullet.
The base specification of our review machine was a pair of single-core Opteron 252 processors running in an HP-branded nVidia nForce Pro motherboard, 4GB of ECC RAM occupying four of the eight DIMM sockets, plus a single Maxtor Atlas 15K II SCSI hard disk spinning at 15,000rpm – twice the spindle speed of a standard fast desktop SATA disk and 50 per cent faster than even Western Digital’s Raptor 10K SATA drives. Graphics come courtesy of dual nVidia Quadro FX 3400 workstation graphics in an SLI setup. With 256MB of RAM per card and 28.8GB/sec memory bandwidth, there’s plenty of grunt on offer.
Inside, all chips on the motherboard are passively cooled, with the exception of the two Opteron CPUs; these both have active copper heatsinks and share a single air duct to bring cooler air in from the front of the machine rather than recycling warm air from the internals. This is a sound idea, since the two Quadro cards – both with their own standard active nVidia heatsinks – become very hot, to the extent that you can easily feel their radiated heat when you take the cover off the machine. HP makes a lot of its intelligent fan-speed system controlling the rear 120mm and front 90mm case fans, but unfortunately it isn’t particularly clever. In use, fan speed jumps in coarse leaps, which can lead to a constant yo-yo change in sound level and pitch that we found far more annoying than a loud machine at a constant level. The Fujitsu Siemens does much better in this regard.
Internals are tool-less, with expansion cards secured in the backplane by a single securing bar held by two easily removed clips. It’s the same story with the hard disk, mounted horizontally on rails at right angles so it will slide out simply by pushing the side clips together and giving it a tug. It falls short of hot-swap capability, but replacing a drive will take only about ten minutes. Three more quick-release bays stand ready to accept drives, but that’s the limit of drive expansion potential, save for a free 5.25in front panel bay for an extra optical drive. As it is, a single Philips dual-layer DVD writer occupies one front bay and a floppy drive the second. A heavyweight Delta 750W power supply completes the picture, although the system stops short of server-style power supply redundancy.
There’s plenty of scope for storage expansion beyond drive bays: four unused SATA II ports give full RAID0, 1 and 0+1 capability, and an unused integrated Ultra320 SCSI port is joined by an LSI Logic 53C1030 Ultra320 PCI-X card, giving one extra internal port plus an external LVD/SE connector. This takes up the only PCI-X/133 slot, with two 100MHz PCI-X slots remaining. A standard PCI slot sits between the Quadro cards, so is effectively unusable.
HP answers questions of the relevance of the ‘workstation’ label with consummate ease. With an overall application benchmark score of 1.28, it’s faster than any desktop machine we’ve tested; the prime factors in that score are the 252 Opterons – the highest-specification, single-core Opteron available – and the mass-storage subsystem. The Atlas 15K drive shows that SCSI isn’t dead when performance is the top priority. And when it comes to graphics performance, the xw9300 is just as impressive, completing our 3ds Max 100-frame 100,000+ polygon animation preview test – a separate test from our standard benchmark suite – in an incredible eight seconds under DirectX, a speed of half real-time. It’s a good indicator of performance when a machine completes a benchmark so fast you realise you’re about to need a new one.