There's Always Tomorrow

The review of Alpha NT systems concludes with a look at workstations from Polywell and Enorex

Over the past several months I've gotten to know some Windows NT Alpha workstations quite well. We went to lunch together, we made fun of the pokey little 300MHz processors from Intel and their clones, and we even tossed back a few beers. So you can imagine how difficult springing the bad news on my 533MHz friends was for me: The Alphas are no match for the wimpy little Pentiums and their kin.

You might think that you can add MHz sequentially to get systems with comparable performance so that, for instance, the performance of a dual 300MHz Pentium II (2 * 300MHz = 600MHz) system would roughly equal the performance of a single-processor 533MHz Alpha system. But the truth is that a single-processor 533MHz Alpha is no match for a dual 300MHz Pentium II. The fastest Alpha I tested for this Alpha NT workstation series of reviews scored a 496 WNT Peak Performance value, using the AIM Technology Workstation Benchmark for Windows NT. The fastest dual 300MHz Pentium II machine I've tested had an AIM WNT Peak Performance value of more than 900.

This canyon-sized performance gap between the single-processor Alphas and the dual 300MHz and 333MHz Pentium IIs I've been playing with in the Windows NT Magazine Lab comes as no surprise. After all, the dual Pentium IIs cost anywhere from two to seven times what some Alpha workstations go for. (Look for reviews of dual Pentium II workstations in the next issue of Windows NT Magazine.)

When I tested a single-processor 300MHz Pentium II system, it scored 515 on AIM's WNT Peak Performance metric--more bad news. Worse still, the fastest Alpha I've tested came with a 10,000rpm Ultra Wide SCSI hard disk, and the Pentium II system I tested had a 7200rpm IDE hard disk. When I used the AIM WNT Sustained Performance metric (the other system-level metric used in this series) to test the Pentium II system, the Pentium II was on par with the Alpha systems: The Pentium II scored 180 application jobs per minute vs. 135 to 191 application jobs per minute for the Alphas. The AIM WNT Sustained Performance results tell me the Pentium II and Alpha systems are comparable in overall performance. However, the AIM WNT Peak Performance results tell me a single-processor 300MHz Pentium II workstation has a slight advantage over a 533MHz Alpha workstation in maximum performance.

In terms of processor performance, the 533MHz 21164PC Alpha processor and the 300MHz Pentium II processor score roughly the same on the Standard Performance Evaluation Corporation (SPEC) 95 integer benchmark. I'm not surprised that these systems run pretty much neck and neck in peak and sustained performance metrics. The Alpha 21164PC processor pulls ahead of the 300MHz Pentium II processor in performing floating-point calculations used in scientific, imaging, and multimedia applications. The SPEC95 floating-point benchmark for the Alpha is nearly twice that of the Pentium II processor. Using this benchmark, I achieved a SPEC95 floating-point value of 15.4 for the Alpha 21164PC processor; the published SPEC95 value for this processor is 16.1. The published SPEC95 floating-point value for a 300MHz Pentium II is 8.15. (For more information about the SPEC95 tests and to see the test results for other Alpha and Pentium systems, go to http://www.specbench.org.)

On the AIM Subsystem test that targets floating-point capabilities, a single-processor 300MHz Pentium II workstation achieves a score of roughly 110,000. A workstation with a 533MHz Alpha scores roughly 45,000--only about a quarter of what you'd expect, if you compare these scores in the same way that you compare the SPEC95 floating-point results for the two processors. What is causing this degradation in the Alpha's performance? The answer I found is simple: The Alpha's floating-point edge doesn't hold when you place the 21164PC processor in an SX or LX series motherboard.

To confirm my conclusion, I spoke with an AIM benchmark engineer, who directed me to tests that AIM has run on an NT workstation fitted with a 21164a processor from an Alpha server. (Most Alpha servers typically use four 533MHz 21164a processors.) The results of AIM's tests were similar to the results I achieved for the 21164PC processors for this Alpha workstation series. I conclude from these results that the 21164PC and 21164a processors are, on a core level, the same processor and, therefore, directly comparable. (AIM floating-point tests for an Alpha server, which multiple processors do not affect, give a score of roughly 200,000. This AIM floating-point value is what I expected to find, given the SPEC95 floating-point results.) Somewhere between the 21164PC's motherboard and the SX and LX series motherboards is a bottleneck the size of the Panama Canal. But don't just take my word for it--go to AIM's Web site at http://www.aim.com for the complete benchmark results on the full range of Alpha systems.

The Alpha workstations' disappointing performance in my testing isn't entirely a surprise. Some industry analysts would say that Pentium IIs and Alphas aren't on the same playing field when running NT 4.0, and I agree. Enhancements to the SX and LX series Alpha motherboards would boost Alpha workstation performance, but you still have to get the 64-bit Alpha processor to effectively run the 32-bit NT operating system (OS). Expecting an Alpha processor to run NT is like tying a prizefighter's hands behind his back and expecting him to win--it's a losing proposition every time. You can run the Alpha version of NT 4.0, which supports the 64-bit Alpha processor, but the OS doesn't run natively at 64 bits. But the times they are a-changin'. With the planned introduction of NT 5.0 late this year and the release of Intel's new 64-bit chip design (code-named Merced), also planned for late this year, the days of NT running in a 32-bit environment are numbered.

NT 5.0 probably won't be a pure 64-bit OS at first, but it likely will have 64-bit support built into the core design. The benefit of a 64-bit design is self-evident: more bandwidth, leading to significantly faster machines. The biggest obstacle to implementing a 64-bit OS is that only 32-bit software currently targets the NT platform. None of today's NT applications would run on a true 64-bit OS.

The NT platform will eventually have a pure 64-bit architecture. The current round of Pentium-based hardware will then be as obsolete as yesterday's 486. Not so the Alpha. Despite the results of my testing, the Alpha processor might be the most lasting hardware technology investment available for the NT platform today. Because we in the Lab believe in lasting investments, we've retained the use of the MaxVision Symbion AXP164SX 533MHz Alpha workstation (see "533MHz Monsters Come On Strong," April 1998). We believe the MaxVision engineers and technical support staff are among the best.

Still, big questions confront the Alpha design. For example, will an Alpha processor exist a year from now? Will an Alpha workstation exist a year from now? Will you be able to purchase an Alpha NT workstation 5 years from now? I answer yes to each question. Microsoft has stated it will support the Alpha design in future versions of NT, and 64-bit platforms will be state of the art in 5 years. Will the Pentium NT workstation exist 5 years from now? The answer depends on whether Merced carries the Pentium nameplate. A far more important question is, will you be able to purchase an NT workstation based on the current 32-bit Pentium design 5 years from now? I give the probability of an affirmative answer the same chance I'd give a prizefighter with both hands tied behind his back.

PolyAlpha 533 SXCompanies that want to upgrade or purchase additional workstations for their engineering departments might want to consider the latest Alpha-based system from Polywell Computers. The flexible PolyAlpha 533 SX workstation offers horsepower, scalability, and a hedge against depreciation and obsolescence. The horsepower comes from the PolyAlpha's 533MHz Digital Semiconductor processor with 1MB of neighboring Level 2 cache. Four PCI slots (two 32-bit and two 64-bit), two ISA slots, and four 168-pin unbuffered DIMMs provide scalability. The PolyAlpha's 64-bit architecture lets you beat the obsolescence curve with a processor and motherboard combination that is likely to fully support future versions of NT.

The PolyAlpha system I tested in the Lab came with 128MB of Synchronous DRAM (SDRAM), a 24X EIDE CD-ROM drive, a Diamond FireGL 1000 Pro PCI video card with 8MB of RAM, a Symbios-based Ultra SCSI-3 adapter, and a 4.5GB Ultra Wide SCSI 10,000rpm hard disk. The only thing missing was a NIC. The SX motherboard uses the Digital Semiconductor-designed 21174 logic processor to optimize communication between the system memory and the PCI bus.

To see how well the PolyAlpha 533MHz SX Alpha processor-motherboard combination worked with the Diamond FireGL 1000 to produce 3-D renderings, I ran three of the five viewsets in the Viewperf benchmark suite from the OpenGL Performance Characterization (OPC) group, as I've done for all the Alpha systems in this series that incorporate 3-D graphics cards. These viewsets--CDRS, which measures modeling and rendering capabilities for CAD; Data Explorer (DX), which measures scientific data visualization capabilities; and Lightscape, which tests a system's ability to realistically reproduce light waves--together provide a useful assessment of a system's 3-D rendering capabilities.

I ran the Viewperf viewset tests using a 21" Intergraph monitor with True Color at a resolution of 1280 * 1024. I averaged three runs of each viewset on the PolyAlpha 533 SX, which scored 21.972 on the CDRS Viewset, 5.857 on the DX Viewset, and 0.398 on the Lightscape Viewset. These scores reflect a value in frames per second. (For a better understanding of the Viewperf tests and to see the results for the PolyAlpha 533 SX in context with Viewperf scores for other systems, visit the OPC Web site at http://www.specbench.org.)

I ran the AIM workstation benchmark tests to measure the PolyAlpha's system performance. I ran these tests after installing a 2-D Matrox Millennium II 4MB video card and setting the monitor resolution to 800 * 600 * 16-bit (in general, 2-D graphics cards provide better AIM test results than 3-D graphics cards provide). The two most telling values from these tests are the AIM WNT Peak Performance value and the AIM WNT Sustained Performance value. The AIM WNT Peak Performance value reflects system throughput by measuring the maximum num-ber of application jobs a system can process in 1 minute. The AIM WNT Sustained Performance value reflects the highest number of application jobs you can run without significantly hindering the system's performance. The PolyAlpha's AIM WNT Peak Performance score was 496 application jobs per minute. Its AIM WNT Sustained Performance score was 191.8 application jobs per minute.

The PolyAlpha's AIM WNT Peak Performance score is the highest for any Alpha system I've tested--and not by a small margin. This score surprised me, considering most Alphas I've tested use the same motherboard and processor that the PolyAlpha uses. The PolyAlpha's SCSI-3 controller and 10,000rpm hard disk help it attain an impressive AIM WNT Peak Performance score. Of course, these hardware wonders push the price up, but you get what you pay for in terms of performance.

Ultra PC 533e
What sets the Ultra PC 533e workstation apart from other Alpha NT workstations is its vendor, Enorex Microsystems. Unlike the other vendors who supplied the Lab with Alpha NT workstations for this series of reviews, Enorex builds Alpha systems exclusively, including servers.

One feature on the Ultra PC 533e that I have not seen on other Alpha systems I've tested is a 120MB 3.5" LS-120 disk drive. The LS-120 drive is backward compatible with existing 1.44MB disks. Enorex plans to make the LS-120 standard equipment on all Enorex Professional series workstations and optional for its Value series workstations. I can't predict this technology's future, but I like the fact that you get a removable storage medium with reasonable capacity without sacrificing an expansion bay.

Otherwise, the Ultra PC 533e is a typical Alpha NT workstation. The PC164SX motherboard uses a Digital Semiconductor-designed 21174 logic processor to interface between the PCI bus and system memory. The Ultra PC 533e has four 128-bit, 168-pin DIMM slots; four PCI slots (two 32-bit and two 64-bit); two ISA slots; 1MB of Level 2 off-chip cache; and a 533MHz 21164 processor. The components on the Enorex system I tested include a 7200rpm 4.5GB Ultra Wide SCSI-3 hard disk, an IntraServer ITI-3140 Ultra Wide SCSI controller, a Toshiba 32X EIDE CD-ROM drive, a 300-watt power supply in an accessible ATX midtower case, 128MB of SDRAM, a NIC, and a Diamond FireGL 1000 Pro video card with 8MB of Synchronous Graphics DRAM (SGDRAM).

You can use the Ultra PC 533e to render 3-D animations, create CAD and computer-aided manufacturing (CAM) designs, and perform other tasks requiring spatial analysis. The Diamond FireGL 1000 video card gives this system its 3-D capabilities but limits the scope of 3-D work. The Diamond FireGL 1000 is an entry-level OpenGL-compliant 3-D video card. You can purchase a Diamond FireGL 4000 card for more intensive 3-D renderings and design.

The Diamond FireGL 1000 (and Diamond FireGL 4000) provides direct support for StereoGraphics Corporation's CrystalEyes virtual-reality goggles. The CrystalEyes Goggles connect to the video card through a three-pin round port next to the video-screen connector. The goggles give viewers the 3-D advantage. For example, mechanical engineers can view how a design works before they build a prototype. Scientists can navigate particles on an atomic level better than they can with a 2-D view. DNA researchers can gain a better understanding of the millions of chromosomal threads entwined in a DNA molecule.

To measure the 3-D rendering capabilities of the Diamond FireGL 1000 Pro video card in the Ultra PC 533e workstation, I ran the CDRS, DX, and Lightscape Viewsets. The Ultra PC 533e achieved a score of 22.177 on the CDRS Viewset, 5.577 on the DX Viewset, and 0.419 on the Lightscape Viewset.

To get a good idea of how all the components in the Ultra PC 533e work together, I ran the AIM workstation benchmarks. For these tests, I used a Matrox Millennium II 4MB video card and a monitor resolution of 800 * 600 * 16-bit. The Ultra PC 533e's AIM WNT Peak Performance score was 372.9 application jobs per minute. Its AIM WNT Sustained Performance score was 168.7 application jobs per minute. These values are consistent with the values achieved by the other Alpha workstations I've tested.

Enorex engineers are quick to point out that a version of Corel's Office Suite now exists for the Alpha architecture. Approximately 3000 native Alpha NT applications exist. During a recent vendor visit to the Lab, a vendor representative made it a point to let us know that developing an Alpha version of his company's software was a high priority. Digital may no longer exist, but the Alpha legacy lives on.

Corrections to this Article: