How today's chips stack up: Cadalyst Labs tests hot single-processor systems

CADalyst, March, 2004 by Ron LaFon

This roundup includes a variety of microprocessors, ranging from fairly standard 3.2GHz Pentium 4 processors to an Intel Pentium 4 Extreme Edition 3.2GHz processor. There are also several AMD64 processors--one of the new 148 chips and a couple of FX-51 chips.

Similarly, a broad range of motherboard vendors and models are used in these systems. This mix of processors and motherboards provides some interesting information in terms of performance and what chips seem best at particular functions--at least at the current time.

Judging from the benchmark test results, the AMD 64-bit chips excel at 3D performance, but tend to earn weak 2D scores. The Intel Pentium processors, on the other hand, are much more even across the board--not the fastest at 3D, but faster on 2D performance. As a result, the benchmark figures tend to equalize, with each processor family gaining points at what it does best. This makes the sort of work you do a significant consideration for the next workstation you purchase.

Early iterations of the AMD Opteron 148 chips varied wildly in their ability to achieve consistent 3D performance figures. I speculate that there might be a bug of some kind that involves the AGP slot on certain motherboards using the AMD processor. It was troublesome and appeared several times during benchmark testing.

What We Requested

For this roundup, we asked for single-processor systems with the fastest processor (Intel or AMD) available from each vendor. Tested systems had to have at least 2GB of RAM and at least 120GB of total hard drive storage--multiple drives were allowed. We also requested both a CD-R/W drive and a DVD-ROM drive, which could be combination CD/R and DVD drives.

We required an incorporated network card and a 3D OpenGL graphics card with at least 64MB of RAM, capable of supporting resolutions as high as 1280X1024 and 24-bit color at a minimum refresh rate of 85Hz.

Systems were to be preloaded with Microsoft Windows XP Professional configured to the older Classic system interface. Systems came with a mouse with a wheel, but we didn't ask for speakers and monitors or include them in the pricing.

Testing the Goods

I began with the C2001 test using AutoCAD 2004 running on each system in Windows XP Professional at 1280X 1024 with 32-bit color at 85Hz. Nothing was different from the past few times I've used the benchmark, although those of you who haven't tried it with AutoCAD 2004 should download the free updated version on the Cadalyst Web site (www.cadalyst.com). I then ran the proe-02 test suite of SPEC ViewPerf 7.01 on each system, selecting the best score from the two runs through the benchmark that we normally do.

Finally, for the first time I ran the MAXBench4 benchmark test with Discreet's 3ds max 6. Though the benchmark ships on one of the CD-ROMs included in 3ds max 6, a graphic required for one part of the benchmark test is missing. Once I added it to the mix, the benchmark ran without difficulty. Locate this file, rhino.jpg, from a previous version of 3ds max and place it in the MAXBench directory with the other graphics files that the benchmark test uses.

Note that there is no MAXtreme column on the systems using NVIDIA graphics cards because the highperformance MAXtreme driver hadn't been released in time for this review. We will use it for future tests. The PowerDraft driver is also being updated to take advantage of AutoCAD 2004.

Coming Attractions

New microprocessors from Intel and AMD are imminent, so performance characteristics will likely ratchet up over the next few months. Also on the horizon is PCI Express, a two-way serial data bus designed to replace the AGP 8X interface. A PCI Express graphics card slot is expected to provide usable bandwidth of 4GB per second, double that of an AGP 8X slot. Systems and graphics card that support PCI Express are expected out this summer.

None of the submitted systems deviated widely from what we've seen before, though test scores continually climb. All the systems offered standard 3.5" floppy drives, and each had the old standard BIOS, so none of the vendors are yet adventurous enough to move into new territory around that technology--although that will certainly come.

While running these benchmarks, I reflected back on the many systems I've tested in the past using these benchmarks or their predecessors. Not only are systems significantly faster, they're more stable than ever before. This is a result of improvements in the hardware and the software drivers. In the past, I dreaded certain parts of the benchmark tests, because there were stumbling blocks that not all systems made it through. Now, it's extremely rare for a system to have a problem with a test. Much has changed, and for the better. Overall quality has increased over the years, something that we're not always aware of because of the slow, steady pace of improvements.

Review summary

These days, the difference between a midrange CAD workstation and a high-end performance system may be only a few hundred dollars, and most of that difference is the video card. Virtually everyone is trying to get the maximum performance out of systems while keeping them at a relatively reasonable price.